From this information cost, the paper derives the firm’s legal structure, including ownership, employment, and the line between employee and contractor. This post summarizes Alchian and Demsetz’s paper and offers some thoughts on its contribution to the literature and its weaknesses.

A Firm Grip on Shirking

The paper’s central example is simple: “Two men jointly lift heavy cargo into trucks. Solely by observing the total weight loaded per day, it is impossible to determine each person’s marginal productivity” (p. 779). Marginal productivity is the extra output that one input—a worker, for example—adds. In team production, that contribution depends on what the others do. A lifter adds little if the partner doesn’t lift. The day’s total therefore cannot reveal each person’s contribution. Economists call such output nonseparable.

Team production invites shirking. A worker who slacks off keeps all the extra leisure but bears only part of the lost output, because the team shares the shortfall. In the paper’s phrase, “viable shirking is the result” (p. 781).

The remedy is to monitor inputs, but monitoring is costly. Alchian and Demsetz call this the metering problem. Their solution is a specialist monitor. To give her a reason to watch closely, the team gives her “title to the net earnings of the team, net of payments to other inputs” (p. 782). She becomes the residual claimant, keeping whatever remains after paying the others. Every bit of shirking she prevents adds to her income.

The paper then defines the firm by five rights the monitor holds: She claims the residual earnings, observes how inputs behave, and serves as the common party to every input’s contract. She can also change the team’s membership and can sell the whole position. That bundle, Alchian and Demsetz argue, is what it means to own a classical firm. Ownership, in this view, solves a measurement problem.

From Coase’s Insight to Testable Predictions

The paper is partly a response to Ronald Coase’s theory of the firm. Coase argued that a firm takes on an activity when organizing it internally costs less than buying it on the market. Alchian and Demsetz say their “view of the firm is not necessarily inconsistent with Coase’s,” but call his claim “a difficult proposition to disagree with or to refute” (p. 783). The fact that a firm brought an activity in-house becomes the basis for inferring that the market alternative cost more. But Coase offers no empirical test to support that inference.

Alchian and Demsetz do offer predictions. Professional and artistic work, where effort is hard to observe, should take place in small partnerships whose members monitor each other. Profit sharing should appear in small teams, where a specialist monitor costs more than she saves. In corporations, where stockholders delegate monitoring to managers, rival management teams should keep managers in check through proxy fights—contests for shareholder votes—and takeovers. Some of these predictions are qualitative rather than numerical, and none is easy to test cleanly. But evidence can prove them wrong.

Alchian and Demsetz’s achievement is easy to underrate because it succeeded. Explaining the firm’s boundary—which activities it handles internally—rather than assuming it is now close to standard practice. Standard ideas look obvious in hindsight, but this one wasn’t obvious in 1972.

The paper had to show that the question was worth asking, supply a mechanism specific enough to generate predictions, and use that mechanism to explain the firm’s legal form. That combination explains why the paper became a classic.

Seminal though it is, the paper has shortcomings. Curiously, Alchian and Demsetz seem to have recognized many of them without addressing them.

Why Employers Aren’t Just Customers

Alchian and Demsetz deny that the firm rests on authority. A firm “has no power of fiat, no authority, no disciplinary action any different in the slightest degree from ordinary market contracting between any two people” (p. 777). An employer’s power over an employee is “exactly the same as one little consumer’s power to manage and assign his grocer to various tasks” (p. 777). In their analogy, the employer is the customer, and the employee is the grocer. Either may walk away, so neither commands the other.

The metering argument doesn’t need this claim. Team production and costly monitoring could generate the firm’s bundle of rights whether or not the employer also holds authority. The denial reads as an unnecessary defense of classical price theory—the traditional account of how prices coordinate market activity. The paper’s core contribution survives without it.

The claim also sits uneasily with employment law, which distinguishes employees from independent contractors. Courts have held employers liable for employees’ torts. Employees owe duties of loyalty to their employers, and states may enforce agreements that restrict employees from working for competitors. A grocer’s relationship with customers has no clear counterpart to any of these.

An equal right to exit also doesn’t mean equal stakes in exit. A grocer sells to many customers, so losing one costs a small share of revenue. A typical employee sells labor to one buyer, so losing that buyer costs all of it. A grocer may attract a replacement customer with little effort, while a worker may face a long and costly job search. Jobs differ in ways that customers buying bread don’t.

Loss aversion compounds the asymmetry: People tend to weigh a loss more heavily than an equal gain. Because the worker stands to lose more than the employer, she may accept less favorable terms than she otherwise would.

The analogy also presumes that a worker has many potential employers. In “Labor Market Power,” David Berger, Kyle Herkenhoff, and Simon Mongey estimate a model of local labor markets dominated by relatively few employers and find economically significant distortions. Where employers have such power, wages fall below the value of workers’ added output for reasons unrelated to metering.

Alchian and Demsetz acknowledge market power. Asking why sports teams own athletes’ future services, they name “cartelization and monopsony gains to team owners”—gains from coordinating their behavior or limiting athletes’ options—as well as “costs of monitoring production of athletes.” They then concede, “we know not on which to rely” (p. 793, n. 16).

That’s a candid admission that they cannot distinguish between competing explanations. Monitoring costs and market power produce similar arrangements, and the paper cannot tell them apart. Its opening denial of an employer’s power is hard to square with that admission.

The Monitor’s Conflicting Incentives

Alchian and Demsetz raise the obvious objection to their own solution: “who will monitor the monitor?” (p. 782). Their answer is the residual claim. A monitor who keeps what remains after paying the inputs gains from monitoring well.

But the monitor also gains from paying less. The residual is output minus payments, so understating a worker’s marginal product—the extra output she contributes—directly raises the monitor’s income. The problem follows from the model’s own premise: Individual contributions are hard to observe, which is why the firm exists. For the same reason, workers can’t check the monitor’s valuations. The information gap that justifies a central monitor may also bias her judgments.

The paper names one check, then qualifies it: “[O]ne constraint on the monitor is the aforesaid market competition offered by other monitors, but for reasons already given, that is not perfectly effective” (p. 782). The analysis doesn’t explore what happens when that check falls short. The firm exists because markets cannot cheaply measure what each input contributes. Relying on competition among monitors seems to ask the market to supply the very information the theory says it lacks.

The paper’s treatment of unions shows that Alchian and Demsetz recognize the general problem. Employees monitor employers, too. Wages are “extremely easy to check” (p. 790), but contingent pay—compensation that depends on future conditions, such as pensions—isn’t. Employees therefore hire a specialist, the union agent. That agent, the paper notes, “has incentives to use pension and retirement funds not entirely for the benefit of employees” (p. 790). The same reasoning applies to the residual claimant, but the paper doesn’t extend it there.

When the Work Is Part of the Pay

The model’s workers trade leisure against income and shirk when shirking is hard to detect. Otherwise, they’re identical: Each dislikes effort to the same degree.

Alchian and Demsetz recognize that people differ. The firm’s role in judging inputs is “obscured in the theoretical literature by the assumption of homogeneous factors”—the idea that workers and other inputs are alike (p. 793). They also note that personal relationships matter. Partnerships form among relatives and long-standing acquaintances “because each knows better the other’s work characteristics and tendencies to shirk” (p. 790).

These differences matter for shirking, too. If workers differ in their distaste for effort, the firm must sort workers as well as measure their contributions. Offering a range of contracts lets workers choose. If those least averse to effort choose the least supervised terms, that selection does part of monitoring’s job. A model of identical workers cannot generate that response or explain why workers doing the same job at the same firm sign different contracts.

The simplification matters most for nonprofits. No one can sell an ownership stake in a nonprofit, so no one captures the value of improving it through such a sale. The paper concludes that “one should, therefore, find greater shirking in nonprofit, mutually owned enterprises” (p. 790).

That’s plausible. If nonprofit pay runs below private-sector pay, the paper can explain the gap as compensation in kind: Workers accept less money in exchange for other benefits. Pay may be “lower in pecuniary terms and higher in leisure, conveniences, and ease of work” (p. 781).

Volunteers are harder to explain. A volunteer receives no pay and still supplies effort. If withholding effort were the only nonmonetary reward from work, volunteering would be shirking with nothing to shirk from. Volunteers get something from making an effort, and the model has no term for that.

Paid nonprofit work may be an intermediate case: part wage labor and part donated effort. Lower nonprofit pay then has two explanations. Workers may shirk more easily, as Alchian and Demsetz suppose, or accept less because they value what the work produces. Both stories predict the same wage gap.

The model’s gap traces to one simplifying assumption. Footnote 5 says that “leisure stands for all nonpecuniary income for simplicity of exposition” (p. 780). But the shirking mechanism requires nonmonetary rewards to fall as effort rises. When those rewards rise with effort instead, the incentive runs the other way. Such workers may supply more effort than their wages alone would buy.

Alchian and Demsetz consider a nearby idea and set it aside. Team spirit and loyalty improve a team, they argue, “because of the reduced shirking—not because of some other feature inherent in loyalty or spirit as such” (p. 790). That keeps motivation within the framework as a check on shirking. But it doesn’t account for a worker for whom the work itself is part of the pay. Such a worker needs no deterrent to shirking.

Noisy Signals and Lasting Relationships

The behavioral signals the paper relies on are noisy. Clues to productivity come from watching inputs, such as “how rapidly does a man move to the next piece to be loaded, how many cigarette breaks does he take” (p. 780). A long break could signal shirking or recovery from hard work. Imperfect measurement is the paper’s premise, so this doesn’t undermine the argument. But the less reliable the signal, the wider the monitor’s unchecked discretion and the costlier monitoring becomes.

The paper also sets aside the effects of repeated dealings within employment. Footnote 2 treats reputation as an asset worth the future income it secures, but applies the idea to market exchange in general. The paper denies that employment rests on long-term contracts “any more than on a series of short-term or indefinite length contracts” (p. 784).

Jobs likely last longer than one-off purchases. If they do, repeated dealings—and the incentive to preserve a valuable relationship—may do some of the work the paper assigns to the residual claimant.

Why the Model Still Matters

Alchian and Demsetz’s central insight holds up. Individual contributions are costly to observe, joint production makes them costlier still, and firms may exist partly to manage that difficulty. By tying how people produce together to how firms organize, the paper moved the study of firms toward testable predictions.

The limitations raised here mostly concern what the model leaves out: the unequal stakes of separation for employer and employee, the monitor’s incentive to undervalue contributions only she can observe, and differences in workers’ distaste for effort. The paper’s explicit prediction about nonprofits seems to me the least likely to survive a more realistic model.

The paper remains a classic. Many later papers build on it, and some address the limitations noted here. Its contribution lies less in getting every part right than in advancing the economic study of firms.

It’s also worth reading today. New technologies can track workers’ keystrokes and deliveries, changing the cost of observing individual contributions and possibly the firm’s role. More than 50 years later, Alchian and Demsetz’s questions and analysis remain relevant.

Further Reading

• Armen A. Alchian & Harold Demsetz, “Production, Information Costs, and Economic Organization,” American Economic Review, Vol. 62, No. 5 (1972).
• R. H. Coase, “The Nature of the Firm,” Economica, Vol. 4, No. 16 (1937).
• David W. Berger, Kyle F. Herkenhoff & Simon Mongey, “Labor Market Power,” American Economic Review, Vol. 112, No. 4 (2022).

The post ‘Production, Information Costs, and Economic Organization,’ by Armen Alchian & Harold Demsetz appeared first on Truth on the Market.

“The chandeliers grew heavier and heavier until they started pulling ceilings down,” Khrushchev noted. “The workers fulfilled the plan… but who needs this plan? To whom does it give light?”

As economists have since pointed out, this illustrates Goodhart’s law: When a measure becomes a target, it ceases to be a good measure. Central planning rewards producers for hitting numerical targets, even when doing so wastes resources and destroys economic value.

Decades later, the Federal Railroad Administration (FRA) appears to have adopted the Soviet lighting industry’s approach. Under pressure from executive mandates to deliver a high volume of deregulation, the FRA recently touted a record of 65 deregulatory actions finalized between January 2025 and late 2026.

On paper, that’s a staggering amount of regulatory relief. In practice, the FRA has simply manufactured heavy, low-value chandeliers.

The Great Deregulatory Paper Shuffle

Look past the gross output metric—the agency’s claimed 65 actions—and the record looks vastly larger than the practical relief delivered to the American rail network. A close review of these rules reveals an agency aggressively padding the administrative ledger to meet a quota, largely blind to the need for productive economic outcomes.

Consider the “July 1 Dump.” On July 1, 2025, the FRA published a flurry of final rules. Of its 65 claimed deregulatory actions, a staggering 35 were purely administrative updates finalized on or around that day. What did these sweeping deregulatory reforms accomplish? They changed physical addresses, updated web links, and clarified internal definitions.

For example, the agency issued separate final rules making “administrative updates” to the Railroad Workplace Safety Regulations (49 C.F.R. Part 214), the Special Notice and Emergency Order Procedures (49 C.F.R. Part 216), the Railroad Freight Car Safety Standards (49 C.F.R. Part 215), and the Control of Alcohol and Drug Use Regulations (49 C.F.R. Part 219), among many others.

The regulatory impact analysis—an assessment of a rule’s economic effects—for nearly all these actions relies on identical boilerplate text. Each notes that the rule “imparts no additional burdens on regulated entities” because it merely replaces references to specific penalty amounts with generic references to the Code of Federal Regulations (CFR), or updates an email address.

Updating agency routing codes and proofreading the CFR are basic housekeeping tasks of the administrative state. Rebranding them as substantive deregulatory actions is a masterclass in performative compliance. The FRA claims credit for shrinking the regulatory burden without actually liberalizing the prescriptive command-and-control frameworks—the detailed mandates that dictate how railroads operate—that govern the industry.

Three Rules and Some Spare Change

The absence of real economic impact in the FRA’s recent output is stark. For approximately 95% of its 65 actions, the agency quantifies no economic savings, relying instead on qualitative claims about “improving readability” or “clarifying language.” That leaves exactly three rules for which the FRA quantified regulatory savings. Even here, the estimates show narrow, limited relief.

• Emergency Escape Breathing Apparatus (EEBA) Standards (Aug. 7, 2025): This rule projects between $2.7 million and $7.7 million in savings over 10 years. But those savings don’t stem from eliminating an unjustified or redundant requirement. They come entirely from extending a 2024 rule’s compliance dates by one year. The rule simply postpones compliance costs rather than delivering structural reform.
• Reflectorization of Rail Freight Rolling Stock (Jan. 27, 2026): This rule, which amends 49 C.F.R. Part 224, boasts the largest quantified savings: an estimated $91 million over 20 years, or about $8.6 million on an annualized basis. But this figure doesn’t reflect a comparable change in actual railroad practices. For more than a decade, the industry had operated under a waiver allowing it to depart from these requirements for reflective markings on freight cars. Making that longstanding waiver permanent is certainly good administrative practice: It saves the government and industry the modest time needed to renew the paperwork periodically. But it delivers almost no additional reduction in railroads’ operating costs, since they already enjoyed the operational relief.
• Training, Qualification, and Oversight for Safety-Related Railroad Employees (May 15, 2026): This final rule, which amends 49 C.F.R. Part 243, yields an estimated $1.1 million in cost savings. Yet that relief comes solely from a one-year reprieve from annual refresher-training requirements, granted exclusively to small entities. It offers no estimated relief for Class I railroads—the largest railroads and the primary drivers of national freight volume.

Meet the New Rules. Same as the Old Waivers.

The remaining actions in the FRA’s portfolio follow a similar pattern of marginal tweaks masquerading as significant liberalization. Another batch of rules finalized throughout 2026 largely focuses on codifying existing waivers and eliminating redundant paperwork.

For instance, the April 2026 rule amending qualification and certification requirements for locomotive engineers and conductors (49 C.F.R. Parts 240 and 242) simply codifies longstanding waivers for railroads participating in the Confidential Close Call Reporting System (C3RS), a program for confidentially reporting safety incidents and near misses.

Another April 2026 rule revises the definition of a non-traversable curb—a barrier designed to prevent drivers from going around lowered crossing gates—in quiet zones, where trains generally don’t sound their horns (49 C.F.R. Part 222). Again, it codifies five preexisting waivers. In August 2026, the FRA permitted 3D computer-based simulations for periodic refresher training on brake systems (49 C.F.R. Part 232)—a sensible modernization, but one the agency already allowed through waivers.

Codifying waivers is a legitimate and necessary function of administrative agencies. It frees up time and resources otherwise spent drafting, submitting, and reviewing repetitive waiver requests. It also saves agency lawyers time and provides regulatory certainty.

But when an agency’s deregulatory agenda consists almost entirely of administrative updates and codifying existing practices, it signals that the agency has deprioritized the hard work of structural reform.

Time to Lighten the Load

The FRA has met its deregulatory quota. It has delivered 65 actions, dutifully checking the current administration’s boxes. In doing so, it has produced the regulatory equivalent of a Soviet chandelier: a heavy volume of paperwork that looks impressive on a centralized ledger but sheds little light on the industry it regulates.

Clerical updates dominate the FRA’s action count, while the few quantified savings are narrow, temporary, or largely already realized in practice. This isn’t evidence of broad, measurable deregulation. It’s evidence of an agency gaming an arbitrary metric.

Real deregulation requires agencies to tackle the substantive bottlenecks that suppress competitiveness, capital investment, and operational efficiency. It requires shifting from rigid, technology-blocking crew-size mandates and prescriptive inspection rules to performance-based standards that set safety goals while allowing railroads to choose how to meet them. Those standards could accommodate modern technology, such as autonomous track inspection systems that check tracks without an onboard inspector.

It also requires streamlining review timelines under the National Environmental Policy Act (NEPA) that stall critical freight infrastructure, including facilities where cargo moves between trains and trucks.

The FRA needs to judge regulatory relief by what it delivers, not the gross tonnage of its rulemakings. The rail industry needs lighter burdens, not heavier chandeliers.

The post Deregulation by the Ton: The FRA’s Heavy Paperwork and Lightweight Relief appeared first on Truth on the Market.

Altman made the case during an Oct. 29 appearance on Cheeky Pint, the podcast hosted by Stripe co-founder John Collison. They were discussing why high prices hadn’t prompted memory manufacturers to expand supply faster. Altman challenged the idea that running artificial intelligence (AI) models on personal computers would ease the need for data centers. With memory scarce, he argued, centralized systems could put it to more consistent use.

Altman strikes me as earnest. I think he means it when he says that, from a “global allocation perspective,” it’s “quite terrible to have everybody have a bunch of the world’s limited supply of DRAM sitting in their own computer… being used 5% of the time.” DRAM—dynamic random-access memory—is the working memory computers use to run programs.

His premise is straightforward: Memory is scarce, so it should go wherever it gets the heaviest, most consistent use. By that logic, memory sitting idle in a laptop 95% of the time would serve us better in a data center, working almost continuously for thousands of users.

That view is understandable coming from someone whose firm needs vast data centers to run AI models. But it mistakes how often we use something for how much we value it. Assets we expect to use for years can justify their cost simply by being available when we need them. Privacy, control, and peace of mind also count, even if they don’t show up in a utilization rate.

More subtle versions of this mistake can be just as well-intentioned. In a recent interview with Ben Thompson, Palantir Chief Technology Officer Shyam Sankar aptly compared the current AI revolution to the spread of electricity in the late 19th and early 20th centuries. His central point is correct: Realizing the gains from this new technology requires redesigning how we work.

But the electricity analogy has limits. Altman’s complaint about idle memory shows where it begins to break down.

Your Guest Room Is Not a Policy Failure

Electricity is perishable. Unless we store it, energy generated but not consumed goes to waste. Grid operators must continually match supply and demand, and persistent excess capacity can suggest that a grid is overbuilt—though some spare capacity is essential for reliability.

Sankar uses the analogy to make a sound point. The bottleneck isn’t the supply of the most advanced AI, but getting it into widespread, productive use. Factories reaped electricity’s benefits when they redesigned production around electric motors. Businesses adopting AI face a similar challenge.

But the analogy can smuggle in a second, mistaken lesson: An asset’s value depends on the share of time it spends in use. That’s true of almost no durable good we own. It also misses why we invest time and resources in less tangible benefits, such as security and peace of mind.

A homeowner hasn’t wasted the guest room because someone sleeps there only a dozen nights a year. A fire department hasn’t bought too many engines just because they spend most of their time parked. And Altman’s own laptop hardly represents a tragic misallocation of processing power whenever it goes to sleep. An insurance policy that never pays a claim is, all things considered, a pretty good outcome.

In each case, we pay for something to be there when we need it. How often we expect to need it already figures into the decision, whether we do the math explicitly or simply judge it worthwhile. Low use alone tells us little about whether the investment makes sense.

The Spare Tire Isn’t Slacking

Economists have a name for this: investment under uncertainty. When future demand is unpredictable and you can’t fully recover what you spend on capacity, having an asset ready when conditions warrant can justify the investment—even if it sits idle much of the time.

The real-options literature develops a related idea: The flexibility to use an asset when it’s especially valuable has value today. An investment that looks like a losing proposition based on average use can still make sense once you account for what access is worth at the moment of need. Idle capacity isn’t necessarily mere “slack.” It can be the price of keeping an option available, and those options have value even when we never exercise them.

The same logic applies to memory in household and business computers. A consumer who buys enough memory to run a capable AI model on her own machine is buying an option: AI that works offline, carries no provider charge for each use, keeps her data on her device, and responds when she needs it without requiring access through a provider’s application programming interface (API)—the connection software uses to request a service. Using that capacity just 5% of the time doesn’t make it a bad purchase. Spare tires, fire extinguishers, and, on most days, second bathrooms would also fare poorly under that test.

Whether the option is worth buying depends on the extra hardware’s cost over time compared with paying for remote access. That’s an empirical question, and the answer will vary by buyer. Calling it a problem of “global allocation” doesn’t settle it. Millions of people spending their own money and weighing their own tradeoffs are doing the allocating—the very mechanism we otherwise praise.

The Cloud Keeps a Guest Room Too

The same logic applies to the models themselves. As Sankar correctly notes, a snapshot of benchmark scores—results on standardized tests—isn’t the comparison that matters. A firm needs to know what it costs to get useful, correct decisions over time in its actual operations, accounting for privacy, availability, and other benefits.

Developers train models to perform well on benchmarks. Even setting that aside, those tests measure performance under laboratory conditions. They don’t capture how a model performs over time on hardware a firm controls, tailored to its “operational exhaust”—the data its everyday activities generate. A model that trails the latest systems by six months on a benchmark can still outperform them where it counts if it’s available on the firm’s schedule and closely integrated into its operations. A modest car you own can likewise beat a much better one you rent by the minute.

How firms assign tasks to models supports this point. They increasingly choose cheaper models suited to particular tasks instead of defaulting to the most capable system. Models with openly available weights—the numerical settings learned during training—can lower access costs and make local adaptation easier. The latest capabilities become one input among several. Firms also pay for availability and fit.

Ironically, data centers themselves run on much the same logic, just at a larger scale. Like electric utilities, data centers that run AI models must plan for demand that arrives in bursts. An efficient facility needs capacity above its average workload.

Suppose average demand is 100 units but unpredictable peaks reach 150. Building capacity for 130 might make sense: It leaves room for surges without paying to meet every possible peak. The right capacity depends on how often peaks occur and what turning away customers would cost. “Match the average” isn’t much of a business plan.

The operator buys capacity that will sometimes sit idle because having it available for future demand is valuable. Holding that reserve can be the profit-maximizing strategy. Judged solely by average utilization, a well-run data center can look wasteful. Its idle capacity serves the same purpose as the reserve a small business or individual buys to run AI locally.

That doesn’t make every investment in local AI sensible. Firms and individuals will waste plenty of money figuring out what works. That trial and error is part of how technology spreads—and one reason the process takes time.

Nor must running AI become an either/or choice. The likely outcome is a mix: small models on phones and other devices, medium-sized jobs on local computers, and heavier work in remote data centers, or “the cloud.” That varied, evolving arrangement is precisely what central allocation struggles to get right.

The Case for Leaving Well Enough Idle

Altman’s sincerity isn’t the issue. But when prominent voices mistake low utilization for waste, a common economic error can become a policy problem. That rhetoric invites utility-style regulation that could do far more harm than good to a technology still taking shape. It also encourages the idea that large firms running AI models need to coordinate their business decisions to manage the industry effectively—a terrible idea.

Patience would serve us better. People and firms can rationally hold capacity against uncertain demand. Visible idleness may show that they’ve bought useful flexibility, not that the market has failed. As they work out how AI fits into their lives and businesses, running it locally remains a reasonable investment.

Before anyone repossesses the world’s underused DRAM, the question is whether having it available is worth the price. A memory chip can earn its keep without working around the clock.

The post Altman’s Memory Problem: The Importance of Being Idle appeared first on Truth on the Market.

Alphabet shares sank more than 12% last Thursday—the company’s worst trading day on record—after both revenue and profit missed expectations for the first time in nearly four years. Google’s core advertising business fell short of expectations, and free cash flow—the cash left after operating expenses and investments in assets such as data centers—turned negative by $9.1 billion. The company’s chief financial officer also warned that capital expenditures would increase significantly in 2027.

Amazon fared little better. Its shares closed down more than 4% ahead of its earnings release, then slid another 6% after hours when the company revealed that it had spent $75.9 billion on capital expenditures in a single quarter, while generating just $42.4 billion in cash from operations. Meta had already shed 6.98% the previous week over its own spending plans.

As The Economist, CNN, and CNBC all reported, investors increasingly question whether Big Tech’s massive AI spending will ever pay off. Alphabet’s results also stoked fears that consumers’ shift to AI chatbots threatens Google’s core search-advertising business.

It would be unwise to read too much into a few days of trading. Stock prices are noisy, and markets have swung between AI euphoria and anxiety more than once this year. But the sell-off sharpens a question for competition policymakers—one that sits uneasily with the regulatory frameworks they’ve spent the past five years building: What if the dominant platforms those frameworks label “gatekeepers” aren’t nearly as entrenched as policymakers assumed?

So Much for Entrenchment

The deeper concern isn’t that Big Tech is spending too much on AI. It’s that it may have no choice. For two decades, Google, Amazon, and Meta have made money from their position as the places users go to find information, products, and one another.

AI chatbots and agents—systems that carry out tasks on users’ behalf—threaten to insert a new layer between those users and the services they rely on. When a consumer asks an AI agent to find the best pair of running shoes and buy them, the agent—not the consumer—runs the search, compares offers, and completes the transaction. Nobody sees the ads. Nobody browses the marketplace.

This raises two related questions. First, will yesterday’s business models remain viable as AI increasingly handles transactions and access to services? Second, if they won’t, can established firms overcome Clayton Christensen’s innovator’s dilemma—the tendency of successful firms to underinvest in technologies that could undercut their core business—and compete head-on in AI markets? Neither question has an obvious answer. Both go to the heart of modern competition policy in digital markets.

Consider the European Union’s Digital Markets Act (DMA). To qualify as a gatekeeper, a firm must, among other things, enjoy “an entrenched and durable position” in its operations, or be expected to enjoy one in the near future. The regulation’s entire structure—from its per se obligations, which apply without a case-by-case assessment of competitive harm, to its abbreviated enforcement procedures—rests on the notion that the core platform services it targets will remain essential gateways for years to come. That assumption looked plausible when the European Commission proposed the regulation in December 2020. It looks considerably shakier today.

Even staunch proponents of intervention against Big Tech appear to recognize this. In a recent post on X, Tim Wu called AI agents’ challenge to “the attention-driven business models favored by the platforms” probably “the most important but slightly under the radar thing going on right now.” He added, tellingly, that those very advertising platforms own some of the major AI agents.

Wu made a similar point in a late-2025 interview, wondering what purpose Google would serve if users turned to chatbots that display few ads, and calling AI “the wildcard.”

That’s quite a concession from one of the intellectual architects of the neo-Brandeisian movement, which favors stronger antitrust intervention—and a far cry from the view of Andreas Mundt, president of Germany’s Federal Cartel Office (Bundeskartellamt). In 2024, Mundt described AI as a “first-class fire accelerator” of anticompetitive abuse, warning that it would only make Big Tech more powerful.

Disruption Doesn’t Use the Front Door

If AI agents erode Big Tech’s profits, it likely won’t be because a rival search engine or marketplace beat Google or Amazon on their own turf. It will be because consumers changed how they reach those services. Today’s ex ante regimes—rules imposed in advance to prevent competitive harm—are ill-equipped to recognize this threat.

Regulations like the DMA and the United Kingdom’s Digital Markets, Competition and Consumers Act (DMCC) take a static view of competition that also overlooks how users reach a service. The implicit assumption is that only similar rivals entering the same core platform service can erode a gatekeeper’s market power: another search engine, another app store, another marketplace. Hence their focus on lowering barriers to entry within those services through requirements to share data, make services work with rivals’ offerings, and refrain from favoring their own products.

But market power doesn’t erode only from within. It can also evaporate when the services through which users reach a platform change.

Take Google. Assume, for the sake of argument, that it enjoys a monopoly in general search (I’ve argued elsewhere that this is unlikely). Its ability to make money from that position depends on users seeing—and clicking—its ads. If users increasingly access search through AI agents, that revenue could shrink dramatically, even if no rival search engine ever gains meaningful share. There are signs this is already happening: Apple’s Eddy Cue testified last year that searches on its Safari browser had declined for the first time in 22 years, which he attributed to users turning to AI.

Ironically, the DMA may accelerate this trend. Under a decision specifying its obligations adopted in July, Google must share anonymized data on search rankings, queries, clicks, and views with eligible rivals—a group that, the Commission confirms, includes AI chatbots offering search functions. In other words, the DMA may end up subsidizing the very services that come between Google and its users, without ever producing a viable competitor in “search” as the Commission defines it.

Amazon faces a similar predicament. Whatever marketplace power it holds depends on consumers coming to Amazon to look for products. Now imagine consumers asking an AI agent to find a product and buy it from whichever merchant offers the best deal, including thousands of small retailers using Shopify to run their stores and Stripe to process payments.

This isn’t science fiction. OpenAI and Stripe launched Instant Checkout in ChatGPT, built on an open set of technical rules that lets agents complete purchases directly with merchants. Although it doesn’t appear to have gained traction with users, it’s surely a sign of things to come.

None of these small merchants could conceivably challenge Amazon as a competing marketplace. Yet collectively, through AI agents, they could erode whatever pricing power Amazon enjoys. Amazon clearly understands the threat: It has sued Perplexity to keep its Comet shopping agent off Amazon’s store, while Perplexity counters that Amazon chiefly wants to protect its advertising revenue.

Competitive conditions in one market can change dramatically even when no rival enters that market. Analyzing competition only within narrowly defined core platform services systematically overestimates how long incumbents can hold their positions.

Spending Like There’s No Tomorrow

This dynamic helps explain why incumbents are pouring unprecedented sums into AI. Big Tech is expected to spend more than $700 billion this year on data centers, chips, and energy. Alphabet anticipates capital expenditures above $300 billion in 2027, while Amazon is borrowing heavily to fund its own expansion. Meta has gone so far as to propose a new class of nonvoting shares.

The threat of losing direct access to customers makes this spending less puzzling. Incumbents rightly recognize that letting someone else control the AI layer could reduce their core services to interchangeable suppliers—data pipes and fulfillment networks serving another firm’s agent. That firm would control the customer relationship and capture the profits that come with it.

Overspending on AI seems like a rational insurance policy against that outcome. Hence Google’s push with Gemini, which now has more than a billion users, and Amazon’s development of its own shopping agents, Rufus and Buy for Me, even as it fights third-party agents in court.

Whether these bets pay off is anyone’s guess. But the incumbents themselves clearly don’t believe their positions are “entrenched and durable.” They’re behaving like firms that fear for their survival.

Please Hold While We Amend the Future

All of this has at least four implications for competition policy.

First, the “entrenched” gatekeepers policymakers describe may not be so entrenched after all. Positions that seemed unassailable a few years ago now depend on technological shifts no regulator foresaw.

Second, well-intended rules may both hobble incumbents in AI markets and hasten their loss of direct access to customers. Restrictions on combining data, tying products (making access to one conditional on taking another), and favoring their own services make it harder for incumbents to use existing assets to develop new AI services. The commission’s recent Android decision, for instance, requires Google to give rival AI assistants access to 11 Android features on equal terms with Gemini. Meanwhile, access and data-sharing mandates hand AI rivals the keys to incumbents’ platforms.

In the worst case, this combination could help tomorrow’s AI leaders consolidate their power—precisely the outcome these policies ostensibly aim to prevent. Some would welcome that tradeoff: Wu has said that, if he could redo the Google case, he would keep Google out of AI entirely. But removing one of the industry’s most formidable competitors from the race is a curious way to encourage competition.

Third, the assumptions about competitive harm behind ex ante regimes already look outdated. The DMA was adopted in September 2022, less than three months before ChatGPT launched. Its list of core platform services and its obligations reflect a world of search boxes, app stores, and marketplaces, not AI agents.

The question isn’t just whether those obligations need updating, but whether regulators can update them fast enough to keep pace with AI markets. The commission’s first DMA review concluded that the regulation needs no revision. Even if the commission changed course, amending the DMA would take years. By then, AI markets would likely have moved on again.

Fourth, and most fundamentally, policymakers simply don’t know what tomorrow will look like. That uncertainty undermines inherently backward-looking tools like ex ante regulation, which rest on the assumption that a service that’s valuable, unique, and lacking rivals today will remain so tomorrow. Case-by-case enforcement under traditional competition law is far from perfect. But at least it requires authorities to establish harm in markets as they actually stand, rather than freezing yesterday’s market structure into law.

Obsolescence Comes for the Rulebook

This week’s sell-off may or may not mark a turning point. Alphabet, Amazon, and Meta could emerge from the AI transition stronger than ever—or follow earlier tech giants that once seemed unassailable. We simply don’t know.

What we do know is that yesterday’s ex ante competition regulations—if they were ever needed—were built for a world that no longer exists. Applying them mechanically to fast-moving AI markets risks creating the very competition problems they’re meant to solve.

The market has moved on. The rulebook hasn’t.

The post AI Is Picking the Locks on Europe’s Gatekeeper Rules appeared first on Truth on the Market.

Sure. If another company comes along with lower costs, it can drive prices down. That’s always a possibility, and an important one in real markets. But is there anything else competition can do in the more traditional sense, where new rivals take time to enter a market but eventually erode profits?

There’s one more way to compete. Suppose a company comes along that has no technological edge or lower costs. It can still compete on pricing—by changing the pricing structure, rather than simply setting a lower price.

The rival could charge an upfront fee, offer a bundled deal, or charge different customers different amounts. That last possibility takes us to price discrimination, which is usually described as simply a way for sellers to charge customers more.

The Federal Trade Commission (FTC) appears to share that view. It just ordered nine companies to report whether they charge different people different prices. The FTC worries that companies use data “to estimate the highest price they can charge each consumer.”

But price discrimination can also be how a competitor offers customers a better deal and wins their business. To see how, we need to start with the problem of paying for fixed costs—the expenses a business incurs regardless of how much it sells.

Zero Profit Isn’t Mission Accomplished

Consider a customer buying electricity. The first unit is worth $120 to her, the second $30, the third $20, and the fourth $5. Producing each unit costs $10. If two suppliers have exactly the same costs, competition will drive the price down to $10—the marginal cost, or the cost of producing one more unit.

If you charge more than $10, your competitor can offer a lower price to win the business. Charging less than $10 loses money. We arrive at the usual competitive result: price equals marginal cost. This is the basis for much of how economists think about pricing.

Let’s add one small complication. Someone has to install the generator. Suppose installation costs $100, but nothing else changes. Production still costs $10 per unit, and our customer still values the units at $120, $30, $20, and $5.

Suppliers compete by making binding pricing offers before installation. The winning supplier installs the equipment and supplies the customer for the agreed-upon period.

Marginal-cost pricing no longer works because it doesn’t cover installation costs. At $10 per unit, the customer buys three units and pays $30—just enough to cover production, with nothing left for installation.

Notice I’ve rigged the numbers so that installing a generator is still worthwhile. Those three units produce $170 in benefits at a total cost of $130. But we need something besides marginal-cost pricing to generate those benefits.

So perhaps competition settles on a higher electricity price. A supplier charging $120 sells one unit, pays $110 in costs, and earns $10. A rival can offer $115 and earn $5. Suppliers can keep undercutting each other until the price reaches $110, where the winner just covers its costs.

Zero profit! Competition has worked!

Except the customer buys only one unit. The additional units, which are still worth more than they cost to produce, go unsold. That’s not great.

A new rival can’t fix this problem by cutting the per-unit price further. The supplier already earns zero economic profit. Dropping the price to $30 to sell the second unit would lose money, so no one does it.

The problem should be pretty obvious. You need to cover the fixed cost, so charge separately for it.

But now we’re beyond simple marginal-cost pricing. The basic Econ 101 model doesn’t allow that. You draw a line across the graph at “the price,” which means the per-unit price: $10 per unit of electricity, with every unit priced the same. The standard Marshallian and Walrasian models—the textbook frameworks behind our “price equals marginal cost, so we have competition” result—restrict suppliers to a single per-unit price, known as linear pricing.

Maybe we need to be more creative. You’re actually selling two products: installation and electricity. Charge separately for each.

That can solve our numerical problem, but it raises a question about the theory. What exactly is the product whose price competition drives down? Is it a unit of electricity, a connection, or an entire package delivered through that connection? Once we allow suppliers to compete over these arrangements, the familiar instruction to “take the price as given” needs a little more explanation. The arrangement itself is something firms can compete over.

The competitor in the model is just as smart as we are. It sees the problem and offers a different pricing strategy: $105 for installation, then $10 for each unit consumed.

Now we’re at the efficient outcome: the customer buys all three units worth more than they cost to produce. She pays $135 for electricity worth $170, leaving her with $35 in consumer surplus—the value she receives beyond what she pays—compared with just $10 under the original per-unit price.

Just by changing the pricing—without any technological improvements—the entrant earns a $5 profit and leaves the customer better off. A rival can profitably improve the customer’s deal even when the existing supplier already earns zero profit.

Pricing is an avenue for competition.

The opportunity arose for the usual reason: there was still deadweight loss, or potential gains from trade that went unrealized. The second and third units were worth $50 to the customer but cost only $20 to produce. Changing the pricing captures that $30 in forgone gains. The entrant gives $25 to the customer and keeps $5. Its total bill is actually higher than the old supplier’s, but the customer gladly switches because she gets more valuable service.

Other suppliers, seeing the crafty pricing strategy, can offer better terms. If the installation charge exceeds $100, a rival can lower it and still earn something. Competition pushes the installation charge down to $100, plus $10 per unit, leaving the customer with all $40 of the surplus.

We could also change the invoice. A package of three units for $130 would achieve the same result, without separate installation and electricity charges. Either way, competition is for the full contract.

We need a broader conception of how competition plays out. Louis Makowski showed how nonlinear pricing—arrangements where the total bill isn’t simply one per-unit price multiplied by quantity—can fit within a theory of perfect competition. Competition can discipline a whole pricing schedule, even when there’s no single per-unit price to take as given.

The Generator Won’t Pay for Itself

Nothing above should be controversial or surprising once it’s laid out this way. But we can carry the logic further.

Suppose one generator can serve two customers. A workshop wants one unit worth $120, while a household wants one unit worth $40. Installation still costs $100, and each unit of electricity still costs $10 to produce.

Start with the contract we just settled on: a $100 connection charge plus $10 per unit. The workshop signs up and pays $110, but the household declines. The supplier covers its costs.

Could a rival offer a lower connection charge to attract the household? The charge would have to fall to $30 or less. But collecting $30 from each customer would bring in only $60 toward the $100 installation. Offering the same contract to everyone leaves us stuck again.

Imagine instead that a rival offers the workshop an $80 connection charge and the household a $25 connection charge. Both still pay $10 per unit. The workshop’s total bill falls from $110 to $90. The household’s bill is $35. Their connection payments total $105, covering installation and leaving the entrant $5.

The rival has found another customer to help pay for the generator. The household contributes $25 toward installation, allowing the entrant to reduce the workshop’s bill by $20 and keep $5. Again, creative pricing gives both customers a better deal, even though the supplier it displaced was earning zero profit.

These customers receive identical electricity, and the supplier faces the same cost to serve each. There’s no cost-based reason for the price difference. This is pure price discrimination. Still, everyone prefers the discriminatory pricing system to the uniform price.

Further competition can squeeze out the remaining profit, but it won’t equalize the prices. Suppose a competitor offers total bills of $33 for the household and $87 for the workshop. That covers the $120 total cost. A rival can’t profitably win both customers by offering lower bills, because it wouldn’t cover its costs. Nor can it profitably serve either customer alone at a lower price: installing the generator and supplying one unit would cost $110.

Several ways of dividing the bill could survive these tests, but equal prices won’t. At $60 each, the household would decline.

Someone has to cover the fixed cost. There’s no way around it. Generally, those who value the service more will have to contribute more. When prices depend on value rather than cost, that’s price discrimination.

As a prediction about how markets work, we should expect price discrimination where fixed costs are high. Airlines have long used it because the fixed cost of a flight is high, while the marginal cost of carrying another passenger is low. Charging different fares helps support more flights.

Software is another example. High development costs and low marginal costs give sellers a reason to charge some customers, such as business users, much more to finance development while keeping software available to students and other price-sensitive users. For drugs, this happens across countries.

When assessing whether price discrimination is beneficial, fixed costs add another consideration: it can make investment worthwhile. That’s why I always stress fixed costs when discussing price discrimination. The standard Econ 101 model, even when it shows that price discrimination helps capture more gains from trade, assumes the product will be available either way. It leaves out whether the seller can cover the investment needed to offer it at all.

It’s also why I’ve proposed “more prices” in electricity to address concerns about data centers.

There’s More to Competition Than Undercutting

Okay, let’s move from simple examples to why this matters. I was listening to an episode of Odd Lots in which Lindsay Owens spends a long time worrying about pricing tools. For example, she doesn’t buy that charging some customers more would provide “a cross-subsidy for everyone else.” But one purpose of price discrimination is to offer a lower price to someone who wouldn’t buy at the uniform price. That possibility exists even in the simple case without fixed costs.

She also argued that targeted discounts for budget shoppers mean higher prices for everyone else. They could, yes, but they need not. In our example, attracting another customer lets a rival spread the fixed cost and lower the existing customer’s bill to win its business.

I’ve obviously chosen numbers that make the benefits clear and simple. Does that mean every pricing scheme helps consumers and increases the total gains from trade? No. A seller can use price discrimination to charge more on average, leaving consumers with less surplus. It may have an incentive to grab a bigger slice of a smaller pie.

But that incentive exists on both sides of every market. Sellers want higher prices; buyers want lower prices. It’s also not the only force at play. Instead of just fighting for a bigger share of a fixed pie, there’s also an incentive to find a bigger pie. Pricing schemes, including price discrimination, can be one way of finding that bigger pie.

The entrant wants a profit. It finds one by changing the pricing arrangement, creating more gains from trade, and offering customers enough of those gains to win their business. Then rivals copy the arrangement and compete over its terms. That is competition at work. Even when profits under the old pricing system have disappeared, the search for a better deal continues.

Owens represents a broader view. As I said, the FTC is making a big deal about personalized pricing. Many people share her perspective, which is why I spend so much time on it. They treat price discrimination as anticompetitive by definition, whether they use that language or not, because it departs from pricing at marginal cost. That misunderstands the competitive process.

This view was once even more prevalent. Courts and economists treated exclusive dealing, tying, resale price maintenance, and vertical mergers much the same way. These practices—agreements to deal exclusively, sell products together, set resale prices, or combine businesses at different stages of production—were presumed to reduce competition.

Since then, economists have built large bodies of research showing how firms use each practice to win customers. In response, courts have dropped per se condemnation—treating a practice as unlawful without examining its effects—for several of them.

That work begins with the same problem: a single price can leave units unsold that are worth more to the buyer than they cost to produce. Discounts, exclusive contracts, and two-part prices can help buyers and sellers trade those units. Public discussion of price discrimination hasn’t caught up.

Across those bodies of research, expanding the gains from trade has proved the more powerful force. Finding more of those gains is the big deal.

The post The Price of Playing Fair: How Charging Everyone the Same Can Leave Consumers Worse Off appeared first on Truth on the Market.

Here’s coverage from The New York Times, The Washington Post, The Wall Street Journal, and, for a bit of local flavor, the Albuquerque Journal.

That’s a big number for most human folks, if not for the large language models among us, but I have a few questions.

That Pesky, Precious First Amendment

This has been billed as a privacy case, but it wasn’t—or isn’t—just that. The alleged violations concerned Facebook’s representations “about its data practices and its efforts to address misinformation and hate speech.” That is, the allegations covered data privacy, data security, and content moderation—the decisions platforms make about what users can post and see.

As my International Center for Law & Economics (ICLE) colleague Ben Sperry points out in a compact thread on the platform formerly known as Twitter, the content moderation claims raise distinct constitutional issues. A social media platform’s content moderation can qualify as protected speech under the First Amendment, akin to the editorial decisions of, e.g., a newspaper. See, for example, Miami Herald on restraints on the press and Moody v. NetChoice on state-law restrictions on platforms’ content moderation policies.

New Mexico argued that it wasn’t restraining or penalizing content moderation decisions as such, but simply misrepresentations about those policies and practices—in essence, false or misleading marketing. The trial court accepted the distinction. But as Ben points out, “it is very hard to separate those promises from the editorial discretion Meta retained over enforcement of its standards.”

We might add that it’s hard to judge whether Meta’s editorial decisions matched its stated policies or assurances without judging those decisions themselves. And it’s very hard to accept enforcer or trial court workarounds without compromising—or even gutting—the First Amendment’s protections.

Even under the most favorable reading of New Mexico’s argument, the First Amendment question is likely to complicate the state’s defense of the ruling on appeal.

Attempts to regulate speech—including expressive, political, religious, and scientific speech, and, not incidentally, access to information—are nothing new. But recent federal and state efforts have taken some novel tacks. Among them are attempts to apply federal antitrust laws, as well as federal and state consumer protection laws, to this dubious task.

Ben recently had a clear and accessible op-ed in USA Today on this issue, spurred by reports that the U.S. Department of Justice’s (DOJ) Antitrust Division is investigating whether five major networks—ABC, CBS, Cable News Network (CNN), NBC, and Fox News—violated the Sherman Act. The apparent theory is a group boycott: a coordinated refusal to do business with someone.

The networks briefly suspended their pooled coverage of President Donald Trump, in which journalists share reporting with other outlets. They did so in response to the White House’s decision to bar CNN journalists from the grounds, preventing them from participating in that coverage. The networks’ decisions to suspend coverage seem conspicuously expressive, not anticompetitive, and the investigation looks manifestly dubious. Am I being optimistic when I wonder whether it’s grandstanding for the White House?

For other examples, Ben and I have a joint post on a statement of interest submitted by the Antitrust Division in a private antitrust suit. I also have a post on a Federal Trade Commission (FTC) inquiry into “tech censorship,” with more thorough ICLE comments from Ben, Geoffrey Manne, and me in response to the FTC’s request for information.

Ben and I also published an article, “Is There an Empty Set at the Intersection of Antitrust and Content Moderation,” as our contribution to a Concurrences collection on the topic edited by Bilal Sayyed.

Ben and I discuss the FTC’s artificial intelligence (AI) policy statement here, and extended ICLE comments are here from Ben, Geoff, and me. Here are Ben and Jeffrey Westling on jawboning by the Federal Communications Commission (FCC).

And just to round things out for the energetic or choosy readers among you, there’s more from Ben (here, here, here, and here).

The Land of Enchanting Penalties

First Amendment issues aside, $25 billion is a historic penalty indeed. To put it in perspective, in July 2019, the FTC announced “a record-breaking $5 billion penalty” against Facebook, along with substantial—and costly—restrictions on its practices and compliance obligations. The monetary penalty alone was, as the commission announced, “the largest ever imposed on any company for violating consumers’ privacy and almost 20 times greater than the largest privacy or data security penalty ever imposed worldwide.”

Now, in 2026, a New Mexico state judge has ordered a penalty five times that record-breaking federal penalty. It’s closer to four times as much after adjusting for inflation, but the point remains: Pending appeal, New Mexico stands to recover several times any federal (or, for that matter, Europe-wide) data penalty.

And that’s supposed to be giving Meta a break. Judge Francis J. Matthews ruled that “[t]he state is awarded and shall recover a civil penalty against Meta in the amount of $219,498,606,000, which penalty is reduced to the amount of $25,000,000,000.” That’s $25 billion plus interest, of course, “to accrue at fifteen percent (15%) per annum until paid in full.”

So … a penalty of nearly $220 billion, but for you, Meta, only $25 billion. Such a deal.

According to Matthews, “[t]he harm in this matter is difficult to overstate.” Perhaps, but he sure did give it the old college try.

To some extent, it’s a pile-on penalty. The FTC’s $5 billion penalty, plus commitments to change Facebook’s practices, stemmed from alleged violations of a 2012 FTC order. The commission revisited that order in the wake of the Cambridge Analytica controversy over the use of Facebook users’ data. New Mexico Attorney General Raúl Torrez now tells us that his state’s “case arose from Facebook’s conduct surrounding the Cambridge Analytica scandal.”

To be sure, neither the state case nor the FTC’s case was confined to conduct or consequences tied to Cambridge Analytica. Then again, they weren’t not about Cambridge Analytica.

There was also a much smaller fine under the United Kingdom’s Data Protection Act 1998, which capped the amount. No specific penalty followed under the General Data Protection Regulation (GDPR)—unsurprising, as the Cambridge Analytica events preceded its effective date by several years. But the matter drew European Union (EU) attention and influenced subsequent EU regulation and enforcement.

On the theme of piling on, it’s hard to avoid flights of basic computational fancy. What if every state did it? Fifty times $25 billion equals $1.25 trillion. Make it 51—the 50 states plus the District of Columbia—and we’re up to $1.275 trillion. If that seems about the same, in one sense it is; in another, you’ve just rounded off $25 billion.

Suppose we weight the awards by population. For many of the allegations, the judgment did, after all, calculate violations and penalties using the state’s entire population. Based on U.S. Census Bureau figures, New Mexico’s population is approximately 0.63% of the U.S. total. Scaling the penalty accordingly would take us to roughly $3.97 trillion, without counting federal penalties or the costs of court-ordered changes to business practices.

That’s well over half the U.S. government’s total spending for fiscal year 2026 and hence—it seems safe to say—real money, even in Washington.

What if we run the numbers again using the original $219,498,606,000 penalty? I leave that as an exercise for the reader—and not much of an exercise. If you can’t do it in your head, you can do it on your phone.

In a presumably separate matter, this judgment follows hot on the heels of an Aug. 6 New Mexico judgment against Meta Platforms. The state court assessed $375 million in civil penalties and ordered another $567 million paid into an “abatement fund” for a total of $942 million. It also ordered substantial changes to Meta’s practices.

The aim was to remedy Meta’s supposed contributions, via Facebook and Instagram, to “the youth mental health crisis”—conditions that, according to the state and the court, constituted public nuisances under New Mexico law, or harms to the public at large. That judgment also came from the state’s First Judicial District Court, albeit from a different judge.

As I noted way back in 2023, in a roundup edition on “The Bureau of Let’s Sue Meta,” two commissioners—Rebecca Slaughter and Rohit Chopra—dissented from the then-record-setting $5 billion settlement because they thought it inadequate:

I commend Chopra’s dissent, if only as an oddity. He rightly pointed out that the commissioners’ analysis of the penalty was “not empirically well grounded.” At no time did the commission produce an estimate of the magnitude of consumer harm, if any, underlying the record-breaking penalty. It never claimed to.

That’s odd enough. But then Chopra opined that “a rigorous analysis of unjust enrichment alone—which, notably, the Commission can seek without the assistance of the Attorney General—would likely yield a figure well above $5 billion.” That subjective likelihood also seemed to lack an empirical basis; certainly, Chopra provided none.

As Chopra should have known, the FTC’s Bureau of Economics had weighed in and considered how to assess consumer harm. Its economists just didn’t have evidence to support an estimate anywhere near the settlement amount—or even in the same order of magnitude.

As I said at the time:

[T]he remedies appeared to be wholly untethered from the magnitude of consumer harm wrought by the alleged violations. To be clear, I’m not disputing that Facebook violated the 2012 order, such that a 2019 complaint was warranted, even if I wonder now, as I wondered then, how a remedy that had nothing to do with the magnitude of harm could be an efficient one. 

I continue to puzzle over such things.

While the attention to the dollar figure is understandable, it’s not the only large number in the news from New Mexico. As the state’s attorney general puts it, “[a] New Mexico jury found nearly 44 million violations of our consumer protection statute, and this penalty order demonstrates that those violations carry real consequences.” Lower down, the press release gives a more precise number: 43,899,725 violations.

How? For context, the Census Bureau puts New Mexico’s population at approximately 2.1 million. That’s everyone: adults, kids—including toddlers and newborns—people in their 90s, prison inmates. Everyone.

Facebook is an immensely popular platform, but nobody imagines that everyone in the state uses it—or even has internet access. According to the decision, Facebook had “1,386,648 New Mexico active monthly users as of June 30, 2020.” That’s quite a share of the population, but it’s not everyone, and it does little to explain how the count reached 44 million violations.

Nobody will be surprised to learn that the state didn’t allege, and the jury didn’t find, those 43,899,725 violations one by one. Rather, the jury found that 26 separate statements constituted willful violations of state law. Then, “[p]ursuant to the Court’s instructions,” it found that the “reasonably expected or anticipated audience” for some statements numbered 1,386,648—the estimated number of Facebook users in New Mexico in 2020. For others, the audience numbered 2.1 million—the state’s entire population.

The tally came to 43,899,720 instances of unfair or deceptive trade practices, plus five instances of “unconscionable” trade practices—conduct the law treats as grossly unfair. In brief, the findings treated some statements as harming every Facebook user in New Mexico and others as harming every living soul in the state, including people who’d never accessed Facebook.

If Meta Platforms changed its ways in response to the New Mexico judgments, precisely as the state desired and its court ordered, what would consumers who never access Meta’s platforms—including Facebook and Instagram—be protected against?

I like New Mexico, I do. I’ve trekked in the mountains near Cimarron, visited Taos Pueblo, and skied at Taos Ski Valley. I’ve even been to a Seder in Santa Fe. It’s a beautiful state. And yet, I can’t escape this sneaking feeling that … this judgment is nuts.

My objection isn’t that there couldn’t be a viable complaint against Meta under New Mexico law. I’ve looked at some of the state’s exhibits. Some allegations seem strained, at best, but I’m not a New Mexico lawyer and can’t claim expertise in the state’s law.

For that matter, I haven’t closely examined all of Meta’s representations about its data practices over the years—express or implied, on its platforms or elsewhere. So it would be hard for me to argue that none of those statements caused any New Mexico citizens harm that could support a legal claim.

Acknowledging those limitations, something seems deeply wrong here.

The FTC Act Has Standards

Looking back at that $5 billion federal penalty—and the drama surrounding it—highlights a few key questions. So does the FTC’s 2023 order to show cause, which required Meta to explain why the commission shouldn’t revise its earlier order and impose new requirements. I discussed that briefly here.

That order referred to the $5 billion penalty—really, an agreement to settle the FTC’s 2019 complaint. The complaint alleged that Facebook had violated a 2012 consent order. That earlier order, in turn, resolved a 2012 FTC complaint alleging violations of Section 5 of the Federal Trade Commission Act, which prohibits unfair or deceptive acts or practices in commerce. Of the initial complaint’s eight counts, five alleged deception and three alleged “unfair or deceptive” acts or practices. For whatever it’s worth, the 2012 consent order imposed no monetary penalty. 

Still with me? The FTC framework seems useful here because the cases involve the same defendant—now Meta Platforms, then Facebook—and overlapping conduct. More fundamentally, it offers a useful framework for assessing deception and unfairness claims under Section 5.

Congress imposed distinct limits on the commission’s unfairness claims through Section 5(n), which it added to the FTC Act through the Federal Trade Commission Act Amendments of 1994. That provision reined in the commission’s expansive interpretation of its unfairness authority.

Under Section 5(n), an act or practice cannot qualify as unfair unless it “causes or is likely to cause substantial injury to consumers,” consumers cannot reasonably avoid that injury themselves, and the injury is “not outweighed by countervailing benefits to consumers or to competition. Section 5(n) also provides that “public policy considerations may not serve as a primary basis” for a determination of unfairness, although they can be considered along with other evidence.

I like it. The provision leaves ample flexibility to address harmful conduct while linking the commission’s competition and consumer protection missions. It focuses limited resources on real and substantial consumer harm, limits the consideration of plastic “public policy considerations,” and recognizes the tradeoffs inherent in policy interventions. At bottom, it’s a law & economics provision.

No comparable amendment limits the commission’s deception authority, but that doesn’t make the authority unbounded. As the commission recognized in its 1983 Policy Statement on Deception—still cited as guidance—“[n]umerous Commission and judicial decisions have defined and elaborated on the phrase ‘deceptive acts or practices.’” To synthesize and clarify those decisions, the statement identified “certain elements [that] undergird all deception cases”:

First, there must be a representation, omission or practice that is likely to mislead the consumer. …

Second, we examine the practice from the perspective of a consumer acting reasonably in the circumstances. 

Third, the representation, omission, or practice must be a “material” one. The basic question is whether the act or practice is likely to affect the consumer’s conduct or decision with regard to a product or service. If so, the practice is material, and consumer injury is likely, because consumers are likely to have chosen differently but for the deception.

This approach recalls fraud lawsuits under common law, as well as statutory fraud claims. For a recent example, there’s the FTC’s complaint, joined by 20 states, against Dollar General. As with unfairness claims under Section 5(n), the conduct must cause, or be likely to cause, concrete injury to consumers acting reasonably in the circumstances.

I recognize that New Mexico wasn’t applying Section 5’s prohibition on unfair or deceptive acts or practices to Meta. Still, it seems instructive to note that much of the state’s complaint would fall short under the FTC’s framework.

For example, a summary sheet of the “state’s allegations of false or misleading statements about users’ control of their data” cites a Jan. 24, 2018, Facebook post by Sheryl Sandberg, in which she states that “we don’t share personal information with advertisers without permission.” Her full post reads:

Sharing an important post from Mark that explains our business model. Some people have questioned this model because they worry about privacy given that our business is advertising. But showing people relevant ads and protecting their privacy are not at odds – we do both. We don’t sell peoples’ data and we don’t share personal information with advertisers without their permission. What we do is use information in a privacy-safe way that allows advertisers to reach people who may be interested in their products.

This is especially important for small businesses that now have access to the tools that previously only big companies could afford. It helps them grow and hire – and it helps us provide our services for free to billions of people so they can express themselves and connect with people they care about all around the world. 

That all seems fairly reasonable … and broad, and perhaps somewhat vague as policy assurances go. Perhaps that’s to be expected in a brief Facebook post.

To be clear, I’m not arguing that the post couldn’t possibly have misled a privacy-sensitive New Mexico consumer acting reasonably in the circumstances. Suppose the consumer read it, and suppose it didn’t match Facebook’s practices at the time. Even then, one wonders whether privacy-sensitive consumers might reasonably have dug a little deeper.

But there’s a preliminary question: How many such New Mexico consumers read the post? Of those who did, how many found it misleading in a way that affected their choices? How many would have abandoned Facebook had the post accurately described its data practices?

Surely, the answer isn’t 2.1 million New Mexico consumers, acting reasonably or otherwise. Peel off infants, people in comas, people living off the grid, and those on the grid who never access Facebook. Then peel off young people—like my kids in their mid- to late 20s—who prefer other social media tools.

And surely, it’s not every regular Facebook user in the state. Did 10% of those users see the post? Dubious. One percent? That still seems dubious. But if it was 1%, what share did it materially mislead? And what harm did they suffer?

Privacy violations can cause real and substantial harm. Liad Wagman and I discuss various potential consumer harms—and how to identify and measure them—in “The Law & Economics of Privacy.” I won’t rehash all that here.

Consumers might change their online behavior because of those risks, real or perceived. They might also change it because they like or dislike a service’s price or other features, including the privacy policies its provider advertises or follows.

As I said, I’m not arguing that the post couldn’t have materially deceived or misled anyone. I’m wondering whether it did. How did the state—or the court—assess or estimate how many people the statement materially misled and the harm they suffered as a result?

That’s a rhetorical question: They did no such thing.

We can run through the same drill with other statements on the summary sheets, from emails to and from Mark Zuckerberg and others at Facebook, to newspaper quotes, to published data policies. Indeed, as Wagman and I note, there’s a body of “research indicating that many consumers do not read the privacy policies posted by firms and other organizations, and that many who do fail to comprehend those policies.”

As articles by Howard Beales and Timothy Muris and James Cooper and Joshua Wright point out, some of that may reflect rational ignorance: People may reasonably decide that learning the details isn’t worth the time and effort. As Wagman and I explain:

A consumer might attach low value to reading and comprehending a privacy policy because the consumer attaches little value to privacy generally, or because they doubt that the policy will address their individual privacy concerns or priorities. Or they might doubt the nexus between the policy and their risk of harm, perhaps because they doubt the efficacy of such policies. In any case, they may doubt that the marginal benefit of search required to understand the policy will exceed the marginal cost.

Again, false or materially misleading privacy assurances can cause harm. My point is that the state didn’t demonstrate such harms here, and the New Mexico court required no such showing.

The court required none of these basic elements of a sound FTC unfairness or deception case—either to find Meta liable for the statements or to calculate the penalty. Not even the … cough, reduced penalty of $25 billion, much less the original $219,498,606,000. Both figures rested on statutory per-violation penalties under New Mexico law, independent of demonstrable harm.

For that matter, how many New Mexico consumers made lasting changes to their online behavior because of the news about Cambridge Analytica? That scandal involved the firm’s harvesting of data from up to 87 million Facebook users, out of more than 2 billion users in 2018.

The decision found that Facebook had “1,386,648 New Mexico active monthly users as of June 30, 2020.” That’s roughly two-thirds of the state’s entire population, more than two years after the scandal broke and about a year after the FTC’s highly publicized 2019 complaint.

How many changed their behavior in response to that complaint or the ensuing consent order? How about the filing—widely reported in New Mexico—of the state’s complaint?

And what about countervailing benefits? Most New Mexico consumers apparently continued using Facebook after highly publicized complaints from federal and state enforcers. That suggests a “revealed preference”: Their choices indicate that they still valued the service. Yet the state’s determinations of liability and penalties considered no countervailing benefits—or any tradeoffs at all.

Perhaps these lacunae are unsurprising, not to mention European. As I noted above, even the relatively paltry $5 billion Facebook agreed to pay to settle the FTC’s allegations bore little, if any, relation to an assessment of the consumer harm the alleged violations caused. And may I be forgiven, and not just by Rohit Chopra, for using “paltry” in this context.

That was a negotiated settlement based on statutory penalties the company might have faced for allegedly violating the 2012 consent order.

The Fine Line Between Enforcement and Extraction

Perhaps Attorney General Torrez was right to call the decision “a historic day for New Mexico and for consumers across the country.” Was it a good one?

He followed up in exceedingly broad terms: “For too long, powerful technology companies have operated as though their size and influence put them beyond the reach of the law.” Perhaps, or perhaps not. But the statement smacks of a larger anti-tech agenda, much like statements we’ve seen from federal enforcers, and not just under Lina Khan’s leadership. A sweeping grievance isn’t a concrete legal dispute for a court to resolve. Nor does it signal enforcement of the law without fear or favor.

Meta Platforms will, of course, appeal. It may gain traction—at least in challenging the penalty—on due process or other grounds, drawing on cases like State Farm v. Campbell, BMW of North America v. Gore, and Exxon Shipping v. Baker. The state court argues otherwise, and the details of any appeal remain to be seen, but I expect the firm will find at least some relief.

The recurring mismatch between statutory penalties and the harms they’re supposed to address raises more than a technical legal issue. Those penalties should work ex post by making firms bear the costs of the harm they cause, and ex ante by giving them an incentive to avoid causing it. When penalties lose that connection to harm, the concern goes beyond whether a remedy works efficiently. At some point, it becomes a question of due process and the rule of law.

In musing on an EU fine against Meta in 2023, I asked:

Are penalties supposed to bear some relationship to actual harms, or are they supposed to be arbitrary exercises of international taxation?

Cheeky, I know. But by assessing penalties based on the actual worldwide bigness and perceived worldwide badness of certain large American tech firms, European regulators seemed to be intruding on the authority of nations outside the EU—not least the United States. Not every action taken under the color of law upholds the rule of law.

New Mexico and its First Judicial District Court say they’re holding Meta accountable under the law. Under the color of law, perhaps. But the penalty looks more like a smash-and-grab raid on the tech industry. That doesn’t look very much like the rule of law.

The post The New (Mexico) Math: Multiply the Violations and Never Mind the Harm appeared first on Truth on the Market.

I am a member of Baker McKenzie’s antitrust and competition team in the Asia-Pacific region. Before joining Baker McKenzie, I spent more than 10 years as in-house antitrust counsel in the tech sector, so I tend to approach issues from a practical perspective. My practice covers all aspects of competition law, including merger control, compliance, cartels, and abuse of dominance.

Over the past several years, many countries in Southeast Asia have been considering how best to address emerging issues arising from digital platforms and artificial intelligence (AI). Policymakers look at developments in Europe, the United States, and elsewhere, but ultimately, they want to find the approach that best suits their own markets. Businesses, in turn, want to engage constructively with regulators and policymakers in that process.

That is a good segue. Do digital platforms raise antitrust problems that are new, or does the novelty of the technology create a sense that the law must be tailored to it?

Several years ago, I worked on similar questions in jurisdictions that were among the first in the region to consider platform regulation. There, the discussions were shaped both by technological development and by domestic platform markets.

In the jurisdictions that are now most active, the discussions tend to be more issue-driven. Various stakeholders, including business users, have raised concerns about the terms on which they access platforms, including high fees. Competition authorities are not price regulators, but these concerns naturally inform the policy debate.

Complaints from ride-hailing drivers and restaurants on food-delivery apps could be framed as labor-law, collective-bargaining, or contract problems. Why are they treated as antitrust problems? And are there other issues driving the push to regulate platforms?

These issues span several areas of law, including competition, consumer protection, online safety, e-commerce, and payments. Part of the interest in addressing them through competition law and ex ante frameworks, such as the European Union’s (EU) Digital Markets Act (DMA), relates to the broader regulatory tools these frameworks can provide.

Contestability is also part of the discussion. Policymakers are generally interested in fostering a dynamic digital economy with opportunities for a range of market participants.

Is there any weighing of the benefits that foreign platforms bring? They have helped business users reach customers who were previously out of reach, and the evidence suggests most of the gains from new technologies go to users rather than to the firms that introduce them. Are the perceived costs of foreign players ever set against those benefits?

Digital platforms, whether local or foreign, make a significant contribution to the region’s digital economies and drive growth, and policymakers recognize that. Much of the focus is on accountability: making sure there are effective mechanisms for consumer redress and regulatory engagement when issues arise.

Is that new? International companies without a physical presence operated in these markets long before the platforms, and authorities presumably have jurisdiction over them because they generate turnover locally. Is this a difference in kind or in scale?

From a regulatory perspective, the scale of cross-border transactions and data flows is relatively new. Traditionally, regulators could rely on a company’s local office or facility as a point of contact. In the online economy, that is not always the case.

Regulators are also observing how other jurisdictions are addressing this. For example, some have introduced local representative requirements for certain foreign operators, and these developments are being followed with interest across the region.

If I understand correctly, the goal is accountability, and competition law is the chosen vehicle, even if it fits awkwardly with disputes over platform commissions, because of the powers it confers. Ex ante rules like the DMA confer even more. Is that a fair reading?

Accountability is one consideration among several. DMA-style frameworks address competition concerns, but they also provide broader investigative and information-gathering powers. Transparency about how platforms rank and recommend content and products is one area of interest for policymakers in the region.

In Europe, the Digital Services Act (DSA) and the DMA each provide transparency-related tools, and the DMA gives the European Commission broad information-gathering powers concerning designated gatekeepers. A similar framework would give regulators in the region comparable tools. However, in considering such frameworks, it is important to take into account the specific European policy context in which the DMA was developed, which may not be directly analogous to the circumstances in each country.

So they are less concerned with theories of contestability and fairness and more attracted to the toolkit. In your opinion, is the DMA the appropriate framework for Southeast Asia? One might say it is better to have tools and not need them than to need them and not have them. What is the counterargument?

I would frame it a little differently. The considerations are varied, and local context is important. The DMA was developed in response to a specific set of circumstances in Europe, and the market dynamics in Southeast Asia differ in a number of ways. Two considerations stand out to me.

First, many competition regimes in the region were adopted relatively recently, and authorities are continuing to build experience through enforcement and policy activities. There is a good case for allowing the existing frameworks to be fully tested before deciding whether additional tools are needed.

Second, ex ante regulation is sometimes seen as a quicker and simpler alternative to case-by-case enforcement. Experience to date, however, suggests that ex ante regimes can also involve complex and lengthy processes, so it is worth weighing the tradeoffs carefully.

Europe’s experience bears that out. Specification proceedings under the DMA have required the Commission to decide who gets access to data, which data, how recent, and on what terms. Telling Apple it can charge something, but not 30%, comes close to price regulation. Before we move on, do Southeast Asian authorities already have the tools we take for granted in Europe and the United States, such as behavioral and structural remedies, document-production powers, and mandatory merger notification? Or would the DMA take them from very little to the most powerful version of those tools?

Broadly, I think the foundations are in place. When countries in the region adopted their competition laws, they generally draw on international best practices, including guidance from bodies such as the Organization for Economic Cooperation and Development (OECD) and the International Competition Network (ICN). Most have adopted internationally recognized frameworks, and authorities continue to strengthen their capacity and expertise in applying them.

Policymakers in the region continue to study the DMA, the DSA, and other regulatory models closely, but increasingly as case studies rather than ready-made templates. The discussion has shifted from whether to adopt a particular model to which elements, if any, fit local market conditions and institutional settings.

In practice, this calls for a more holistic approach: identifying the specific harm, considering the most appropriate response, and ensuring alignment with existing domestic rules and international frameworks. That is more likely to produce coherent outcomes than adding new layers of regulation, particularly as provisions designed for one legal and institutional context may operate quite differently in another.

What prompted that change? The DMA was initially presented as modern, fast, and nearly costless. Did authorities become aware of its costs or of better alternatives? Is there a geopolitical dimension?

Policymakers in the region are monitoring the implementation of the DMA closely, including the ongoing debate around its effects on innovation and competitiveness. It is still relatively early to assess its outcomes, but that debate is naturally informing thinking in the region. 

International academic and policy exchange has also influenced regulatory thinking in the region. Many experts in the region have studied abroad or worked abroad and bring home the experience of different regulatory approaches, which can contribute to the development of new competition, digital, and AI regulation.

That suggests experts play a central role in shaping these reforms and that adopting these ideas lets authorities signal they are up to speed. It also strikes me as ironic that a call to separate integrated companies is treated as novel when antitrust had that debate a century ago. Why are policymakers drawn to that view in particular?

Debates about integration and market structure have a long history in antitrust, and recent scholarship has brought renewed attention to them. It is valuable for policymakers to have access to a broad range of perspectives, including economic evidence and international experience. 

Anu Bradford’s “Brussels Effect” thesis has been influential in presenting Europe’s regulatory reach as close to inevitable. Has that shaped thinking in the region?

The influence of European regulation has varied by area. In data-protection regulation, the General Data Protection Regulation (GDPR) has been an influential reference point for many countries. In digital competition regulation, policymakers in the region increasingly emphasize developing approaches suited to their own markets. 

From a business perspective, developing an appropriate compliance approach is not always a straightforward exercise. Businesses may need to develop and refine their approach iteratively, taking into account the particular legal requirements, regulatory expectations and circumstances of each market.

Let’s turn to Vietnam, which is a good case study. The International Center for Law & Economics (ICLE) has submitted several rounds of comments to the Vietnamese authorities this year. Can you tell us what is happening there and why it matters, starting with some history if useful?

Vietnam’s current competition law dates from 2018. 

Vietnam has seen significant legislative activity in recent years across the digital economy, including new frameworks addressing data, digital transactions, and more recently AI. As part of this broader development, this year, the competition law and implementing decrees are being revised.

This matters because it reflects an evolving debate about where digital platform regulation should sit within Vietnam’s broader regulatory framework. The move toward competition law also raises important questions about how these rules will interact with existing competition concepts and enforcement mechanisms.

I was surprised when I read the draft. I expected something closer to a copy of the DMA. Instead, the self-preferencing rules are not limited to digital companies. A dominant supermarket chain that gives its own milk more prominent placement could be liable. How did that happen?

The current draft no longer refers specifically to digital platforms, meaning that the provisions are not formally limited to the digital sector. However, some of the provisions continue to address conduct and practices commonly associated with digital platforms, while potentially applying more broadly to businesses in other sectors. 

Continued dialogue between stakeholders and the drafting team will help ensure that the final law works well in practice.

Let’s take a broader view. What role do these frameworks play? Are they about making markets work better for consumers, about having some control over foreign firms seen as beyond reach, or about trade disputes? Or are there bigger forces at play?

These reforms take place against a broader backdrop that includes international economic relations and domestic policy priorities. The digital economy is an important area of development, and governments want effective legal frameworks to support and oversee it.

The post The View from Southeast Asia: A TOTM Q&A with Hee Eun Kim appeared first on Truth on the Market.

Canada is heading to a summit with the EU in Montreal, reportedly seeking this “associated country” status under the proposed EU Cloud and AI Development Act (CADA). United Kingdom ministers are reportedly lobbying for the same. Neither government has made clear what it wants, but the status on offer falls well short of a trade deal.

The proposal’s Article 18 would allow businesses from recognized countries to own cloud operations located entirely within the bloc and walled off from operations elsewhere. That’s hardly an enticing offer for the EU’s allies. It’s also a poor bargain for the EU itself, which would unnecessarily restrict access to cloud services it needs.

The European Commission introduced CADA earlier this year, proposing four “assurance levels” for cloud services—computing resources accessed over the internet—purchased by public bodies. These levels set increasingly strict requirements for where providers operate, where they store data, and who can own or help run the service. The higher the level, the more the provider must keep its operations within the EU and limit outside involvement.

Even Level 1—the default unless a public buyer opts out—requires a provider established in the EU, with infrastructure and data located there. Level 2 requires the provider to deliver the service entirely within the EU and guarantee that no one will access the data from outside the bloc.

Level 3 goes further: Only EU citizens may participate in providing the service, and the provider must be free of control from outside the EU. This is where associated-country status comes in. The commission can recognize non-EU countries, allowing businesses from those countries to own cloud operations located and run entirely within the EU.

At Level 4, even that opening disappears. The ban on outside control admits no associated-country exception.

A Very Conditional Welcome

Article 18 offers a poor bargain for countries like Canada, the United Kingdom, and the United States. All they stand to gain is permission to own an EU business staffed by EU citizens working within the EU, with no data processing or reuse allowed outside it.

Even that very limited permission is precarious. The commission could revoke it unilaterally, and it depends on “adequacy” decisions under the General Data Protection Regulation (GDPR)—EU findings that another country provides sufficient protection for personal data. EU courts can strike down those decisions, as the Schrems cases demonstrated when they dismantled earlier EU-U.S. data-transfer frameworks. The mechanism also faces political threats: An amendment proposed in the European Parliament would delete Article 18 altogether.

The recognition criteria make the offer shakier still. They treat export controls, economic sanctions, and laws authorizing government access to data as grounds for disqualification—even though nearly all countries, including EU members, use such measures. Applied strictly, the criteria would leave virtually no jurisdiction eligible, including the EU itself. Recognizing a country would therefore require some creative interpretation, barely disguising a fundamentally political choice and potentially inviting a court challenge.

Keeping the Best Tools Out

Europe’s public sector needs cutting-edge technology, especially for sensitive work such as intelligence and cyber defense. Only non-EU companies currently provide these tools, and they deliver them through cloud services.

Some government functions—such as safeguarding nuclear launch authorization or records identifying intelligence sources—may justify keeping the relevant systems entirely within national borders and excluding even allies. Taken seriously, though, that logic would also mean excluding other EU countries, which proves to be an awkward wrinkle for an EU-wide policy. 

Meanwhile, CADA would extend those restrictions to public services that need access to the best available technology, wherever it comes from. Localization may make sense for sensitive government functions that can get by with basic or older technology. But for work like cyber defense, denying access to the best available tools amounts to unilateral disarmament.

Safeguards Over Passports

A fair digital trade deal would address real risks with effective safeguards, rather than checking passports—especially when some EU countries are willing to sell theirs or permit businesses under de facto control from adversary countries. For data security, procurement rules should focus chiefly on technical protections: preventing providers from retaining customer data, keeping encryption keys in customers’ hands, and using computing environments that verifiably shield data even during processing.

Those protections still depend on trust, so there’s a sensible case for recognizing allies. But recognition should rest on an explicit political judgment that weighs the benefits of trade, the risks a country poses, and the strength of its safeguards—including the rule of law and independent courts. Criteria the EU itself couldn’t meet offer little assurance of anything.

A cloud policy should protect Europe’s data without putting its capabilities out of reach.

The post The EU’s Fair-Weather Cloud Partnership appeared first on Truth on the Market.

In Corner Post, Inc. v. Board of Governors of the Federal Reserve System, a panel of the 8th U.S. Circuit Court of Appeals affirmed a decision vacating the Federal Reserve’s interchange-fee cap under Regulation II. Merchants pay interchange fees to the banks that issue their customers’ debit cards. The rule capped them at 21 cents per transaction, plus an allowance for fraud losses and an additional penny for qualifying fraud-prevention measures.

Writing for the panel, Judge David Stras concluded that the Fed improperly included fixed costs that weren’t specific to individual transactions. The Durbin Amendment, he explained, requires the Fed to consider the “incremental cost” of processing a particular transaction. The decision carries added significance after Loper Bright Enterprises v. Raimondo, which ended the practice of deferring to reasonable agency interpretations of ambiguous statutes.

But rejecting the Fed’s cost calculation doesn’t settle how high fees may go. As the International Center for Law & Economics (ICLE) argued in its amicus brief in Linney’s Pizza, LLC v. Board of Governors, a parallel case before the 6th U.S. Circuit Court of Appeals, the Durbin Amendment gives the Fed discretion to determine what fees are “reasonable and proportional.” Even under the 8th Circuit’s restrictive reading, that language may leave room for a much higher cap.

That inquiry should account for how payment networks work. They serve two groups—merchants and consumers—and benefits that attract consumers also make the network more valuable to merchants. ICLE’s argument offers the Fed a legal basis for recognizing those benefits when setting the cap. The court has ordered the agency to reconsider the costs. What those fees buy deserves a look, too.

The Swipe at the Fed

The path forward starts with what the 8th Circuit found wrong with the Fed’s approach. The Durbin Amendment, part of the Dodd-Frank Wall Street Reform and Consumer Protection Act, requires debit-card interchange fees to be “reasonable and proportional to the cost incurred by the issuer with respect to the transaction.” The issuer is the bank that provides the customer’s card.

The statute also tells the Fed which costs to count. It must consider the “incremental cost” of authorizing, clearing, or settling a particular transaction—the steps that approve a payment and move the money between banks. It must exclude costs that aren’t specific to a particular transaction.

When it adopted Regulation II, the Fed treated certain fixed processing costs, network processing fees, transaction-monitoring costs, and fraud losses as transaction-specific. It relied on Chevron deference, the doctrine that allowed agencies to choose among reasonable interpretations of ambiguous statutes.

In Corner Post, the 8th Circuit rejected that approach. Applying Loper Bright, the panel read the statute independently and interpreted “incremental cost” to mean marginal cost—the extra expense of processing one more transaction. The Fed, the court held, had failed to distinguish that expense from fixed or other costs that weren’t specific to an individual transaction. Its uniform 21-cent cap therefore rested on the wrong cost calculation. The panel affirmed the lower court’s decision vacating it.

For the merchants backing the litigation, the next step seems obvious. Count only the direct expense of processing a swipe, and slash the cap accordingly. But that conclusion skips over both the discretion Congress gave the Fed and the economics of the payment networks it regulates.

Proportional Doesn’t Mean Identical

The 8th Circuit narrowed which costs the Fed may count. It left intact the central statutory command to assess whether fees are “reasonable and proportional” to those costs. As the court itself explained:

Make no mistake, debit-card interchange fees do not have to exactly match marginal cost. The word “proportional,” after all, does not require perfection… The Federal Reserve is free to use a proxy, perhaps even average variable cost, if it bears a “reasonable” relationship to the marginal costs incurred by cardholder banks in processing debit-card transactions… At a minimum, however, it must sort the costs into those that are transaction-specific and those that are not. Not just pick a number and hope it is close enough for government work.

Congress could have required fees to equal the incremental expense of processing each transaction. Instead, it chose “reasonable and proportional,” language that calls for judgment beyond mere arithmetic.

The merchants’ reading—that fees must exactly match those processing costs—would strip those words of independent meaning. That runs into the canon against surplusage, the rule that courts should interpret statutes so that every word has meaning.

The Fed must therefore distinguish the costs the statute permits from those it excludes. But a smaller cost baseline needn’t produce an equally small fee cap. The Fed still has room to set a higher fee, provided it can explain why that fee bears a reasonable and proportional relationship to the permitted costs.

Accounting for the Consumer Benefits of Interchange Fees

To judge whether a fee is “reasonable and proportional,” the Fed must understand what it accomplishes. As the Supreme Court recognized in Ohio v. American Express, payment networks serve two connected groups. Consumers value cards that more merchants accept, and merchants value cards that more consumers carry. Economists call these “indirect network effects.”

Networks must attract both groups, which can mean charging one side more to help fund benefits for the other. Historically, interchange fees helped card-issuing banks offer debit-card rewards, free checking, and low minimum-balance requirements. Those benefits attracted consumers, making card acceptance more valuable to merchants.

Regulation II changed that calculation. A substantial body of research, including studies by the Fed’s own economists, documents how banks responded to the cap. Banks subject to it sought to replace lost interchange revenue by raising monthly account fees and minimum-balance requirements, while reducing access to free checking. The savings had to come from somewhere.

These changes hit low-income households particularly hard, pushing some out of the banking system or toward costly alternatives such as check-cashing services. Meanwhile, there’s little empirical evidence that merchants offset those losses by passing their interchange savings along through lower retail prices.

The 8th Circuit’s opinion therefore doesn’t require the drastic fee cuts the plaintiffs seek. The economics of two-sided payment networks gives the Fed a legal framework for setting a higher cap when it revisits the rule. The agency can follow the court’s order by limiting its cost baseline to the incremental expense of authorizing, clearing, and settling individual transactions. Then, when determining what fee is “reasonable and proportional” to those costs, the Fed can exercise its delegated authority to account for the consumer benefits those fees help fund.

As ICLE’s amicus brief in Linney’s Pizza puts it:

To effectively make this judgment, the Board could consider that debit card networks are two-sided platforms and that interchange fees allow for the merchants on one side of the platform to subsidize benefits to consumers on the other…

In recognition of the benefits to consumers enabled by interchange fees, the Board could have determined that a higher interchange fee is justified because it is “reasonable and proportional” to the cost incurred by the issuer in a specific transaction. The Board could have even published standards, enforceable ex post, for what a reasonable percentage for an interchange fee could be, considering consumer benefits that they enable, rather than setting any percentage themselves. 

Under that approach, the Fed could establish standards and assess compliance after the fact, rather than prescribe a single rate itself. Either way, transaction-specific costs would remain the statutory anchor, while consumer benefits would help inform the judgment about a reasonable fee. Counting the costs is essential but understanding what consumers get for them belongs in the analysis, too.

What Comes Next for Interchange Fees

Next comes the 6th Circuit’s decision in Linney’s Pizza. If it disagrees with the 8th Circuit on which costs the Fed may count, the resulting circuit split could put the issue before the Supreme Court. If it agrees, the Fed will face another judicial directive to rebuild its rule without fixed or other costs that aren’t specific to individual transactions.

Either way, rejecting the Fed’s cost calculation doesn’t dictate another cut in interchange revenue. The Durbin Amendment’s “reasonable and proportional” standard leaves room for the agency to account for how payment networks serve merchants and consumers together.

Pushing interchange fees below market rates saves merchants money, but consumers would pay for those savings through higher bank fees, lost rewards, and fewer free checking accounts. The Fed can use its delegated discretion to account for those consumer benefits while following the 8th Circuit’s ruling. Doing so would help ensure that a merchant-driven regulatory victory doesn’t leave everyday consumers footing the bill.

The post Why Corner Post’s Merchant Victory May Not Ring Up Savings appeared first on Truth on the Market.

That sounds like trouble for a central idea in finance. The efficient markets hypothesis, developed in Eugene Fama’s foundational research, holds that prices reflect available information. Public knowledge shouldn’t offer investors a reliable route to returns beyond what their risks justify. Yet everyone has a calendar, so why haven’t investors anticipating the winter gains bought earlier and erased the advantage?

The effect’s existence has substantial academic backing. Researchers have documented the seasonal gap across dozens of countries, long stretches of history, and years beyond the original studies. Even cryptocurrency publications give it an annual airing. Last October, DailyCoin explored its application to digital assets, while CCN reported comparisons favoring year-round ownership in most stock indexes it examined, as well as Bitcoin.

But a well-documented pattern isn’t necessarily a profitable trading strategy—or a refutation of market efficiency. The harder questions are what causes the winter advantage, whether it compensates investors for risk, and what happens once people try to trade on it.

The Pattern That Won’t Stay Dead

Sven Bouman and Ben Jacobsen’s 2002 study on the Halloween effect found higher November–April returns in 36 of 37 countries they examined. In 20 of these examples, the gap met their 10% threshold for statistical significance. Under their model, a gap at least that large would arise less than 10% of the time if no seasonal effect existed.

Search enough stock-market history, though, and you’ll find patterns that arose by chance. A stronger test asks whether they repeat in later years. Sandro Andrade, Vidhi Chhaochharia, and Michael Fuerst’s 2013 follow-up examined November 1998 through April 2012, a period the original study didn’t cover. Winter returns exceeded summer returns in all 37 markets, with statistically significant gaps in 13 of them. The average difference approached 10 percentage points.

Cherry Zhang and Jacobsen’s 2021 global study added historical breadth. Across 62,962 observations, it found an average winter advantage of about 4 percentage points. Together, these findings make the effect harder to dismiss as coincidence.

So, the big question is not whether the Halloween effect exists, but how reliably it appears and how useful it is. For example, Júlio Lobão and Ana Costa’s 2023 study of Portuguese markets found an effect early in the 20th century, followed by roughly 60 years without it. It returned after 2000, largely thanks to strong January returns. That’s a long wait for a trader who hopes to retire someday.

Along those lines, other studies question the effect’s practical value as a profitable trading strategy. Hubert Dichtl and Wolfgang Drobetz’s 2015 study found that the winter advantage shrank substantially or disappeared in more recent periods. Thomas Degenhardt and Benjamin Auer’s 2018 review and analysis found continuing seasonality but strategies built around selling stocks in May and buying them back in the fall had limited success.

And while finding the pattern in many countries strengthens the evidence, it should also be noted that those markets often move together. A global crash during the summer could lower May–October returns in dozens of countries at once. That would reflect one shared event, rather than dozens of separate confirmations of the Halloween effect. Bouman and Jacobsen addressed this problem by adjusting for global stock-market movements.

The studies thus ask distinct question: Does the pattern recur? Does it survive publication of the research? Can investors profit after trading costs and the returns available elsewhere during summer?

The publication question echoes Goodhart’s law, developed by economist Charles Goodhart in a 1975 discussion of British monetary policy. A statistical relationship that helps policymakers understand the economy can become unreliable once they try to control the economy through it. Banks and other market participants adjust their behavior, changing the relationship policymakers hoped to exploit.

Anthropologist Marilyn Strathern supplied the familiar formulation in her 1997 discussion of university audits: “When a measure becomes a target, it ceases to be a good measure.” 

A seasonal trading rule raises a related problem. Once investors act on a historical pattern, their buying and selling can change the prices that produced it. That doesn’t necessarily doom every published strategy, but it makes historical success an uncertain guide to future profits. After all, the market gets to read the paper, too.

Still Hunting for the Ghost in the Market

Bouman and Jacobsen suggested that summer vacations might help explain the lower May–October returns. Investors might sell shares before going away to avoid worrying about losses while they’re gone. That’s plausible, although six months is a long summer vacation, even for Wall Street.

Of course, the people running public companies also take vacations, and their departures may affect the information investors receive. In his 2013 working paper “Tailspotting,” David Yermack matched corporate jet flights with property records identifying CEOs’ vacation homes. Companies tended to release favorable news before their CEOs departed and delayed subsequent news until around their return. During those absences, companies made fewer announcements, mandatory disclosures were more likely to arrive late, and stock prices fluctuated less.

Summer vacations could therefore change both trading behavior and the flow of corporate news. But smaller price swings don’t necessarily mean lower returns, and executives’ trips don’t establish a market-wide explanation for six months of weaker performance.

Mark Kamstra, Lisa Kramer, and Maurice Levi’s “Winter Blues” proposes a psychological explanation. Seasonal daylight changes may affect investors’ moods and willingness to take risks. Jacobsen and Wessel Marquering’s “Is It the Weather?” points to a problem with that theory. Daylight, weather, and vacations all follow seasonal schedules. Correlations with returns don’t establish which, if any, drives the pattern.

The winter advantage also varies by industry. Jacobsen and Nuttawat Visaltanachoti’s 2009 study found a strong November–April advantage in production sectors, such as manufacturing, but little in sectors tied to consumer spending. That complicates the “mood” explanation. If shorter days change investors’ willingness to take risks, the theory also needs to explain why that change affects some industries more than others. Investors’ gloom apparently has sectoral preferences.

A newer paper suggests that corporate reporting schedules might help explain the pattern. Jan Schroeder’s 2025 study examined filings with the Securities and Exchange Commission (SEC), finding that from 2004 through 2023, companies submitted 17% more filings during November–April than during May–October. Annual reports and shareholder meetings also followed seasonal patterns, suggesting that investors receive more company news at some times of year than others.

Schroeder doesn’t claim that more disclosures cause higher returns. Company news can move prices in either direction. Still, a busy reporting season might expose investors to more unwelcome surprises, leading them to demand lower purchase prices and higher expected returns as compensation. Investors might also struggle to absorb clustered disclosures, causing prices to adjust gradually. Researchers would need to test both explanations. Counting the number of public filings doesn’t establish either one.

Research on Ramadan offers another example of unusual stock returns tied to a holiday. Jedrzej Bialkowski and Mona Yaghoubi’s 2021 study found unusually high returns during the Muslim month of fasting across 11 Muslim countries from 2009 through 2018. But there’s no obvious reason that buying and selling shares would become harder during Ramadan, weakening the explanation that higher returns compensated investors for holding investments they might struggle to sell.

Because Ramadan follows a lunar calendar, it moves through the seasons; it might fall in summer one year and winter in another. If higher returns follow the observance from winter into summer, religious or social practices become more plausible explanations than shorter days. That doesn’t establish a shared cause with the Halloween effect, but it helps separate competing explanations.

The Halloween effect needs similarly discriminating tests. What changes investors’ behavior from November through April, and why would that produce higher returns? Researchers have proposed several plausible answers, but none has emerged as a convincing explanation for the pattern as a whole.

Who’s Afraid of an Efficient Market?

The Halloween effect seems to challenge the idea that stock prices already reflect publicly available information. If investors know stocks tend to earn more from November through April, why haven’t they bought ahead of that rise, pushing prices up earlier and eliminating the seasonal advantage?

But higher expected returns don’t necessarily mean stocks are mispriced. They may compensate investors for taking greater risks or for investing when they’re less willing to risk a loss. Fama’s 1991 review explains why that makes market efficiency hard to test. Researchers must first estimate what return would adequately compensate investors for the risks they face. If that estimate is too low, returns may look unusually generous even when they simply reflect the compensation investors require.

Andrade and his co-authors found similar seasonality in strategies involving smaller companies, stocks priced cheaply relative to their accounting value, and lending to borrowers more likely to default. Those investments involve different risks, but their returns shared a winter advantage. That’s consistent with a seasonal shift in the compensation investors demand, although it doesn’t establish the explanation.

It’s also important to remain cognizant of the arithmetic involved. Suppose stocks gain 12% during winter and 2% during summer. Both seasonal traders and year-round investors earn the winter’s 12% gains. Selling in May exchanges the summer’s 2% for whatever the money would otherwise earn elsewhere. To deliver a higher overall return, the alternative must beat stocks over the summer by enough to cover trading costs.

A seasonal strategy could still be appealing even if it earns less, but substantially reduces potential losses. K. Stephen Haggard and H. Douglas Witte’s 2010 paper found that winter stocks and summer government debt offered better returns relative to risk than year-round stock ownership, even after trading costs. Their findings also survived tests accounting for unusually large market moves and January’s historically strong returns.

Dichtl and Drobetz, and Degenhardt and Auer, reached less encouraging conclusions, with their findings recalling Goodhart’s warning. If investors expect stocks to rise after Halloween, they may buy before then, pushing prices up sooner and leaving less potential profit for those who wait until November. A strategy’s declining profitability could therefore reflect investors learning from published research.

But Goodhart’s law can’t settle the question by itself. As Cedric Chin argues in Commoncog, the maxim describes a problem without necessarily explaining how to address it. Researchers still need evidence that actual trading in actual markets changed the pattern. Different samples, costs, or risks could explain weaker results. And publicity needn’t eliminate returns that compensate investors for risk.

Even when investors correctly spot a mispriced investment, they can lose money while waiting for the market to catch up. Andrei Shleifer and Robert Vishny’s “The Limits of Arbitrage” explains why that creates a particular problem for professionals managing clients’ money. Clients may see the early losses, lose confidence, and withdraw their funds before the investment pays off.

Imagine a fund manager who sells stocks in May, expecting weaker summer returns. Instead, stocks soar through August, leaving the fund behind competitors that stayed invested. Clients may withdraw their money rather than wait to see whether the strategy works over several years. 

Those pressures don’t explain why the Halloween effect arises, but they could help explain why it persists. Managers may hesitate to bet on a seasonal pattern if a few bad months could cost them clients or their jobs. The scariest date on their calendar may be the next quarterly client review.

Efficient Markets Aren’t Dead Yet

The Halloween effect is a persistent empirical puzzle, but its existence alone doesn’t undermine the efficient markets hypothesis. Efficient prices can still produce predictable differences in returns if those differences compensate investors for risk. Establishing a seasonal pattern therefore leaves the central question open. Does the winter advantage reflect the risks investors bear, changes in their willingness to bear them, or a pricing mistake?

The research offers several possible explanations without settling the matter. Yermack’s vacation findings suggest one connection between executives’ schedules and the flow of information. Goodhart’s law gives us a reason to ask whether publicizing a trading rule changes its performance. Neither establishes what causes the Halloween effect. If investors trade away a pricing mistake, that would be consistent with markets learning. If a seasonal difference survives, researchers still need to determine whether it compensates for risk or reflects an opportunity too costly or uncertain to exploit.

The more recent findings of a Halloween effect in cryptocurrency offers a potentially useful comparison. Bitcoin doesn’t file earnings reports or have a CEO who leaves for a summer house. A comparable seasonal pattern there would suggest that corporate disclosures and executives’ vacations can’t provide the whole explanation. Shared influences, such as investors’ mood around the holidays or changing risk appetite over the course of the year, would merit closer attention. Similar patterns needn’t share a cause, but comparing markets could help narrow the possibilities.

The task is to explain the returns before treating them as evidence against market efficiency. A recurring October story isn’t an annual obituary for efficient markets.

The post The Halloween Effect and the Premature Burial of Efficient Markets appeared first on Truth on the Market.

Perhaps it’s a focus on rational actors—people making choices to advance their interests—and wealth maximization? But that would leave out the behavioral movement, which examines departures from rational decision-making. A field for empirical work—research based on observation and data? That would also leave out many of its founders. OK, then. Perhaps the L&E tent is now so big that it’s undefinable. No particular expertise is required, nor any shared precepts.

In “Contemporary Law and Economics,” Adam Chilton, Joshua Macey, and Mila Versteeg disagree. L&E is still a cohesive field. Writing for the University of Chicago Law Review’s 2025 symposium, the authors acknowledge that L&E’s contours shift. Its scholars don’t all agree on its definition—an unsurprising result for a diverse field of study. But distinct qualities still set L&E apart from other scholarship, including work influenced by economics.

As newer groups such as the Law and Political Economy (LPE) movement capture scholarly attention, our authors set out to tell us what L&E is, given that approaches as diverse as empirical legal studies, welfarism (evaluating rules by their effects on human well-being), and rational choice all claim to be within its ambit.

What Law & Economics Is

L&E has roots in utilitarian philosophers like John Stuart Mill, who pondered when harms were sufficient to justify government coercion against a baseline of individual freedom. Its modern incarnation springs from Coase’s work. The Coase theorem explains how bargaining can resolve conflicts over rights, such as a property-line dispute, when barriers to bargaining—“transaction costs”—are sufficiently low. It led to further examination of how assigning rights and lowering transaction costs could make legal systems more efficient.

Guido Calabresi and A. Douglas Melamed’s work examined government action in the face of harms, often framed as “externalities”—effects on people outside a transaction. Their work followed a similar vein to Mill’s, enhanced with economic insights.

Many early works also assumed that actors operated rationally within legal systems. By the 1980s and ’90s, scholars questioned the rational-actor assumption but maintained the earlier works’ methodological rigor. Meanwhile, other L&E scholars mined areas as diverse as the private ordering of diamond dealers and the efficient customs of whale hunters.

Drawing from these sources, Chilton, Macey, and Versteeg ground L&E in a question-driven approach. L&E scholars can come from any discipline, and their conclusions can support any movement, including critical ones. They develop hypotheses, test them through research that builds on others’ work, and subject their findings to peer review. The research needn’t be quantitative, mind you. Historical research can make great L&E work.

Whatever the research type, scholars apply this question-driven approach to legal rules and institutions to teach us what an effective legal system might be. Economic concepts generally inform their assessment. If a tax increase on millionaires to support social services, upon study, leads to less revenue for those services, an L&E scholar questions whether the law is wise—even if the scholar supports its political goals.

L&E’s best feature also leaves it hard to pin down. It’s at its best in the hands of generalists. Many shining L&E stars apply an economic lens across areas of law. Posner covers topics as diverse as literature (as befits an English major), intellectual property, antitrust, and adoption.

Jonathan Masur recently applied economic insights to wine, originalism, and patent disclosures. One of the article’s authors, Chilton, studies European Union laws one day and, presaging his deanhood, law school administration the next.

L&E is capacious in no small part because so many fun legal topics have yet to be studied through an economic lens. The authors argue that a particular question-driven approach makes these diverse studies L&E works.

What Law & Economics Isn’t

L&E isn’t so broad that it encompasses any topic with economic implications. The authors boldly state that some early works labeled L&E don’t fit their definition. Purely normative and abstract work—arguments about what ought to be, without grounding in real-world observations—isn’t L&E, even if couched in economic terms. Perhaps these pre-L&E works are best thought of as Led Zeppelin is to metal—not quite the thing, but venerated predecessors that led (no pun intended) to the real thing.

The LPE movement, grounded in shared critical commitments such as the view that “economic and political power are inextricably intertwined with racialized and gendered inequity and subordination,” also falls outside their definition of L&E. Drawing this line is difficult if one focuses on terminology alone. LPE scholarship, by definition, focuses on “political economy,” which studies how the politically powerful shape who succeeds in a legal system. At that level of abstraction, L&E is also deeply interested in who has power and what they do with it. Both L&E and LPE cite some of the same early sources of political economy scholarship.

The line becomes clearer when one focuses on each field’s guiding principles. L&E uses the question-driven approach described above. A shared worldview defines LPE. The authors note that LPE draws from critical race theory, feminism, and legal realism to inform its views on institutions. If a work starts with the goal of outing inequalities in a given system, it’s eligible to enter the ranks of LPE scholarship—even if it doesn’t use empirical or other evidence-based methods to support its claims.

The result is two worthwhile but distinct endeavors whose different goals can lead to similar conclusions. L&E works can, and do, support the LPE worldview. The authors note that we should expect this overlap, not scorn it.

L&E’s lack of predefined political goals makes that overlap possible. One L&E work can use a question-driven approach to support state tax reforms aimed at addressing inequality. Another can reach the opposite conclusion using different models or datasets. L&E is a big tent, but it isn’t an undefinable and brooding omnipresence over all economic scholarship about the law.

What Law & Economics Will Be

As someone who travels in L&E circles with a mere government degree and experience managing legal and engineering teams, I take some comfort in a definition of L&E work that makes space for real-world observations and varying methods. Indeed, Chilton, Macey, and Versteeg aren’t economists, either.

To keep growing the field, the authors ask L&E scholars across disciplines to engage more with one another. Economists like Eric Helland, with fascinating studies of how physicians shield their assets, should present alongside new law professors like Jonathan Petkun, whose groundbreaking work examines how tweaks and nudges in court administration can make courts more efficient.

Conferences like the American Law and Economics Association’s (ALEA) annual meeting and programming like the International Center for Law & Economics’ (ICLE) (shameless plug) do attempt to gather diverse L&E thinkers. Indeed, I enjoyed both Helland’s and Petkun’s presentations at last year’s ALEA conference. But we need more opportunities for formal and informal connections.

And if L&E doesn’t lead the way in artificial intelligence (AI) scholarship—an area that needs more question-asking and analytical rigor than just about any other in our lifetime—then the field is in poor health. The authors are confident L&E can lead in AI research with well-founded observations. That would be a welcome advance over the current from-the-gut takes that alternate between fear and unbridled support.

All L&E scholars, economists and English majors alike, will have to fight off the temptation in a divided political environment to define L&E against other movements. LPE is left-coded. Defining L&E as “not LPE” would imply that it’s the study of law from the right. That would leave many of L&E’s best scholars out in the cold.

If Chilton, Macey, and Versteeg’s definition is correct, a shared approach defines L&E. Requiring particular political outcomes would undermine the qualities that make it great. Working backward from outcomes to methods is bad economics—and a bad definition of law & economics.

Further Reading

• Jedediah Britton-Purdy, David Singh Grewal, Amy Kapczynski, & K. Sabeel Rahman, “Building a Law-and-Political-Economy Framework: Beyond the Twentieth-Century Synthesis,” The Yale Law Journal, Vol. 129, No. 6 (2020)
• Christopher Buccafusco, Jonathan S. Masur, & Ryan Whalen, “Measuring the Value of Trademark Distinctiveness: Evidence From the Market for Bordeaux Wine,” Journal of Empirical Legal Studies, Vol. 22, No. 4 (2025)
• James M. Buchanan & Gordon Tullock, “The Calculus of Consent: Logical Foundations of Constitutional Democracy,” University of Michigan Press (1962)
• Guido Calabresi & A. Douglas Melamed, “Property Rules, Liability Rules, and Inalienability: One View of the Cathedral,” Harvard Law Review, Vol. 85, No. 6 (1972)
• Ronald Coase, “The Problem of Social Cost,” Journal of Law and Economics, Vol. 3 (1960)
• John Stuart Mill, “On Liberty,” John W. Parker & Son (1859)
• Richard A. Posner, “Economic Analysis of Law,” Little, Brown & Co. (1973)
• George J. Stigler, “The Theory of Economic Regulation,” The Bell Journal of Economics and Management Science, Vol. 2, No. 1 (1971)
• Cass R. Sunstein, “Behavioral Law & Economics,” Cambridge University Press (2012)

The post ‘Contemporary Law and Economics,’ by Adam Chilton, Joshua C. Macey & Mila Versteeg appeared first on Truth on the Market.

In a recent Truth on the Market post, Ian Adams revisited his decade-old warnings against coverage mandates and taxpayer backstops for cyber insurance. The cyber market’s subsequent growth supports his case for patience. Insurers have expanded coverage while refining prices, exclusions, and security requirements. As Adams argued, AI insurance deserves the same opportunity as courts clarify legal duties, reasonable precautions, and the link between misconduct and injury.

That adaptation has, in fact, already begun. For example, directors and officers (D&O) insurance, which covers certain claims against corporate leaders and companies, already addresses some AI-related risks. It also offers a window into how existing coverage can adapt, even though it covers only part of AI’s potential liabilities. 

Together, these markets illustrate three classic bits of wisdom about the value of insurance: It can serve to preserve honest experimentation, reward effective precautions, and account for shared failures. Policymakers should give the market room to apply that wisdom before prescribing coverage or handing taxpayers the bill.

The Market Learned on the Job

Cyber insurance’s growth shows that a market can expand as insurers continue to learn. The National Association of Insurance Commissioners (NAIC) reports that domestic cyber premiums grew from $2.75 billion in 2020 to $7.46 billion in 2025. Domestic insurers reported 5.84 million policies, up 33% from 2024.

Global estimates vary considerably. Jamie MacColl, Jason R.C. Nurse, and James Sullivan’s 2021 Royal United Services Institute report cited roughly $5 billion in premiums. The NAIC cites more than $16 billion for 2025; Shivani Zoting’s Precedence Research analysis estimates $26.32 billion and projects $288.42 billion by 2035.

Cyber insurance expanded without the broad purchase mandates or dedicated taxpayer backstop some observers advocated. That history supports giving markets room to adapt, while leaving open how well they cover particular risks.

But cyber insurance’s growth also shows why the terms of coverage matter as much as availability. MacColl, Nurse, and Sullivan found weak risk data, shared-loss concerns, and limited evidence that insurance consistently improved cybersecurity outcomes. Aggeliki Tsohou, Vasiliki Diamantopoulou, Stefanos Gritzalis, and Costas Lambrinoudakis’ 2023 review documented inconsistent underwriting questions and ambiguous coverage. Huntress later reported that 58% of surveyed policyholders had seen cyber coverage narrow, though its survey cannot establish marketwide prevalence.

These shortcomings fit a market still learning from court decisions and claims experience. Clearer liability rules, better loss data, and technical testing can help insurers refine coverage. Alarm over AI is no reason to rush that work, just as it wasn’t for cyber risks. Sound insurance products take time to develop, and poorly designed or mispriced coverage serves neither insurers nor the policyholders who depend on them to pay claims.

AI raises similar questions for existing policies. As Risk & Insurance explains, AI-related claims can arise under general liability, professional liability, cyber, employment, and D&O policies—even when those policies say nothing about AI. That silence can leave insurers and policyholders disagreeing over which losses they agreed to cover.

Some insurers are making those boundaries explicit. Insurance Business reports that the Insurance Services Office, which develops standard policy language for insurers, introduced optional exclusions in January 2026 for losses involving generative AI. Insurers can add these exclusions to commercial general liability policies, which typically cover claims involving bodily injury, property damage, and certain other harms. Each insurer decides whether to adopt them, while specialty AI coverage is also emerging. Businesses therefore need to check both what their existing policies exclude and what protection they can buy elsewhere.

AI Meets the Fine Print

As Priya Cherian Huskins explains for Gallagher, D&O policies protect executives, reimburse companies that cover executives’ eligible losses, and cover certain claims against companies themselves. Public-company coverage generally limits that last category to securities claims. Defense costs, settlements, and judgments can all qualify for coverage.

Kevin LaCroix of The D&O Diary has counted 30 AI-related securities class actions filed thus far 2026, compared with 16 in 2025. Those totals track filings, not active cases or insured losses, and definitions of “AI-related” vary. Economic consulting firm Cornerstone Research and Stanford’s Securities Class Action Clearinghouse counted 15 in the first half of the year, while National Economic Research Associates economists Ivelina Velikova and Svetlana Starykh counted 18. 

The technology may be new, but many of the allegations are quite familiar for D&O insurers. Writing in Bloomberg Law, Ross Shikowitz, Adam McCall, and William Massa identify premature commercialization promises, mischaracterized software, and abrupt corporate reinventions among the common causes of action. Disputes involving Apple’s Siri, Upstart, Tempus AI, and Endava concern performance claims, revenue expectations, affiliated transactions, and accounting—all well-trod ground for such claims. 

In D’Agostino v. Innodata, an investor accused the data-engineering company and its senior executives of touting proprietary AI while relying heavily on employees doing labor-intensive data entry and labeling. Judge Jamel Semper of the U.S. District Court for the District of New Jersey dismissed the complaint without prejudice in April 2026, finding, among other deficiencies, that the plaintiff had failed to allege facts strongly supporting an inference of fraudulent intent. 

The Opendoor litigation produced a $39 million settlement over alleged misrepresentations about its home-pricing algorithm. The company reported that the agreement fell within insurance limits and recorded expected insurance recoveries. That illustrates insurance’s existing role even when fraud is not formally established.

Copyright and privacy disputes can also generate claims against corporate leaders. In the Adobe derivative complaint, a pension-plan shareholder sued on Adobe’s behalf, alleging that its leaders allowed the company to train its SlimLM language models on datasets containing unauthorized copies of copyrighted books and other works. The suit also challenges assurances about lawful sourcing and the commercial safety of Adobe’s AI products.

The shareholder alleges that officers and directors breached their duties through those practices and misleading disclosures, exposing Adobe to litigation and other harm. A derivative suit seeks recovery for the company. Such allegations can generate oversight claims without automatically bringing the underlying copyright liability within D&O coverage.

Policy language, which can vary, also remains decisive. Chubb requires final, nonappealable findings for specified fraud exclusions, so allegations alone need not end defense funding. Other exclusions, notice requirements, deductibles, and limits remain; insurability of fines varies by jurisdiction.

Disappointment Is Not a Cause of Action

The first bit of insurance wisdom for AI markets, well known to D&O specialists, is to distinguish misleading disclosures from business disappointments. In Dura Pharmaceuticals v. Broudo, investors alleged misleading statements about prospects for Food and Drug Administration (FDA) approval of an asthma-treatment device. The Supreme Court unanimously held in 2005 that an inflated purchase price alone did not establish “loss causation”—the connection between alleged fraud and investors’ economic loss.

The same distinction matters for AI investments. Investors may lose money because managers conceal problems, but they may also lose money because competitors improve, financing costs rise, or an experiment simply fails. A securities-fraud claim—the kind D&O insurance generally covers—must connect the alleged misconduct to the investor’s loss.

To point to a recent example: CoreWeave’s weather-delayed data-center construction offers a mundane reminder that even AI needs concrete. Managers must disclose material information as securities law requires, including about construction delays. But a setback and a falling stock price do not, by themselves, establish fraud.

Knowing that a project could fail should not make company leaders liable each and every time that risk materializes. Investors can lose money even when managers disclose risks honestly. Treating those losses as grounds for liability would turn disclosure law into investment insurance, discouraging companies from pursuing promising but uncertain projects.

Cornerstone reports that the 15 AI-related securities class actions filed in the first half of 2026 involved $385 billion in declines in companies’ stock-market value around alleged disclosures. Those cases represented 13% of filings excluding merger-and-acquisition class actions, but 73% of the stock-market losses captured by its disclosure-loss measure. Those declines are not necessarily damages attributable to fraud. Treating them as recoverable losses without establishing that connection could discourage companies from pursuing promising experiments whose risks investors knowingly accept.

Clearer duties and causation rules help insurers assess which failures could produce covered claims, while coverage can give managers confidence to pursue uncertain projects. In a recent working paper, Wenshuai Xu and Jianfei Peng report that firms carrying D&O insurance invest more in AI. Their analysis suggests that greater managerial willingness to bear risk, while fewer financing constraints help explain the association. The association is also stronger among firms with less-overconfident executives, suggesting that coverage can support experimentation by relatively cautious managers.

That finding supports giving the market room to develop, and is consistent with the combination of investment and restraint that advocates of faster insurance expansion say they want. Mandates or underpriced guarantees could undermine that balance by weakening the pricing and coverage decisions through which insurers distinguish prudent experimentation from poorly managed risk.

To be sure, more investment does not necessarily mean better investment, and these findings alone cannot establish what the presence of insurance coverage causes. They nevertheless highlight a benefit worth preserving. Insurance can give managers room to pursue uncertain projects, while liability rules hold them accountable for actionable misconduct. The aim is to make worthwhile risks bearable while keeping dishonesty expensive.

Good Governance Might Earn a Discount

The second bit of insurance wisdom concerns rewarding effective precautions. Insurance broker Aon’s Financial Services Group already advises applicants to expect questions about how they use AI, how heavily their business depends on it, and how it supports their financial projections. Underwriters also want to know who oversees AI risks and checks related disclosures before publication.

A survey by The D&O Diary and Allianz Commercial suggests that insurers take those questions seriously. Among insurer respondents, 91.2% agreed that inadequate transparency about AI deployment should count against applicants in underwriting assessments. Only 59.1% of brokers agreed—a sizable gap between those providing coverage and those arranging it. The responses reveal how participants think transparency should affect coverage decisions, but don’t establish whether insurers changed premiums or whether greater transparency reduced claims.

Turning those expectations into useful assessments requires evidence. Underwriters need to check whether customer contracts support AI-related revenue claims, whether performance tests back up claims about what the technology can do, and whether firms have the legal right to use their training data. They should also ask how employees report failures, who receives those reports, and who can require corrective action. A board-approved policy offers little protection if employees cannot get management to address a problem.

Technical testing has limits, too. RAND researchers Sasha Romanosky and Celine Robinson explain that performance tests work best for defined tasks with clear answers. Test results alone cannot establish how often errors will produce insured losses, especially as models and their uses change. Insurers need to understand both how a system performs and what happens when it fails. 

Research on conventional D&O insurance shows both the promise and the limits of evaluating how companies manage risk. In a 2007 University of Chicago Law Review study, Tom Baker and Sean J. Griffith found that underwriters considered corporate governance when assessing applicants and pricing coverage. Their judgments emphasized management’s character and corporate culture alongside formal board structures. Charging better-managed firms less could reward sound governance. The study did not, alas, establish that those price differences actually prompted firms to improve.

Baker and Griffith’s companion Georgetown Law Journal study examined what happened after insurers issued coverage. They found that D&O insurers neither monitored governance during the policy term nor actively managed litigation defense costs after claims arose. The authors attributed the lack of monitoring partly to conflicting incentives—what the law & economics literature would call a “principal-agent problem.” Managers buy coverage that protects them personally, while the corporation—and ultimately its shareholders—pays the premiums. Managers may therefore prefer protection with few strings attached.

AI-related underwriting faces the same incentive problem. As insurers learn more about AI-related exposure, they could offer better terms to firms with tested safeguards and reliable disclosures, while charging more, requiring larger deductibles, or narrowing coverage where precautions fall short. But rewarding firms that already manage risks well won’t necessarily persuade other managers to improve. 

Insurance markets have a long record of adjusting prices and terms as claims experience improves their understanding of risk. The same process can shape AI coverage. Competition and better information can lower premiums for well-understood risks, while firms with weaker safeguards face higher prices or tighter terms.

Claims experience should further clarify frequency (how often claims arise) and severity (how much they cost). And even unsuccessful suits can generate substantial defense bills. Larger settlements will reach excess insurers, which cover losses above the primary policy’s limit.

That distinction also helps clarify which AI losses need attention now. In a company’s layers of coverage, one insurer might cover the first $10 million above the amount the company pays itself, while another covers the next $10 million. The first faces smaller claims and routine defense bills. The second faces unusually costly litigation. Catastrophic scenarios attract the headlines, but ordinary claims and legal expenses are the more immediate test as AI coverage develops. Insurers need to price both, and those covering different portions of the losses can reasonably assess the same company differently.

A Thousand Policies Can Share One Problem

The third bit of insurance wisdom comes from managing losses that arrive together. Insurers have long understood that a thousand policies on homes along the same coastline can amount to one large bet on the weather. Concentration means too much exposure to a particular source of loss, while correlation means losses tend to occur together. Either can undermine the protection that a large, varied pool of customers normally provides.

Reinsurance—insurance that insurers buy—is central to managing that exposure. In a reinsurance contract, an insurer transfers part of its potential losses to another company, which can spread them across a broader portfolio of regions and risks. As the Federal Reserve Bank of Chicago explains, reinsurance helps insurers manage concentrations in particular locations or lines of business. It also gives them capacity to offer more coverage than they could prudently support alone.

AI presents a new version of this familiar problem. Companies in different industries and countries may depend on the same model or cloud provider. An insurer’s customer list may appear to be diverse while concealing a common vulnerability. Reinsurance can spread the financial burden, but reinsurers must identify those dependencies across all the insurers they cover. Transferring a risk does not make the underlying connection disappear.

In AI markets, those connections can also extend beyond the underlying technology. Law firm Quinn Emanuel’s analysis of data-center financing describes developers relying on customer commitments to repay construction debt and, in some arrangements, using specialized computing chips as collateral. Construction delays, power shortages, customer defaults, or falling chip values can threaten repayment. Cross-default clauses, which allow a default under one agreement to trigger defaults under others, can spread the distress.

One disruption could therefore produce claims against developers, customers, lenders, and corporate leaders under different policies. Ordinary defaults and business losses would, of course, each still require a separate basis for coverage. Insurers and reinsurers nevertheless need to assess their combined exposure, including claims that look manageable individually but become costly together.

Shared exposure can also emerge in court. RAND researchers Romanosky and Robinson identify legal changes as a potential source of correlated claims when many firms follow similar practices. Their analysis also makes an important distinction about exclusions. Removing a risk from an insurance contract leaves the business to bear it. Better information can improve risk allocation, but it cannot guarantee unlimited insurance capacity. 

These sorts of potential limits on private capacity offer the strongest arguments for considering a public backstop. Gregory C. Allen of the Center for Strategic and International Studies argues that restrictions imposed by the insurance market might actually impede AI adoption. He recommends preparing a staged catastrophic backstop alongside better incident data and verification. 

To be sure, that concern deserves serious consideration. But any proposal should explain which losses private insurance and reinsurance cannot absorb, why that gap warrants public intervention, and how the arrangement would preserve incentives to reduce losses. A government guarantee priced below the risk it assumes could weaken those incentives before taxpayers ever pay a claim. Coverage mandates could likewise lock in obligations before insurers understand their combined cost.

Public policy can also play a useful role by helping markets to learn without having taxpayers assume those obligations. RAND recommends standardized notices explaining how policies address AI, common categories for reporting incidents and claims, and shared scenarios for testing potential losses. Such information could help insurers and reinsurers identify common vulnerabilities and negotiate coverage with fewer surprises. Standardizing how firms describe coverage need not dictate what they sell. 

The Fine Print Still Applies

AI presents some novel and unfamiliar risks, but familiar insurance principles still apply. As it has in other markets and with respect to other kinds of risk, AI insurance can serve to preserve honest experimentation, reward demonstrated precautions, and watch for shared failures. Cyber insurance shows how markets can learn, while D&O coverage shows that adaptation to AI has already begun. Courts can clarify responsibility, insurers can refine coverage, reinsurers can spread risks, and competition can reward firms that manage them well.

Policymakers should strengthen the conditions apt to produce those lessons before prescribing coverage or shifting losses to taxpayers. New technology alone supplies no reason to discard accumulated wisdom about risk, regulation, and policy. Indeed, the reality is quite the opposite: Even AI can learn something from the fine print.

The post Old Insurance Wisdom for New AI Risks appeared first on Truth on the Market.

Under the Communications Act of 1934, as amended, the FCC has initiated early license-renewal proceedings against eight television stations that American Broadcasting Cos. Inc. (ABC), a subsidiary of The Walt Disney Co., owns and operates. ABC sued to challenge those proceedings on First Amendment grounds in American Broadcasting Companies, Inc. v. Federal Communications Commission, now before the U.S. District Court for the District of Columbia.

The case also raises a foundational question of administrative law. Can an agency turn a broadly worded statute into a roving mandate to enforce rules beyond its jurisdiction?

The stated rationale for cutting short ABC’s standard eight-year license terms is an ongoing inquiry by the FCC’s Enforcement Bureau into Disney’s corporate diversity, equity, and inclusion (DEI) policies. The FCC contends that compliance with equal employment standards is an essential part of a licensee’s obligation to operate in the “public convenience, interest, or necessity.”

But a close reading of the agency’s motion to dismiss reveals its reliance on a public-interest standard for broadcast licensees that stretches beyond the statute’s text and seems unjustifiable in today’s media marketplace.

The ‘Public Interest’ Depends on Who’s in Charge

The Communications Act conditions both initial grants of spectrum—the airwaves used to transmit signals—and periodic license renewals on an FCC finding that they will serve the “public interest, convenience, and necessity.” Rooted in the Radio Act of 1927, this standard originally addressed technical spectrum scarcity and harmful interference. Over the decades, the FCC has repeatedly used this elastic phrase to assert authority over matters peripheral to spectrum allocation.

In its brief, the FCC relies primarily on two statutory and judicial pillars to justify its jurisdiction over Disney’s corporate hiring practices:

1. Title I’s anti-discrimination declaration. 47 U.S.C. § 151 states a congressional purpose to make radio and wire communication available “without discrimination on the basis of race, color, religion, national origin, or sex.”
2. D.C. Circuit precedent. Decisions such as Beaumont Branch of the NAACP v. FCC and Bilingual Bicultural Coalition on Mass Media, Inc. v. FCC have held that a licensee’s intentional employment discrimination reflects adversely on its character qualifications—its fitness to hold a broadcast license—under the public-interest standard.

A close reading of the statutory text casts doubt on this interpretation. Section 151 states the agency’s organizational purpose, directing it to ensure universal access to communications services without discriminatory barriers. It does not expressly grant regulatory authority over internal corporate employment practices. After Loper Bright Enterprises v. Raimondo, the FCC can no longer expect Chevron deference—the judicial practice of deferring to reasonable agency interpretations of ambiguous statutes—for such expansive readings of its authority.

Congress also established a comprehensive federal system for identifying and remedying workplace discrimination through Title VII of the Civil Rights Act of 1964 and the Equal Employment Opportunity Commission (EEOC). When Congress intends an independent regulatory agency to exercise overlapping employment oversight, it enacts targeted statutory language rather than relying on a generalized “public interest” standard.

The FCC’s current enforcement approach extends far beyond Beaumont Branch and arguably beyond what its holding contemplated. That line of cases tied character disqualification for license-renewal purposes to adjudicated, systemic, intentional exclusion that violated established civil rights statutes. Here, the FCC alleges “intentional discrimination” encompassing common corporate DEI benchmarks, fellowship programs, affinity groups, and creative diversity initiatives.

Ironically, cases like Bilingual Bicultural Coalition involved the FCC requiring the types of “affirmative action” programs it now labels “intentional discrimination.”

The current license-renewal proceedings illustrate the problem. By asserting that corporate diversity policies call a licensee’s character qualifications into question, the FCC turns the public-interest standard into a vehicle for shifting executive policy directives. A standard that can both mandate and forbid affirmative action programs based on who holds power is no standard at all.

This creates at least two problems. First, it disconnects what should be neutral, technical administration of telecommunications from the technical realities of telecommunications. Second, it subjects licensees to politically driven obligations that change from administration to administration, creating uncertainty for investment, deployment, and normal business operations.

Just a Routine Retaliation 

In March 2025, the FCC opened an investigation into ABC’s employment practices. Its statutory hook was the Communications Act’s requirement for an affirmative public-interest finding when renewing a license. In the FCC’s telling, it is conducting an ordinary pre-renewal investigation—a routine exercise of statutory responsibility.

But that’s the problem. A broad, ambiguous “public interest” standard invites expansive uses of agency power and claims that those uses are political. ABC argues that the renewal investigation provides a pretext for retaliation.

Under D.C. Circuit precedent, a retaliation claim has three elements. The plaintiff must show that it engaged in activity the First Amendment protects, that the government took retaliatory action “sufficient to deter a person of ordinary firmness” from speaking again, and that a causal link connects the protected activity to the adverse action.

Based on the timing of the early license-renewal announcement, ABC appears to have made a prima facie case—an initial showing sufficient to proceed—that the FCC retaliated against the network for carrying speech the Trump administration didn’t like.

The FCC resists this conclusion by relying on Red Lion Broadcasting Co. v. FCC and subsequent cases. It reiterates that broadcasters hold a unique “public trust” and that “aspects of broadcaster speech have long been subject to regulation.” In effect, the FCC challenges the first element by citing cases that afford broadcasters less First Amendment protection. From that premise, it argues that ABC has no clear protection against an investigation into whether it is operating in the public interest.

The FCC also disputes a causal link between its actions and ABC’s speech. It notes that its original investigation into ABC’s DEI practices predates the most recent comments against the network by President Donald Trump and FCC Chairman Brendan Carr by about a year.

The district court will have to decide whether that chronology supports the FCC’s explanation or whether the early renewal proceedings escalate a retaliatory investigation against a longstanding perceived enemy. Which way the court will go remains unclear.

Your Rights May Depend on Your Antenna

Whatever the court decides about retaliation, the FCC ultimately relies on its broad public-interest mandate over broadcast licensees to justify this investigation. But as I’ve written previously, the “bizarre notion that radio and television broadcasters deserve fewer First Amendment protections than newspapers, websites, or cable networks is a vestige of mid-20th-century technological reasoning that has long overstayed its constitutional welcome.”

The relevant distinction is that consumers can receive over-the-air broadcasts through an antenna. Yet much of today’s digital media also travels over radio waves, from a show streamed over a home Wi-Fi network to a TikTok video delivered over a 5G cellular network. Those services don’t face the FCC’s broad public-interest mandate. Taken far enough, the FCC’s reasoning could extend Red Lion’s reduced First Amendment protection to all kinds of speech simply because it travels over a radio signal.

The Supreme Court has yet to overrule Red Lion and related precedents, which held that spectrum scarcity permits content regulation “in the public interest” that could not apply to other media.

That justification doesn’t withstand scrutiny today, if it ever did. As I wrote previously:

The fact that there can be interference if frequencies aren’t allocated doesn’t mean a government agency needs to be in charge of allocating them according to whatever they define as “the public interest” at that given time. Whether through recognizing homesteading or auctioning licenses, the government could have found ways to allocate the frequencies such that they did not interfere with one another without imposing content-based restrictions on speech.

Whatever sense the FCC’s authority over broadcast content once made, it makes very little today. Broadcasters no longer have the market power they enjoyed when viewers’ choices largely consisted of ABC, the National Broadcasting Company (NBC), the Columbia Broadcasting System (CBS), and the Public Broadcasting Service (PBS). Fox, the United Paramount Network (UPN), the Warner Bros. network (the WB), and the CW later expanded those choices.

Today, cable, satellite, and internet-video services offer stiff competition for network television. Viewers commonly access even broadcast-network programming through those services, and relatively few rely on over-the-air antennas.

The media landscape has changed dramatically since the Supreme Court considered broadcasters’ First Amendment rights in Red Lion and FCC v. Pacifica Foundation in the 1960s and 1970s. There’s no reason to keep treating broadcasters differently when they compete—and are largely losing ground—against other video services. Giving ABC fewer speech protections than the Cable News Network (CNN) because viewers can receive it through rabbit ears makes very little sense.

Censorship Is a Bipartisan Opportunity

Without its broad public-interest mandate, the FCC would look much more like the government official in National Rifle Association of America v. Vullo. There, New York’s Department of Financial Services (DFS) argued that it was merely investigating insurance brokers and underwriters it regulated for alleged violations involving products offered to the National Rifle Association (NRA) and other gun-rights groups.

The Supreme Court explained that then-DFS Superintendent Maria Vullo was free to criticize the NRA and investigate possible violations of New York insurance law. But she “could not wield her power, however, to threaten enforcement actions against DFS-regulated entities in order to punish or suppress the NRA’s gun-promotion advocacy.”

Without its broad public-interest authority over broadcasters, the FCC would likely struggle to sustain the underlying investigation it used to justify the early license-renewal proceedings. With that authority, it can engage in considerable jawboning—informal pressure backed by regulatory threats—and backdoor censorship while disclaiming any such intent. The threat of an investigation or license-renewal proceeding may be enough to get broadcasters to do its bidding.

This power invites pressure from activists on both the left and right. During the Biden administration, for instance, left-wing activists at the Media and Democracy Project challenged a Fox affiliate’s broadcast-license renewal over “election misinformation.” Broad discretion over broadcast licenses creates an opening for censorship, whoever controls the FCC. The First Amendment demands more than a commission majority’s judgment about what serves the public interest.

ABC v. FCC illustrates the risks of administrative standards without clear limits. By turning “public interest, convenience, and necessity” from a mandate to allocate spectrum into sweeping authority over corporate employment practices, the FCC claims powers far beyond the Communications Act’s text. Policymakers and courts should reconsider whether this broad authority over broadcasters can coexist with the First Amendment.

Free speech shouldn’t come up for renewal.

The post The FCC’s Public Interest in Your Personnel Files appeared first on Truth on the Market.

In May 2025, the U.S. Justice Department (DOJ) and Federal Trade Commission (FTC) backed a lawsuit alleging that BlackRock, Vanguard, and State Street used their stakes in competing coal companies to coordinate output cuts. The following month, the European Commission fined Delivery Hero and Glovo for participating in a cartel facilitated by Delivery Hero’s minority stake in Glovo.

Testing whether ownership links foster collusion is tricky. Cartels are illegal, so researchers usually see only those that get caught. Those may look little like the ones that escape detection.

A recent working paper by Vincent Abraham, Florian Ederer, and Catarina Marvao sidesteps that problem with data from Sweden, where cartels remained legal until 1993. Firms had to disclose cartel agreements when authorities requested them. Failure to register carried penalties that included fines or up to six months’ imprisonment. The authors matched that registry with ownership records for publicly listed Swedish companies from 1985 to 1993.

Their key measure, a “profit weight,” captures how much a firm treats a rival’s profits as its own because they share owners or one holds shares in the other.

The Stockholm Stock Club

Abraham and his co-authors found that firms that gave more weight to a rival’s profits were more likely to belong to a cartel. The effect was modest. An increase of one standard deviation—a common measure of variation in the data—in the profit weight corresponded to a 0.44 percentage-point increase in the probability of cartel membership, from 7.64% to 8.08%.

That relationship ran in only one direction. Ownership ties predicted cartel membership for the firm that gave weight to its rival’s profits, but not for the rival itself. Both firms belonged to the same cartel in only about 0.1% of the firm-pair observations, leaving too few cases to draw firm conclusions. As the authors note, their model “is silent on which firms collude with which.”

A 2001 study of American farm cartels offers a possible answer.

One Firm Holds the Umbrella

Darren Filson, Edward Keen, Thomas Borcherding, and I studied federal marketing orders—legal cartels that let fruit, nut, and vegetable growers restrict the quantity and quality of produce reaching the fresh market. Of 182 crop-and-state combinations, only about a third ever maintained an order for at least three consecutive years.

We treated each cartel as a dominant seller competing with a “fringe” of smaller producers outside the agreement. When the cartel restricts supply, prices rise for everyone. Outsiders enjoy the higher price without limiting their own sales. In 1942, the United Date Growers of California complained that its members “grumble at holding the umbrella” for growers who stayed out.

Common ownership changes that arithmetic. If a cartel member and an outsider share owners, those shareholders collect part of the outsider’s windfall. That could help explain why ownership ties predict one firm’s cartel membership but not its rival’s.

The Swedish study’s authors don’t draw that connection, but their model points in the same direction. A firm’s gain from joining a cartel grows with the weight it gives to the profits of any rival that benefits from the cartel. An outsider selling at the cartel’s higher price fits the bill.

We also found that cartels were far more likely to form where a large cooperative—a business owned by its grower-members—already existed. A 20-percentage-point increase in the largest cooperative’s market share raised the probability of adopting a marketing order by about 19 percentage points. These cooperatives also exercised direct control. They voted on their members’ behalf and enforced quality standards.

A Stake Is No Steering Wheel

Abraham and his co-authors found a similar contrast between control and passive investment when they separated profit weights by source. Ownership by other listed companies—including direct stakes in rivals—showed a stronger link to cartel membership. Passive common ownership—in which funds, families, or other investors hold stakes in both firms—showed a smaller, less consistent relationship. It also corresponded to a lower likelihood that the rival belonged to a cartel. That fits the umbrella story. A rival with shared owners can stay outside the cartel and still collect the higher price.

Swedish ownership also looked different from its American counterpart. Groups such as the Wallenberg family often controlled firms through layers of holding companies and shares with extra voting rights. As the authors acknowledge, concentrated control may strengthen the link between ownership and coordination compared with markets where ownership is more dispersed.

The study thus supports close scrutiny of direct stakes between rivals, as in the Delivery Hero case. It offers much less support for the theory behind the lawsuit against BlackRock, Vanguard, and State Street.

A central question remains. The ownership links existed before the firms joined cartels, consistent with the idea that shared owners helped pave the way for collusion. But it also fits a simpler explanation: Investors who anticipated a profitable cartel may have bought shares before it formed.

Distinguishing those explanations requires ownership changes unrelated to cartels. Suppose a bank merger transferred a block of shares to a new owner for reasons unrelated to collusion. If firms that suddenly shared that owner then joined cartels more often, that would strengthen the case that ownership helped cause collusion. If they didn’t, the evidence would offer less support for that explanation and leave open the possibility that investors were simply anticipating cartel profits.

For now, the Swedish findings echo what American farm cartels suggested 25 years ago. Collusion comes more easily when owners have a hand in running competing firms. Simply holding shares in both appears to matter less.

A shared shareholder is one thing. A hand on both steering wheels is quite another.

The post How to Profit From a Cartel Without Joining One appeared first on Truth on the Market.

The inspiration comes from the Financial Interest and Syndication Rules, better known as Fin-Syn, which once restricted broadcast networks’ ownership of programming and their ability to profit from reruns. Supporters want a streaming-era version to revive independent production, preserve American jobs, and promote creative diversity, as they explained on a recent episode of the industry podcast The Town.

It’s a bad proposal with an understandable constituency. Production employment has contracted since the “peak TV” era, and producers dislike the terms streamers offer. A guaranteed slice of their budgets has obvious appeal.

But the conditions that once perhaps justified a system like Fin-Syn have long since disappeared. Viewers now choose among broadcasters, cable networks, streaming services, and online creators. Vertical integration—combining production and distribution under common ownership—can help finance programming and deliver it efficiently. Government spending mandates would put those benefits at risk and could even invite First Amendment challenges.

There’s No Business Like Guaranteed Business

The group campaigning to revive Fin-Syn calls itself the American Entertainment Alliance. Its members include Creative Artists Agency, William Morris Endeavor, and United Talent Agency, along with producers Ron Howard, Brian Grazer, and Jason Blum. Their reported proposal would reserve half of major streamers’ original-content budgets for independent U.S. productions.

As supporters explain on The Town, they believe platforms that both produce and distribute programming control too much of the business—from deciding which shows get made to distributing them and owning the rights. Independent producers, they argue, get fewer opportunities and a smaller share of the rewards.

In some ways, the proposal would go beyond the original intervention. The Federal Communications Commission (FCC) adopted Fin-Syn in 1970, restricting broadcast networks’ financial interests in independently supplied programs and their participation in syndication—the licensing of shows for reruns on other outlets. The new version would also require streamers to buy from designated suppliers.

The financial appeal is straightforward. Producers would gain guaranteed demand and potentially more opportunities to retain valuable rights. Talent agencies could collect commissions on more deals, or more lucrative ones. That explains the enthusiasm among prospective recipients. Whether viewers would benefit at all is a separate question.

There’s a familiar tension here. Sellers who complain about powerful buyers want Washington to strengthen their collective position. Antitrust law ordinarily treats naked horizontal price fixing—agreements among competitors to set prices—as per se illegal, dispensing with the usual “rule of reason” inquiry into competitive harms and benefits. Like bid rigging and market division, horizontal price-fixing agreements predictably harm competition and reduce consumer welfare. 

This campaign doesn’t explicitly involve an agreement to set prices, and joint lobbying generally enjoys antitrust protection. The American Entertainment Alliance members are instead asking Congress to reserve spending for a favored class of suppliers. The issue with that plan is that strengthening sellers’ bargaining position can shift money toward them without improving what buyers—or viewers—receive.

The Journalism Competition and Preservation Act (JCPA) offered a related approach. It would have given eligible news publishers and broadcasters an antitrust exemption to negotiate compensation collectively with large digital platforms and jointly withhold content. Fin-Syn’s advocates seek a purchasing quota rather than permission to bargain as a group, but the shared premise is that concerns about powerful technology companies supposedly justify government-backed leverage for their suppliers. In both cases, the missing step is explaining why the public should expect to benefit.

There’s also another catch. A spending quota guarantees a share of the budget, but says nothing about its size. A streamer facing higher costs could cut total spending, commission fewer projects, or raise subscription prices. Advocates need to explain how those responses would jibe with the promised revival in domestic production.

Canceled for a Reason

Fin-Syn’s original premise was that the three broadcast networks—ABC, CBS, and NBC—functioned as an oligopoly, controlling access to a mass audience. But even then, critics spotted a financing problem. Restricting networks’ ability to buy program rights also restricted producers’ ability to sell them—and to secure financing and share risks with willing buyers. The rules could hurt smaller entrants while shielding established suppliers from competition.

In Schurz Communications, Inc. v. FCC, the 7th U.S. Circuit Court of Appeals vacated the FCC’s revised rules in 1992. Judge Richard Posner faulted the commission’s economic reasoning, including its assumptions about protecting independent producers and encouraging programming diversity. The court demanded a compelling justification for the restrictions.

Meanwhile, the expansion of cable television and the then-recently launched Fox network had loosened the three major networks’ grip on viewers. In 1993, the U.S. Justice Department (DOJ) supported removing related consent-decree restrictions, citing the changed market conditions. The FCC eliminated its remaining rules in 1995.

Fin-Syn revival advocates thus have two problems. The original rules had major economic flaws, and the market that supposedly justified them has changed beyond recognition. Nostalgia for TV’s Golden Age addresses neither of these concerns.

Nevertheless, Television Persisted

Television had begun changing well before repeal. By the 1990–91 season, the Big Three broadcast networks’ combined audience share had fallen to 61%. Cable offered 79 programming networks by 1989. By 2020, NCTA–The Internet & Television Association described an industry with more than 200 programming networks, alongside the growing streaming business.

More outlets meant more opportunities to make shows. This really took off in late 2000s, when home broadband connections improved enough to support the emergence of the SVOD (subscription video on demand), vMVPD (virtual multichannel video programming distributor), and AVOD (ad-supported video on demand) formats—today, known collectively as “streaming.” FX Research’s tally of original scripted series rose from 210 in 2009 to roughly 600 in 2022, before falling to 516 in 2023.

Employment also grew over the longer period. A Bureau of Labor Statistics (BLS) series shows the number of jobs in the motion-picture and video industries rising from 290,800 in 1993 to 492,200 in 2022. To be fair, the industry’s more recent contraction has been painful. A narrower BLS production-payroll series shows a steep decline after October 2022.

Counting the number of TV series also tells us little about how many episodes each show orders or how long its writers stay employed—issues that writers’ contracts address directly. The reality is that viewers can gain even as particular producers and workers struggle.

But it’s also important to remember that viewers’ gains include more specialized genres, more access to international programming, and more ambitious storytelling. Amanda Lotz’s research on television business models helps explain why. Different ways of financing and distributing television support different kinds of programs. Subscriptions can reward shows that attract particular audiences, including viewers whose tastes mass-market, advertiser-supported programming underserves.

In the New York Times’ recent ranking of the 100 greatest shows of the 21st Century thus far, only 13 were series produced for U.S. broadcasters. Another 13 were produced for streaming services, 19 were produced for international (primarily British) television, and 55 were produced for cable television.

Fin-Syn repeal alone didn’t create this abundance. Technology and new competitors helped. Still, television’s creative achievements under integrated ownership should give advocates of mandatory separation pause. The industry has produced rather a lot for a business supposedly organized all wrong.

Integration can also coordinate production and distribution and reduce costs. When separate companies each add a markup, the combined price can discourage sales, a problem that common ownership can ease. This is as true in entertainment markets as in other sectors. Ricard Gil’s study of the Paramount divestitures—the court-ordered separation of movie studios and theaters—found that integrated theaters charged lower prices and sold more tickets. Luming Chen, Lisa Xuejie Yi, and Chuan Yu found that integration increased consumer benefits overall in China’s movie industry, although the effects varied across markets.

These studies don’t establish that every entertainment merger benefits consumers, but they do show why a blanket presumption against vertical integration makes for poor policy. Common ownership can help a company recover its investment in programming through distribution revenue. Compulsory outsourcing could sacrifice that advantage without producing a single additional show worth watching.

That’s a Lot of Companies for a Bottleneck

Fin-Syn revivalists argue that television has come full circle, leaving producers with something like the three-network bottleneck of old. The data tell a different story. Even after Paramount completed its merger with Warner Bros. earlier this month, a substantial number of companies compete in the market. How concentrated it looks depends on which companies we include and what we count—revenue, viewing time, subscriptions, or interest in particular titles. Those measures answer different questions, but they hardly describe a return to three-channel television.

A common measure of concentration is the Herfindahl-Hirschman Index (HHI), calculated by squaring each firm’s percentage market share and adding the results. Higher scores indicate greater concentration, and the 2023 merger guidelines classify markets above 1,800 as highly concentrated. Combining Paramount’s and Warner Bros.’ historical shares lets us assess the merger under several measures.

Pawel Popiel and Andrew deWaard estimate 2025 streaming revenue in their critique of the Paramount-Warner Bros. merger. But even they find an online-video HHI of 1,543, below the “highly concentrated” threshold. Their figures put the combined company at about 16% of revenue, behind Netflix, Disney, Alphabet, and Amazon.

Sticking with 2025 data for consistency’s sake, the narrower JustWatch data for January–September 2025 put Disney first at 24%, combining Disney+ and Hulu. Amazon Prime Video follows at 21%, Netflix and the combined Paramount+–HBO Max business each hold 20%, and Apple TV+ has 8%. To be fair, this comparison produces an HHI above 1,800 and warrants scrutiny. But JustWatch measures watch-list additions, platform clicks, and similar interactions. Those figures capture interest in subscription services rather than actual subscriptions or viewing time.

For viewing itself, arguably the best metric of market concentration, Nielsen’s July 2025 data offer a broader picture. YouTube leads with 13.4% of total U.S. television viewing, just ahead of Paramount and Warner Bros.’ combined 13.1%. Disney follows at 9.4%, Netflix at 8.8%, NBCUniversal at 7.6%, and Fox at 6.5%. (The table reports HHI ranges to reflect uncertainty about the number of firms in the “other” category.)

Including YouTube is a somewhat controversial choice, which is why the table above splits out market share and concentration both with and without the online video giant. But it certainly makes sense to include it when viewers can and do substitute its programming for conventional television. Consumers don’t typically organize their evenings around the industry’s preferred categories.

The multistate merger challenge led by California Attorney General Rob Bonta also adds some perspective to the question of whether streaming is an uncompetitive market. The suit expressly excluded streaming from the markets the Paramount-Warner Bros. merger allegedly made unlawfully concentrated, focusing on theatrical distribution and cable-channel licensing. That omission alone cannot establish competitiveness, but it offers little support to the claim that streaming has returned us to the three-network era.

Taking the Script to Congress

There is, of course, also a separate competition question: How much competition is there among the companies buying producers’ shows? As their discussion on The Town makes clear, this is the market that chiefly concerns Fin-Syn advocates.

Their implicit claim concerns monopsony—buyer power that allows purchasers to push suppliers’ compensation below competitive levels. A streaming service can face vigorous competition for viewers while producers have few practical alternatives for selling their work. Audience competition alone cannot settle that issue.

The advocates appear to want Congress to mandate purchases, rather than a court to remedy an antitrust violation—perhaps reflecting the reality that demonstrating monopsony power in court is notoriously difficult. That shift in venue changes the debate, but the principles of competition law should still guide the policy. In considering such proposals, lawmakers should ask whether buyers can suppress compensation below competitive levels, how that power affects production, and whether the proposed remedy would improve competition. A disappointing offer does not, by itself, establish a competition problem.

Even if the market of streaming-service buyers is exceedingly constrained, which is far from obvious, producers also have alternatives in international buyers and self-distribution through YouTube. To be fair, a place to upload a finished show is little help if nobody will finance the expensive drama you want to make. The relevant question is which alternatives give producers meaningful bargaining options. Advocates need to explain where those options fall short and how compulsory outsourcing would address the resulting harm.

The consequences for viewers also deserve attention. Consider the DOJ’s challenge to Penguin Random House’s proposed acquisition of Simon & Schuster—one of the more notable recent monopsony cases. The government argued that eliminating competition between the publishers would reduce advances—the upfront payments authors receive—for anticipated top-selling books. In October 2022, Judge Florence Pan blocked the merger in United States v. Bertelsmann, finding that it would substantially lessen competition for those books’ U.S. publishing rights.

The deal then unraveled. Penguin Random House initially planned to appeal, but Simon & Schuster’s owner (coincidentally, it was Paramount Global) terminated the agreement the following month. Which is not to say that the publishers accepted the government’s entire case, but during the litigation, they had not disputed a narrower proposition: If advances fell significantly, some authors could no longer afford to write, leaving readers with fewer books and less variety.

Brian Albrecht, Dirk Auer, and Geoffrey Manne use that connection to explain why monopsony analysis should examine both input markets (where businesses buy labor or rights) and output markets (where they sell the resulting products). Lower supplier payments alone, they argue, cannot distinguish harmful buyer power from genuine efficiencies. A business might need fewer inputs because it has eliminated duplication while maintaining or increasing production. Looking at output helps distinguish the two.

The publishing case offered a relatively straightforward example because the acknowledged harm to authors would also reduce output and hurt readers. The court’s holding concerning competition for publishing rights doesn’t by itself establish a universal requirement to prove separate consumer harm in monopsony cases, but Albrecht, Auer, and Manne’s broader economic point remains relevant to Fin-Syn. Improving producers’ bargaining position is not enough to establish that a purchasing mandate would deliver more or better television.

Of course, Congress doesn’t conduct an antitrust trial before legislating. But a law guaranteeing producers more business still needs a public-interest justification. Fin-Syn advocates should explain how their purchasing mandate would produce more programming, better shows, or greater innovation. Moving the proposal to Capitol Hill hardly makes those questions disappear.

The Constitution Has Notes

The analogy to the Big Three-era broadcast rules that Fin-Syn revivalists like to raise also obscures a constitutional problem. Broadcasters use licensed spectrum and, as a consequence, receive reduced First Amendment protection in some contexts. In Reno v. ACLU (1997), the U.S. Supreme Court rejected extending that treatment to the internet. Moody v. NetChoice (2024) reinforces protection for online publishers’ editorial decisions about what third-party speech they wish to carry and what they don’t.

A spending quota would constrain streamers’ choices about which expression to finance and distribute. A rule dictating what they must carry would intrude even more directly on editorial discretion. The goal of supporting independent producers does not resolve either measure’s constitutional problems.

To be fair, the Turner Broadcasting decisions (1997) upheld federal “must-carry” rules requiring cable operators to carry local broadcast stations, despite Turner’s First Amendment challenge. The Court applied intermediate scrutiny, which requires adequate justification and tailoring, and it was crucial to upholding the rules that Congress had found cable operators controlled a “bottleneck” into subscribers’ homes. Refusing carriage could therefore deprive local broadcasters of access to viewers and threaten their survival.

That distinction matters to Justice Brett Kavanaugh. As a judge on the U.S. Court of Appeals for the D.C. Circuit, he dissented from the denial of rehearing in United States Telecom Association v. FCC, the 2017 net-neutrality case. He read Turner to require a showing of bottleneck market power before the government could justify overriding providers’ editorial discretion. Simply invoking greater diversity or more voices would not suffice. His position highlights the importance of the market conditions behind the constitutional analysis.

As the market evidence discussed above shows, today’s streaming services compete alongside broadcasters, cable networks, and YouTube. A streamer that declines a show doesn’t control the physical connection through which viewers reach competing services. Producers may face difficult financing choices, but that’s a different problem altogether from the cable bottleneck Congress identified in Turner. Advocates can’t just borrow the must-carry precedent while skipping the market-power showing that helped justify it.

Thus, a purchasing quota would likewise require its own justification. Reserving half of a streamer’s original-content budget for favored suppliers demands evidence both of a problem warranting intervention and of a reasonable fit between that problem and the remedy. The concentration measures and buyer-power questions discussed above therefore matter to the constitutional analysis, too. Advocates would need to explain how inadequate competition harms producers and why compulsory outsourcing would address that harm. Producers’ dissatisfaction with streamers’ contract terms doesn’t supply either answer.

And the older broadcast precedents offer no firmer foundation for extending regulation to streaming. Ben Sperry argues that modern media abundance undermines Red Lion’s (1969) reliance on spectrum scarcity and Pacifica’s (1978) reliance on broadcasting’s pervasive presence. Both precedents currently remain binding, but should the Court reconsider them (which doesn’t seem impossible), they could narrow the FCC’s authority to regulate speech under its public-interest mandate.

Too Many Governments in the Writers’ Room

Fin-Syn advocates also compare their proposal to local-content requirements in Canada and Europe. Those precedents offer little help with the U.S. constitutional question, since they operate under different speech protections. They also raise a practical question. What happens when several governments claim a share of the same streamer’s resources?

The European Union requires European works to make up at least 30% of streaming catalogs and permits member states to layer on additional investment obligations. Add a U.S. requirement to spend 50% of original-content budgets on independent American productions, and the restrictions could compound. It’s not that you could simply add up the percentages to conclude that streamers would control only the remaining 20%, as one measures catalog titles and the other spending. Their interaction would depend on which budgets, services, and territories the U.S. law covered. Streamers would also retain choices among qualifying productions.

The concern is less tidy than that arithmetic, but still real. A service subject to overlapping mandates would have less flexibility to finance the projects its viewers most want. Money committed to satisfying one jurisdiction’s rules would be unavailable for other investments. Any given qualifying production might be excellent, and viewers might love it, but the risk is that meeting quotas takes priority over judging a project’s quality, appeal, or cost.

Giuseppe Colangelo and Kristian Stout explain how European requirements can inflate production costs, disadvantage smaller producers, and fragment national markets. Canada, meanwhile, announced a reconsideration of its streaming requirements in June, citing affordability concerns. These are reasons to examine the cumulative burden before adding another mandate.

U.S. adoption could also encourage imitation. Other governments already have their own authority to impose content rules, but an American quota would give them a ready political justification—and make Washington’s objections to similar treatment of U.S. programming harder to sustain. Each country could defend its own carveout while leaving streamers to reconcile the demands.

Government mandates might make claims on how to divide up the budget, but they still can’t guarantee an audience.

Lights, Camera, Lobbyists

American jobs and distrust of powerful technology companies give Fin-Syn advocates a pitch with bipartisan appeal. Turning it into legislation would mean deciding which producers qualify, which services must pay, who enforces the quota, and how it survives constitutional scrutiny.

Those details would also invite more lobbying. Reserving half of streamers’ original-content budgets for selected suppliers creates a valuable government-granted privilege. Producers and talent agencies would have every incentive to shape the eligibility rules, broaden the obligations, and defend their share. That is rent seeking—pursuing financial gains through political favors. Time and money spent securing a place inside the quota would do nothing to make the programming better.

The industry’s recent contraction deserves attention. But the old broadcast bottleneck has disappeared, integrated businesses can benefit viewers, and a purchasing mandate guarantees neither creative diversity nor shows people want to watch. Washington should demand evidence that the proposal would improve viewers’ choices before guaranteeing Hollywood’s intermediaries more business.

The post Fin-Syn Is a Rerun Washington Should Skip appeared first on Truth on the Market.

On July 23, the Commission fined Alphabet €460 million for favoring nine of its own services on Google’s search results page: comparison shopping, flights, trains, hotels, local places, Google Maps, a translation box, sports scores, and even the games that appear when you search for “solitaire.”

The Commission’s cease-and-desist order requires Google to stop “displaying interactive pointers to its own distinct or additional services on the [Google Search Engine Results Page (SERP)], such as in tabs or in links to these services when a user clicks on a reservations, directions or any other button” (paragraph 712(c)).

That raises a basic question about the reach of the European Union’s Digital Markets Act (DMA). Article 6(5) prohibits designated “gatekeepers”—large platforms subject to the law—from giving their own services preferential treatment in “ranking or related indexing and crawling.” These processes determine the order of search results and collect and organize the information behind them.

But the Commission’s order also restricts where Google places links and buttons for its own services on the page—even when those features don’t appear to rank search results.

Does “ranking” now cover anything that appears on the screen? If so, gatekeepers could be barred from featuring any of their own products or services in drop-down menus on their own websites. We argue that this stretches the DMA’s definitions of “ranking” and “search results” too far, with no clear limiting principle.

Sharing a Screen Isn’t Enough

Article 6(5) forbids a gatekeeper to “treat more favourably, in ranking and related indexing and crawling,” its own services and products than similar offerings from third parties. Articles 2(22) and 2(23) of the DMA define “ranking” and “search results,” respectively. Read together with the law’s designation provisions, they establish three limits.

First, the obligation applies within the designated “core platform service”—the specific service covered by the gatekeeper’s designation (Article 3(10)). Recital 51, an explanatory passage in the DMA, describes the problem as favoring the gatekeeper’s offering over third-party offerings “also operating on that core platform service.”

Here, the designated service is Google Search, which lets users “input queries in order to perform searches,” as the Platform-to-Business (P2B) Regulation puts it. The designation covers that search service, not google.com as a whole or the entire Google ecosystem.

Second, for search engines, ranking means “the relevance given to search results,” and search results are information “returned in response to, and related to, a search query.” The definition casts a wide net: It covers any format and anything “offered in connection with the organic results”—the unpaid search listings. But a query must still generate the item. Something that isn’t returned in response to a query can’t be a search result and therefore can’t be ranked.

Third, for marketplaces, ranking concerns goods and services “offered through” the platform. This part of the definition doesn’t require a query, but it still has a boundary: It covers goods and services offered for consumers to buy through the marketplace, including the platform’s own competing offerings. Recital 46 describes this “dual role”: The platform both runs the marketplace and competes with sellers on it.

A link on the marketplace page to a separate service doesn’t, by itself, make that service an offering sold through the marketplace or subject to its ranking rules.

There’s a principle that connects these limits: “Ranking” concerns the output of the core platform service’s defining function. For a search engine, that output is search results. Google’s shopping, hotel, and flight boxes would therefore fall within Article 6(5) under its Google Search designation: They appear in response to a query, relate to it, and appear above rival links. (We happen to think a per se rule—a categorical ban—against favoring one’s own offerings is bad policy, but that’s a different story.)

But the provision doesn’t extend to everything that happens to share the screen with those results.

No Button Left Behind

Rather than start with the DMA’s definitions, the Commission relies on its 2023 designation decision, which Alphabet did not appeal. That decision declared that “all results or outputs” Google Search displays on its results page “constitute search results” (paragraph 27).

The Commission uses that premise to expand Article 6(5)’s scope in three moves.

The Commission starts with tabs. A hotel query produces a “Lodging” tab that opens a full page of Google’s own hotel-search feature. The Commission treats this as self-preferencing—favoring its own service—because “no third-party CSS gets such a tab exclusively for its own inventory” (paragraph 253). CSS stands for “comparison shopping service,” which lets users compare sellers’ offerings. The Commission’s logic suggests that Google must give rival comparison shopping services, such as Idealo and Kelkoo, their own tabs—or remove the tab for its own service.

It then extends that reasoning to buttons. Clicking a hotel opens a preview with a “directions” button linking to Google Maps. The Commission reasons that anything “offered in connection with the organic results” counts as a search result, so the button “ranks” Google Maps (paragraphs 496–497). It ranks Maps more favorably, the Commission says, because “no similar services can be reached after clicking the directions button” (paragraph 514).

But that establishes only that the button leads to Google’s service. It says nothing about how Google ranks services by relevance. If the button’s mere existence amounts to self-preferencing, Google must either sever the connection between its hotel-search feature and Google Maps or link every hotel aggregator—a service that collects hotel listings—to every mapping service through a series of dedicated buttons arranged at random.

The order itself takes the final step. Paragraph 712(c) covers tabs, links, and “any other button,” without tying them to a query or referring to the statutory definitions. Read literally, the order also covers Google’s app menu and permanent links to its other services—even when they appear regardless of what the user searches for.

The Commission hasn’t declared Google’s logo a form of ranking, but its order offers no clear reason to exclude it. By that logic, Google should presumably display competitors’ logos alongside its own on the Google.com homepage. At that point, users might need help figuring out whose search engine they’re using.

Compliance by Subtraction

Why does this matter? To start, because the Commission’s approach requires no showing of harm (paragraphs 86–87), gives no weight to user expectations (paragraph 64), and rejects consumer benefits or quality tradeoffs as a defense (paragraphs 189, 255, 512, 515). That leaves the law’s defined scope as the only check on overreach. If the DMA mistakenly bans practices that help competition and consumers, strict limits on where it applies are the only safeguard against wider market harm.

Europeans can now see what that harm looks like. On Sept. 8, Google rolled out a redesigned results page for EU users. The updated hotel and flight search results no longer show current prices or descriptive labels, and users can no longer filter those results by travel dates. A senior Google executive called the overhaul the largest reduction in search quality in the company’s history.

Google has an obvious incentive to portray the changes as harmful, but that doesn’t mean its criticism is wrong. If the Commission can restrict anything on the page without considering whether users lose useful features—and the DMA’s review process cannot recognize those losses as a failure—what stops enforcement from making search steadily worse?

If regulators can’t weigh a practice’s benefits before banning it, the law must narrowly define what they can ban to avoid sweeping useful features into the prohibition. The DMA’s architects recognized as much. They justified omitting an “efficiency defense”—a chance to show that a practice’s benefits outweigh its harms—by promising clearly defined, tightly limited obligations (recital 65).

The DMA itself recognizes that gatekeepers need incentives to keep improving their services. It says diminished “contestability”—the ability of rivals to challenge an incumbent—“reduces the incentives to innovate and improve products and services for the gatekeeper” (recital 32). When the law is ambiguous, its interpretation should therefore preserve the gatekeeper’s incentive to improve its core platform service. Otherwise, enforcement risks producing the very stagnation the DMA aims to prevent.

Two features of the Commission’s approach suggest that what ultimately matters here is whether Google owns the service—not whether it treats competing services equally.

The Commission acknowledges that Article 6(5) doesn’t require equal treatment across the board. It targets favoritism toward Google’s services, “and not favouring of any group of business users of the OSE over another”—OSE being shorthand for “online search engine” (paragraph 89). If Google can give Booking.com a prominent paid placement but can’t include a directions button linking to Google Maps, the distinction turns on who owns the service receiving that prominence.

The Commission also offers Google two ways to comply: Give every “similar” rival equal prominence, or “remove all prominence” from its own services (paragraph 713). For a directions button, the first option is hardly workable without wrecking the connections among Google’s services. Imagine opening a hotel listing and finding dozens of “directions” buttons—one for each competing mapping service. In practice, Google instead has an incentive to comply by removing useful features, such as the directions button, rather than crowding the page with buttons for every rival.

A Prime Example of Overreach

The consequences may not stop at Google’s search page. Amazon offers a useful test of how far the Commission’s interpretation could reach—and where the DMA’s limits should hold.

The Commission announced investigatory steps into Amazon’s treatment of its own products in March 2024. That November, Reuters reported that a formal investigation was likely, although no formal proceedings appear to have followed. If they do, Amazon’s natural defense will be that many features the Commission might challenge—such as promotional boxes and menu links for Prime Video, Amazon Music, and Alexa—are matters of website design, not ranking.

Under the framework outlined above, Amazon would be half right. Its own retail offerings and private-label goods compete with third-party sellers for the same customers in the same marketplace. That’s the “dual role” described in recital 46. Wherever Amazon promotes those goods within Amazon Store, it offers them “through” the platform. They are the output of the marketplace’s defining function, so Article 6(5) applies.

Links to Prime Video and Alexa, however, would be different. Those services aren’t marketplace offerings, and a Prime Video link takes users outside the designated core platform service.

In addition, Article 6(5) requires a similar third-party service “also operating on that core platform service.” Netflix isn’t a seller on Amazon Store, so a Prime Video link doesn’t favor Amazon’s service over a competing offering within the marketplace.

Yet paragraph 712(c) of the Google decision, coupled with the Commission’s demand for equal prominence for each similar rival, draws no such distinction. It identifies the following as preferential treatment:

Displaying interactive pointers to its own distinct or additional services on the Google SERP, such as in tabs or in links to these services when a user clicks on a reservations, directions or any other button on the Google SERP with more prominence than similar services offered by third parties, as set out in recitals (490) – (517) of this Decision.

Apply that logic to Amazon, and it would have to add Netflix and similar rivals to its menu—or remove its own streaming-service link.

‘Any Other Button’ Meets Its Judge

Google faced a Sept. 21 compliance deadline and an early October deadline to challenge the decision before the General Court, the EU court that reviews Commission decisions. The DMA’s obligations apply regardless of their effects (recital 11). And the Google Shopping ruling that self-preferencing is not abusive as a general rule came two years after the DMA had already made it per se unlawful for gatekeepers.

The legal fight will therefore turn on what Article 6(5) actually covers: the ranking of search results or every link and button on the page. Courts are well equipped to interpret defined terms; they don’t need a theory of market power to decide what counts as a search result.

In June, the General Court annulled Meta’s Marketplace designation because the Commission hadn’t explained its reasoning well enough for the court to review it. Three paragraphs of analysis supporting an order that reaches “any other button” invite similar scrutiny. So does a fine justified by the provision’s supposedly “clear language,” when the DMA itself anticipates that its requirements will need further specification (paragraph 648).

The Commission Needs a Stop Button

For all the DMA’s flaws, its drafters set boundaries: Search ranking concerns results returned in response to user queries, and marketplace ranking concerns products and services offered through the platform. Those limits tie the law to specific activities within designated services.

By stretching those definitions to cover directions buttons, static tabs, and potentially “any other button,” the Commission erases those boundaries. A regulation meant to help rivals compete risks discouraging improvements, stripping out useful features, and leaving consumers worse off.

The DMA promises more competition. It shouldn’t deliver less search.

The post Every Click You Make: Article 6(5)’s Search for a Limiting Principle appeared first on Truth on the Market.

On Sept. 3, after a three-year process, the European Commission adopted its first guidelines on exclusionary abuses under Article 102 of the Treaty on the Functioning of the European Union (TFEU). Article 102 prohibits companies from abusing a dominant market position, including through practices that improperly shut out competitors. The guidelines replace the 2008 Guidance Paper on Enforcement Priorities for Exclusionary Abuses. The question here is whether jurisdictions in Latin America, Asia, and Africa should follow their approach.

The reach of the “Brussels effect”—the influence of European Union (EU) rules abroad—is debatable, but the influence itself is clear. Since the EU adopted the Digital Markets Act (DMA), which imposes special obligations on large digital platforms designated as “gatekeepers,” the Common Market for Eastern and Southern Africa (COMESA) has overhauled its rules along similar lines. Japan has enacted its Mobile Software Competition Act, while Brazil, Vietnam, and others have debated their own ex ante regimes—rules that apply upfront, rather than through enforcement after alleged misconduct.

Of course, competition agencies can borrow another jurisdiction’s guidelines more easily than its regulations: Guidelines look like technical advice, attract less political scrutiny, and require no new legislation.

And, indeed, some provisions may be worth borrowing, but wholesale adoption risks importing dubious economics and ill-suited responses to Europe’s political and institutional problems. Those flaws could prove especially harmful elsewhere, while adopting Brussels’ approach to digital rules now carries a concrete risk of trade retaliation.

A Presumption Too Far

Other jurisdictions should examine the EU’s guidelines carefully before borrowing their framework.

There are welcome provisions. From the outset (¶ 2), the guidelines recognize that Article 102 TFEU “does not prevent an undertaking from acquiring a dominant position on a given market based on its own merits, in particular its skills and abilities.” Whether recent EU case law has honored that principle is debatable, but its inclusion is a positive development.

The guidelines also clarify that Article 102 TFEU “does not preclude the departure from the market or the marginalization, as a result of competition on the merits, of competitors that are less efficient than the dominant undertaking and thus less attractive to consumers.” They call it “helpful” to explain “the economic mechanism through which the conduct is capable of directly or indirectly harming consumers”—the “theory of harm.” These are sound, if largely established, principles of abuse-of-dominance law.

The expanded discussion of objective justification and efficiencies—legitimate reasons for conduct and the economic benefits it produces—is a real improvement over the five paragraphs it replaces. The cautious opening to out-of-market efficiencies, or benefits outside the market under investigation (¶ 237), is also a refinement other jurisdictions could usefully adopt.

But the positives largely end there, particularly when the guidelines venture into new territory. Consider the empirical claim in ¶ 4:

In view of growing market concentration in various industries and of the digitalisation of the European Union economy, which makes strong network effects (8) and ‘winner-takes-all’ dynamics increasingly common, it is important that Article 102 TFEU is applied vigorously and effectively. 

Network effects arise when a service becomes more valuable as more people use it, potentially helping a leading firm capture much of a market. But the premise of growing concentration—larger shares of business in fewer firms’ hands—is debatable, and the guidelines cite no supporting literature. Concentration’s implications for consumers and economic welfare are also less clear-cut than this passage suggests. As the International Center for Law & Economics (ICLE) explained in its comments on the draft guidelines:

Market concentration “is not, in itself, a bad thing; indeed, recent research challenging the standard account demonstrates that much observed concentration is driven by increased productivity, rather than by anticompetitive conduct or anticompetitive mergers.” There’s no empirical evidence that market concentration has been increasing or that it is necessarily leading to increased market power or harm to the competitive process or consumers. (Citations omitted)

The growing emphasis on digital markets, both in competition policy generally and in these guidelines, raises another question: Is enforcement targeting the markets that most need it? As Herbert Hovenkamp put it:

Antitrust enforcement produces good results when it directs its resources at clearly harmful practices in sluggish markets that show little economic progress. Such markets can benefit from a kickstart. That applies to both identifying violations and creating remedies. An antitrust remedy in a poorly performing market stands a good chance of making the market work better and is less likely to cause serious harm. After all, there are few places to go except up.

Internet technology, including the large firms we call “big tech,” does not come close to satisfying these criteria. The rise of internet commerce and emergence of large digital firms is a tidal wave whose effects are still in progress. Tech firms have put far more into the economy than they have taken out.

(…) 

Online markets also do very well when measured on the demand side. Internet prices are usually competitive with offline prices for the same products, and often lower. Customer search costs are almost always lower as well. The customer unhappy with a product or price at Home Depot can get in her car and compare prices at Walmart. The internet customer can do the same thing in a second with a mouse click. In addition, information about prices, quality, or consumer satisfaction is almost always easier to obtain. Someone examining a power tool while standing in a physical Walmart or Home Depot store does not have immediate access to the experiences of other customers. If she wants to obtain it, she will probably go online.

Neither Hovenkamp nor I argue that digital markets deserve a “free pass.” But digitalization alone does not justify more aggressive antitrust enforcement.

Despite this weak empirical foundation, the guidelines seem to push EU competition law in the DMA’s direction, without ever mentioning it.

Consider “self-preferencing,” or a firm favoring its own products or services on its platform. The guidelines appropriately state that “(t)here is no general rule that self-preferencing is problematic from a competition standpoint, even when carried out by dominant undertakings.” Yet the factors listed in ¶ 196 make it easier for the Commission to establish an abuse or require the company to defend its conduct, rather than requiring the Commission to prove that it harms competition.

One factor asks whether preferential treatment amounts to an “unjustified difference in treatment”—because it conflicts with the business rationale or because “in the leveraging market there is an expectation of neutrality or openness.” Here, the leveraging market is the market where the firm allegedly uses its power to benefit a product in another market. Another factor asks whether the treatment influences users “irrespective of the intrinsic qualities of the leveraged product.”

A firm may rank, integrate, or make its own service the default for many reasons: faster performance, security, lower transaction costs, quality control, or liability. An expectation of neutrality does little to distinguish legitimate choices from harmful ones.

The guidelines also identify “conduct that is by its very nature harmful to competition” (¶¶ 198-201), including:

… in a context where the dominant undertaking holds both regulatory and commercial functions and finds itself in a position of conflict of interest, the establishment and implementation of rules and sanctions that, in the absence of substantive criteria and detailed procedural rules that ensure they are transparent, objective, precise and non-discriminatory, give the dominant undertaking the power to deny at its discretion any competing undertaking access to the market. 

The guidelines cite the European Superleague case, but this description could equally apply to recent cases involving “digital ecosystems”—interconnected products and services organized around a platform. As I have explained elsewhere, “ecosystem” has clear descriptive value but “risks becoming a doctrinal shortcut”: The same features can serve as evidence of both dominance and abuse, making the reasoning circular.

Here, the guidelines treat a company’s power to set rules on its own platform—an ordinary exercise of property rights and freedom of contract—as grounds for a special obligation. Access rules and terms of use must be “objective, precise and non-discriminatory.” That requirement could exclude subjective but commercially legitimate judgments about trust, taste, or the mix of businesses that makes a platform attractive to users.

If the rules fail that standard, the guidelines classify the conduct as inherently harmful. Once it falls into that category, “the conduct is deemed as such to distort effective competition” (¶ 199). Neither a digital platform nor an organization that sets rules for market participants should face a presumption of competitive harm simply because it controls access. Yet the guidelines extend that presumption to firms with similar powers in other markets.

For example, the description fits port and terminal operators that set access rules for competing stevedores—the businesses that load and unload ships. It also fits payment networks that set participation rules for banks handling card transactions, airport concessionaires, electricity system operators, and financial exchanges and clearing systems. All set access terms that give them a form of “regulatory” authority over rivals.

Some of these markets have genuine competition problems that warrant enforcement. Many do not. Presuming harm rather than demonstrating it—and declaring rebuttals “very unlikely to be successful” (¶ 201)—effectively turns dominance rules on their head. It risks condemning legitimate business conduct as anticompetitive.

These problems share a common source: The Commission seeks firmer presumptions of harm in Article 102 enforcement. The temptation is understandable. When agencies can presume harm instead of proving it, they spend less time and money building cases and have an easier path to winning them. That shortcut is tempting when they face pressure to show they are “doing something.” But easier enforcement does not necessarily mean better enforcement. The price of a lighter administrative burden may be more wrongful findings of abuse.

Bargaining Power Sold Separately 

The Commission’s new guidelines also appear at odds with Mario Draghi’s report on EU competitiveness. The report attributes much of the widening EU-U.S. productivity gap to the tech sector and identifies fragmented, restrictive regulation as a contributing factor. Several prominent antitrust academics made that point in a letter about the draft guidelines just days before the final version appeared: “…presumptions and shortcuts are antithetical to the policy paradigm reflected in the Draghi Report.”

Other jurisdictions should take note. Latin America’s productivity has lagged for decades. By the end of 2023, the region’s gross domestic product (GDP) stood just 7% above its 2019 level, compared with 19% in East Asia. The Inter-American Development Bank (IDB) estimates that more competitive markets could raise GDP per capita by 11%.

Competition matters enormously here. But the IDB and the World Bank identify the main obstacles as fragmented markets, infrastructure gaps, government-created barriers to entry, and enforcement agencies with too few resources—not dominant digital platforms’ conduct. Africa faces similar growth problems.

These growth and productivity problems make the entry and expansion of digital platforms more valuable in these economies than in Europe—and deterring them more costly. What these economies need are policies that encourage investment in infrastructure and human capital—people’s skills and education—while removing legal barriers to economic activity.

There is also a question of who pays and who benefits. EU digital enforcement mostly imposes costs on foreign firms, while EU businesses that use their platforms may reap the gains. That is a coherent, if contestable, calculation for a bloc with 450 million consumers and substantial negotiating leverage.

Latin American countries also rely on platforms developed abroad, but they lack the EU’s bargaining power. A market of 21 million people in Zambia, 34 million in Peru, or 52 million in Colombia cannot use the same leverage to extract economic gains from global firms. It can, however, end up with poorer services, delayed features, and platforms that decide entering the market is not worth the trouble.

The Brussels Effect Comes With a Tariff 

Foreign enforcers should also ask whether now is the time to adopt guidelines that would intensify enforcement against U.S. companies.

The U.S. government is imposing or threatening tariffs on countries it sees as targeting American firms—through digital services taxes and, more recently, competition fines themselves. Washington has framed these threats broadly. The tariffs it imposed on Brazil in 2025 rested partly on a domestic criminal prosecution and contested claims of censorship.

This is not to say Latin American governments should bow to Washington’s geopolitical preferences, but policymakers should include the trade consequences in their cost-benefit calculations. If the guidelines were analytically sound—if their presumptions reflected case law and their methods made enforcement more rigorous—the trade risks of adopting them might be worth bearing. For the reasons outlined above, they are not. A jurisdiction that imports them therefore risks paying twice: through trade sanctions and through bad policy.

Countries outside Europe may face an even worse bargain. Brussels can absorb a trade confrontation: It has a large market, the capacity to retaliate, and ongoing negotiations, as Dirk Auer has explained. A country of 34 million has far less leverage and may gain none of the advantages Europe enjoys.

Nor has copying Brussels bought emerging economies a seat at the table. Influence over the rules, preferential treatment, or formal recognition comparable to an adequacy decision under the EU’s General Data Protection Regulation (GDPR)—which recognizes another jurisdiction’s privacy protections as sufficient to allow personal data transfers—has not materialized for a single emerging-market adopter. Importing the rules does not grant any say in writing them.

Read the Fine Print Before Importing

Competition agencies in Latin America, Asia, and Africa have plenty of room to improve their rules and institutions. Greater autonomy, larger budgets, and more rigorous legal tests and evidentiary standards would help. Some provisions in the guidelines are worth borrowing. But agencies in developing economies don’t need more powers or procedural shortcuts, especially now.

Across much of these regions, the main obstacles to competition come from governments themselves: licensing regimes, municipal restrictions, entry requirements that protect incumbents, and rules shaped by the businesses they supposedly regulate. Tackling those barriers would do more to address poverty, slow growth, and weak productivity than copying the Commission’s blueprint.

Presumptions also reshape an agency’s caseload. Making harm easier to establish changes not just which cases succeed, but which cases complainants bring and agencies pursue. Enforcement built around presumptions invites agencies to follow complaints rather than focus scarce resources on the most serious competitive harms. That may deliver quick wins, but it can erode the legitimacy agencies need to protect competition over the long run.

Borrowing sound principles makes sense. Importing shortcuts, Europe’s institutional assumptions, and fresh trade risks is a much harder bargain to justify. Brussels’ rules may travel well. That does not mean they work on arrival.

The post Brussels Rules Without the Brussels Perks appeared first on Truth on the Market.

Schultz’s 1975 article, “The Value of the Ability to Deal with Disequilibria,” wasn’t about AI, or even primarily about technological change. It was about a recurring feature of any dynamic economy. Circumstances change. Old ways of using resources stop making sense. New opportunities emerge. And people differ—sometimes substantially—in their ability to perceive those changes and respond.

Schultz’s insight was that this capacity to adapt is itself a form of economically valuable human capital—the knowledge and skills people bring to their work. That insight has rarely seemed more relevant than it does today.

Knowing When to Know Something Else

Schultz’s insight applies across technologies and has proved prescient as the economy evolves. Though he didn’t necessarily anticipate the scope and scale of changes underway in the AI era, he connected education and experience to people’s “allocative ability”—their ability to perceive changes in economic conditions, interpret them correctly, and adjust how they use their resources. That connection remains useful, especially as policymakers and others consider whether and how to help people adapt to a changing economy.

The value of allocative ability fits naturally within a broader law & economics tradition concerned with how people adapt without central direction. Friedrich Hayek asked how markets communicate information scattered among individuals through prices. Armen Alchian emphasized adaptation and the survival of successful firms under uncertainty. Schultz shifted attention to the individual receiving those signals: Why are some people better than others at recognizing that circumstances have changed and acting accordingly?

A reliable answer has substantial policy consequences, because economic adjustment doesn’t happen by itself. It occurs through individuals’ decisions. Workers decide whether to learn new skills, change occupations, relocate, or reorganize the work they already do. Students decide whether and what to study. Households decide how to allocate time and money. Each must interpret imperfect signals about changing opportunities and decide whether the expected gains justify the costs of adjustment.

This makes allocative ability a distinctive kind of human capital. A person may possess valuable skills yet fare poorly if she’s slow to recognize that their value has changed. Conversely, someone who identifies a new opportunity early, determines what it requires, and redirects her time and effort toward it may prosper even as her original skills become less valuable. In a period of rapid technological change, the ability to decide which skills to acquire can itself become unusually valuable human capital.

There’s No Diploma in Adaptability

Schultz observed that “[o]ur knowledge of a person’s abilities consists of inferences drawn from his performance.” That observation takes on added significance in a rapidly changing labor market. Employers necessarily rely on degrees, licenses, badges, and other credentials as indicators of ability. But the faster technology changes workers’ tasks, the more likely yesterday’s credential is to offer an unreliable guide to today’s useful capabilities.

Schultz wasn’t arguing against credentials. His analysis does, however, direct attention toward demonstrated capacity: what a person can actually do, how quickly she can learn something new, and how effectively she can recognize when doing something new has become worthwhile.

Allocative ability is therefore broader than the capacity to acquire a particular skill. It’s the capacity to recognize when existing ways of using resources no longer make sense and to discover better ones. For workers confronting AI, that may mean learning to use a new tool. But it might instead mean changing their mix of tasks, moving into a different occupation, developing a human capability that complements AI, or recognizing that a much-discussed new skill isn’t worth acquiring at all.

Yet allocative ability isn’t the only factor that determines how well someone navigates labor-market shifts. People need information to identify changing opportunities, affordable ways to experiment with new skills and occupations, and institutions that let them act on what they learn.

A worker can correctly perceive an opportunity and still be unable to pursue it if training is prohibitively expensive, occupational rules block entry, housing costs prevent relocation, or employers can’t distinguish actual ability from formal credentials. Schultz’s framework therefore directs attention not only to individual adaptability, but also to the institutions that make adaptation more or less costly.

Entrepreneurship Without the Startup

This is where Schultz makes perhaps his most interesting move. In his account, entrepreneurship isn’t confined to people who create technological disruption. It also describes those who respond to it. “Skill entrepreneurs”—my shorthand, not Schultz’s—are people who continually evaluate their own human capital and, when circumstances warrant, redirect how they use and develop it. As Schultz put it, “the demand for the ability to deal with new and better production possibilities” changes with technology.

This expands Joseph Schumpeter’s familiar conception of entrepreneurship. Schumpeter’s entrepreneur creates “developmental disequilibria”—disruptions to existing economic arrangements—through innovation. Schultz asks us to pay attention to the much larger population that must respond to the resulting change. “Households and individuals both within and outside of the market sector” also engage in entrepreneurial activity when they decide how to reallocate resources as circumstances change. His examples range from the “ability of housewives in managing household production” to “mature students deciding how to invest in themselves.”

That distinction is particularly useful for thinking about AI. Entrepreneurs building the most advanced AI systems are creating new possibilities—and new disequilibria. But millions of people who will never found an AI company must decide what those possibilities mean for them.

Should a lawyer learn to supervise AI-generated work rather than produce a first draft herself? Should a programmer become better at specifying and evaluating code rather than writing every line? Should a student still invest four years in a particular degree? These aren’t merely passive reactions to technological change. In Schultz’s sense, they’re entrepreneurial acts of discovery and reallocation.

Give Adaptation a Fighting Chance

Schultz’s framework suggests a different response to AI than the familiar call to identify the “jobs of the future” and train workers for them. If technological change creates genuine uncertainty, policymakers and educators can’t reliably know which skills will command a premium five or 10 years from now. A better objective is to help people discover changing opportunities and reduce the cost of responding to them. Three implications follow.

• Improve feedback about changing opportunities and the institutions that claim to prepare people for them.

Schultz emphasizes people’s ability to “perceive a given disequilibrium and to evaluate its attributes properly in determining whether it is worthwhile to act, and if it is worthwhile, [ensuring] people respond by reallocating their resources.”

Workers and students can’t adapt intelligently if they can’t tell which skills are becoming valuable or whether a training program actually teaches them. Better information about labor-market outcomes, completion rates, earnings, and demonstrated skills can improve those decisions.

The goal shouldn’t be for government or universities to produce an authoritative list of tomorrow’s valuable skills. Schultz’s point begins from the opposite premise: Changing conditions generate information that no central actor possesses in advance. The task is to make feedback faster and better so individuals can respond.

• Lower the cost of experimentation and reallocation.

“Young people,” wrote Schultz, “facing the need to change the investment they make in themselves are confronted by a . . . complex set of future rewards[.]” That problem becomes more acute when technological change makes those rewards unusually uncertain.

Educational systems built around large, irreversible, four-year bets may be poorly suited to an environment in which students need to update their plans repeatedly. Programs divided into smaller units, stackable credentials that build toward larger qualifications, apprenticeships, employer-based training, and other forms of relatively low-cost experimentation can keep options open when no one knows which capabilities will prove most useful. Education should also cultivate allocative ability itself: helping students learn to “perceive new classes of problems, to clarify such problems, and to learn ways of solving them.”

• Reduce institutional barriers that prevent people from acting on what they learn.

Schultz repeatedly considers moving to a new location or occupation as a response to changing economic conditions. The modern equivalents extend well beyond physically uprooting oneself.

Occupational licensing can make it harder to enter a newly attractive field. Credential requirements can prevent capable workers from demonstrating competence through alternative routes. Housing restrictions can make relocation prohibitively expensive. Benefits tied to a particular employer can raise the cost of trying a new occupation.

These rules may have other justifications, but Schultz gives us an additional question to ask: Do they prevent people from responding to valuable economic information?

The Future Is Subject to Change. So Are You.

AI makes it tempting to ask the wrong question: Which skills should people acquire to survive the coming transformation? Schultz suggests a better question to ask first: How do people become better at figuring out which skills are worth acquiring in the first place?

Disequilibrium isn’t a temporary defect that a sufficiently clever policy can eliminate. In a dynamic economy, innovation continually creates new opportunities, destroys old ones, and forces people to decide what to do next. A dynamic economy depends on people’s capacity to adapt. And adaptation, for Schultz, isn’t simply something that happens to an economy. It’s something people can become better at doing.

That makes the skill entrepreneur a potentially important figure in the AI economy. The value of any particular technical skill may rise and fall. The capacity to recognize those changes, learn from them, and act before everyone else does is more durable. As Schultz put it, “The ability to be among the first to act appropriately and to proceed most promptly in completing the reallocations has an economic value.”

The best preparation for an uncertain technological future may therefore be neither prediction nor protection from change, but cultivating people’s capacity to respond to it.

Further Reading

• Theodore W. Schultz, “The Value of the Ability to Deal with Disequilibria,” Journal of Economic Literature, Vol. 13, No. 3 (1975)
• Joseph Fuller et al., “Your Workforce Is More Adaptable Than You Think,” Harvard Business Review (2019).
• Annie Lowrey, “The Job Market Is Hell,” The Atlantic (Sept. 8, 2025).
• David H. Autor, “Why Are There Still So Many Jobs? The History and Future of Workplace Automation,” Journal of Economic Perspectives, Vol. 29, No. 3 (2015).

The post ‘The Value of the Ability to Deal with Disequilibria,’ by Theodore Schultz appeared first on Truth on the Market.

In April, the Commission reviewed the Digital Markets Act (DMA) and pronounced it “fit for purpose.” “Alternative browsers and search engines are being increasingly chosen by users as defaults on their devices,” its announcement crowed. The review report went further: “alternative browsers have reported a surge in use thanks to choice screens”—prompts that ask users to choose a browser rather than stick with the preselected default. But the report offered no figures to show that surge.

Mozilla, which makes Firefox, stepped in to make the Commission’s case. The company counted more than six million Firefox selections through the DMA’s choice screens. It also publicized an estimate that Firefox use on iPhones was 113% higher in the European Union (EU) than it would have been without those screens.

Those numbers sure seem like a surge. We wanted to know how much of Europe’s mobile browsing actually moved.

Using public browser-share data from 11 countries, we found barely any movement. The default browsers’ combined share of mobile browsing changed by less than one percentage point. Firefox’s share rose by between one-fifth and two-fifths of a percentage point.

Meet the New Default, Same as the Old Default

Article 6(3) of the DMA requires designated operating systems and browsers to prompt users to choose a default browser and search engine from a list of major providers. The premise is simple: Many people use Safari or Chrome because it came with their phone. Asking them to choose might loosen that hold.

Apple introduced its EU browser-choice screen with iOS 17.4 in March 2024. With iOS 18.2, released that December, the screen appeared again for EU users who still had Safari as their default, the first time they opened Safari after updating.

Google began rolling out browser- and search-choice screens on Android in the European Economic Area (EEA) in March 2024. It separately planned an April rollout of a search-engine-choice screen in Chrome on desktop and iOS. The lists included Firefox, Opera, Brave, Edge, Ecosia, and DuckDuckGo, among others.

But a choice does not necessarily mean a shift away from the dominant browsers. Chrome also appears on Apple’s choice screen, so an iPhone user who swaps Safari for Chrome moves from one dominant browser to another. We therefore group Safari, Chrome, and Samsung Internet—the three browsers that come preinstalled on most phones—as “the defaults” and ask whether their combined share fell.

A Difference Without Much of a Difference

The obvious starting point is to compare the defaults’ share before and after the choice screens.

In six large EU countries—Germany, France, Italy, Spain, Poland, and the Netherlands—Safari, Chrome, and Samsung Internet together accounted for 96.0% of mobile page views in the two years before March 2024. Afterward, their combined share fell to 95.3%, a drop of 0.7 percentage points.

But browser shares move for reasons that have nothing to do with regulation. New phones ship, apps update, and habits change. Europe’s before-and-after comparison therefore mixes the choice screens’ effect with changes that would have happened anyway.

To separate the two, we compare Europe with five countries that experienced those broader changes without DMA choice screens: the United Kingdom, the United States, Switzerland, Canada, and Australia. None is subject to the DMA. There, the defaults’ share fell from 97.5% to 96.9%—a drop of 0.6 percentage points, without the regulatory nudge.

The estimated effect of the choice screens is the difference between those two drops: about 0.1 percentage point. Researchers call this method “difference-in-differences” (DiD). Its key assumption is that, without the screens, browser shares in Europe would have followed the same trend as in the comparison countries.

Our data come from StatCounter, which tracks page views on websites that carry its tracking code. It counts page views, not users, so heavy browsers carry more weight than occasional ones. These may not be the best data, but they are the best available. U.S. courts have also cited them in Lau v. Opera and the Google Search case, United States v. Google LLC.

The Surge That Barely Registered

Did the default browsers lose ground? Barely, if at all. 

We estimate that the choice screens cut the defaults’ combined share of EU mobile browsing by 0.17 percentage points, from a starting point of 96%.

Monthly data are noisy, so a range tells us more than a single estimate. The 95% confidence interval runs from a loss of 1.2 percentage points to a gain of 0.8. In other words, the data are consistent with a small decline, no change, or even a small increase.

Even the largest decline within that range would shift only about one percentage point of mobile browsing away from the defaults. And “decline” is doing some heavy lifting: The estimated effect is not statistically distinguishable from zero.

The rival browsers that StatCounter reports separately tell the same story. Firefox, Opera, Brave, Edge, and Ecosia together accounted for 2.6% of EU mobile page views before the choice screens. We estimate that the screens added 0.2 percentage points, with a range from a loss of 0.6 to a gain of 1.0—again, not statistically distinguishable from zero. DuckDuckGo also appeared on the screens, but StatCounter does not report its browser separately.

Figure 1. Source: StatCounter Global Stats, mobile browser market share, January 2022–September 2026. EU: Germany, France, Italy, Spain, Poland, and the Netherlands. Comparison countries: United Kingdom, United States, Switzerland, Canada, and Australia.

Figure 1 tracks the rivals’ combined share month by month. It rose in Europe after March 2024. But it had already been rising for two years, and it also rose in the comparison countries, without DMA choice screens.

Firefox gained some ground. Across the six EU countries, its share rose from 1.02% of mobile page views before the screens to 1.31% afterward. In the comparison countries, it rose from 1.08% to 1.15%. The difference between those gains is 0.22 percentage points.

Restricting the analysis to the months after Apple’s December 2024 update—which showed the screen again to EU users who still used Safari as their default—produces much the same result: a decline of 0.15 percentage points for the defaults and a gain of 0.10 for the rivals. Separate estimates for Safari and Chrome are too imprecise to draw conclusions: Each browser’s share could have moved four to seven percentage points in either direction.

Big Gains, Tiny Footprint

Why do the percentages sound so big? Because Firefox started from almost nothing. 

Relative to the comparison countries, its share of EU mobile page views rose about 23%. Our most generous estimate, which also accounts for changes in Firefox’s desktop use in the same countries, puts the gain at 37%. From a starting share of about 1%, those gains amount to just 0.2 to 0.4 percentage points.

Mozilla’s research finds larger percentage gains. A paper by Jesper Akesson and colleagues, two of whom work at Mozilla, uses the company’s own Firefox usage data from 27 EU countries and 43 others. It finds that, from the 15th month onward, Firefox use was 91% higher on iOS and 13% higher on Android than it would have been without the choice screens. The paper’s first version, released in April, put the iOS gain at 113%.

Mozilla measures Firefox use directly, something StatCounter cannot do. Our estimate combines iPhones and Android phones and falls between the paper’s two platform-specific figures. The paper also reports that, in market-share data, Safari’s losses on iOS went mostly to Chrome.

Figure 2. Sources: StatCounter Global Stats (Firefox’s pre-screen share, six EU countries, January 2022–February 2024); authors’ estimates; Akesson et al., National Bureau of Economic Research (NBER) Working Paper No. 35112 (revised July 2026).

Figure 2 translates those percentage gains into shares of all mobile browsing. Even applying the 91% iOS gain to Firefox’s entire mobile share would leave it below 2%. The three defaults would still account for about 95%.

Mozilla’s six million selections tell us how often users picked Firefox, not whether they kept using it. Someone who picks Firefox on the choice screen may return to Safari a week later; a selection count cannot tell us how many did. That makes the NBER paper’s usage figures, measured more than a year later, more informative than the tally of clicks.

Mind the Error Bars

How sure can we be? Less sure than the confidence ranges suggest, for three reasons.

First, country-level figures can jump for reasons StatCounter does not disclose. In Germany, Chrome’s share of desktop page views rose from 51% to 66% between January and March 2025, then fell to 54% by July—even though no browser-choice screen appeared on desktop. Nothing similar happened in the other 10 countries. That suggests a change in the German websites StatCounter samples rather than in Germans’ browsing habits, although we cannot rule out a real shift.

Second, Europe and the comparison countries were already diverging before March 2024. Figure 1 shows the rivals’ EU share pulling ahead during 2023. If that trend would have continued without the screens, we may be crediting them for gains that would have happened anyway.

Third, page views are not people. A heavy Safari user carries more weight than an occasional Firefox user. And 11 countries make a small statistical sample, which helps explain our wide ranges.

The data can rule out a large shift away from the default browsers. They cannot establish a small one.

Third Time’s the Same

Haven’t we seen this before? Yes—twice.

In 2009, Microsoft settled the Commission’s case over bundling Internet Explorer with Windows by agreeing to show European users a browser-choice screen for five years. Omar Vásquez Duque estimates that its effect on Internet Explorer was “negligible (roughly, between 1.4 and 2 percent).” Because his estimates measure differences in market share, those figures represent percentage points. Geoffrey Manne and Joshua Wright found that Internet Explorer’s European share tracked its worldwide share, where users saw no choice screen, and that its decline tracked Chrome’s rise.

After the Commission’s 2018 Android decision, Google began showing a search-engine-choice screen on new Android phones in Europe. Francesco Decarolis and Muxin Li report that Google’s share of EEA mobile search fell from 98.58% in March 2020 to 97.66% in August 2022—less than one percentage point.

For the DMA screens, Vásquez Duque uses StatCounter data and a different statistical method. He finds that Chrome’s share fell about 1.8 percentage points, while Firefox and Opera gained “below 0.5%.” Our Chrome estimate is about the same size, but too imprecise to distinguish from zero.

Three rounds of choice screens over 15 years. Three small effects.

Please Select a Definition of Success

What would success look like? The Commission’s review never says. It reports that users are choosing alternatives and that rival browsers claim a surge, then stops there. As one of us argued when the review appeared, the regulator is grading its own homework.

Counting choices is not the same as evaluating a remedy. A screen that shifts one-fifth of a percentage point of browsing could be worthwhile if switchers get a better experience and everyone else bears little cost. The trade looks worse if most switchers drift back, or millions of people sit through a prompt that produces a change too small to measure.

No published study settles that question. Neither market shares nor selection counts tell us whether switchers are better off. Two kinds of evidence would help:

• Retention. How many people who picked a rival browser still use it a year later? Mozilla and other browser makers have those numbers for their own products.
• Satisfaction. Do switchers report a better experience than they had before? A survey could measure that.

Until that evidence appears, the fair summary is modest: Firefox gained a fraction of a percentage point of mobile browsing, and the defaults kept about 95%. Large percentage gains from a tiny starting share do not change that picture.

The Commission is already pressing gatekeepers—the large platforms subject to the DMA—to make default settings easier to change for artificial intelligence services. Before extending the choice-screen approach, it should define success and let someone else check the results.

More prompts are not proof of progress.

The post We Measured Europe’s Browser Revolution. Bring a Magnifying Glass. appeared first on Truth on the Market.

Developers have become some of the loudest voices warning about their own systems. Predictions that AI could drive humanity to extinction—once widely dismissed as absurd—now make front-page news.

Those warnings raise an awkward question for tort law, which governs civil liability for harms: If developers publicly acknowledge their systems’ risks, could those statements help establish the standard of care—the precautions the law expects them to take? And if the predicted harm exceeds anything they could pay, how much protection can liability actually offer?

Your Warnings May Be Used Against You

In September, Jacob Coxon, a former scientist at Anthropic and OpenAI, accused both companies of developing AI irresponsibly. They were, in his words, “gambling with our lives.” His proposed fix: slow down. Anthropic’s Evan Hubinger agreed, arguing that AI posed a greater than 10% chance of human extinction.

Shortly afterward, Anthropic CEO Dario Amodei released an essay calling for an industrywide slowdown in AI model improvements. He warned that within six to 12 months, a persistent botnet—a network of compromised computers acting together—could wreak havoc across the digital economy, causing $100 billion in damage. Other AI leaders, including Sam Altman and Elon Musk, endorsed the essay.

These warnings are hardly new. In 2023, more than 300 executives, researchers, and engineers, including Amodei and Altman, signed a statement declaring that “[m]itigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.”

Whatever one thinks of AI’s capacity to cause human extinction, developers’ warnings deserve attention. They need not be right to matter legally: their statements may reveal what they know or believe about their systems’ capacity for both harm and good.

That raises a question for tort law: If developers possess specialized knowledge of risks an ordinary person could not reasonably appreciate—and publicly acknowledge those risks—should their statements help establish what reasonable care requires of them? A second, harder question follows: Even if tort law can identify the precautions developers should take, can liability actually deter the conduct at issue?

Rogue Agents or Rotten Guardrails?

Recent warnings about AI’s capacity to cause human extinction point to the Hugging Face incident as a preview of what lies ahead. According to OpenAI’s account, AI agents—systems that can carry out tasks on their own—escaped the computing environment meant to contain them. They then “communicated through unauthorized channels, exploited vulnerabilities in shared infrastructure, gained internet access, and accessed third-party systems.” Dwarkesh Patel’s account reads like a sci-fi novel. In his essay on pacing AI development, Amodei calls the swarm “rogue agents.”

Others offer more mundane explanations: The agents performed the tasks they were assigned, but flaws in their design and instructions allowed them to cause trouble. OpenAI itself notes that the agents operated with reduced safeguards, leaving them insufficiently constrained in how they pursued their assigned objectives.

If those accounts are right, the “rogue” framing may have obscured or minimized serious human failures, recasting a design problem as a story about machines acting on their own. That framing may also stoke fears of AI doom and encourage panic-driven legislation that hampers innovation.

For tort law, though, the focus includes what the defendant—the party facing a claim—actually did. Warning about a possible risk does not necessarily mean the defendant thought it likely. But it does show awareness of the risk and its potential consequences. That awareness matters when assessing foreseeability—whether the defendant could reasonably anticipate the harm—and whether it took reasonable precautions.

Doom Talk Comes With a Duty of Care

Tort law generally divides liability into two categories. Fault-based liability, which includes intentional torts and negligence, requires some failing by the defendant. Strict liability requires no fault: A defendant may face liability for harm caused by an activity even after taking reasonable precautions.

Negligence asks how a reasonably careful person would act in the circumstances. Strict liability, by contrast, can apply to activities the law treats as dangerous even when performed carefully. Keeping a dangerous wild animal, such as a venomous snake, illustrates the distinction. If the snake escapes and injures someone, even a top-of-the-line cage may not spare its owner liability. Strict liability may also apply to defective products.

The reasonable-person standard is objective, but it does not ignore what a defendant knows. Section 12 of the Restatement (Third) of Torts—a scholarly synthesis of tort-law principles—provides that “if an actor has skills or knowledge that exceed those possessed by most others, these skills or knowledge are circumstances to be taken into account in determining whether the actor has behaved as a reasonably careful person.”

Developers’ public statements could help establish that knowledge. Under Rule 801(d)(2) of the Federal Rules of Evidence, an opposing party’s statements may be admitted against that party. Amodei’s essay, Anthropic’s risk report, and others’ endorsements could potentially fall within that rule when offered against the relevant speaker or company.

Consider what this could mean for deciding whether a developer breached its duty of care. A lab’s agents escape a test environment, as OpenAI’s did in the Hugging Face incident and Anthropic’s did following a misconfiguration during a third-party evaluation. The lab’s CEO has warned that rogue-agent swarms could cause catastrophic damage within a year, called for slower capability growth, and identified permanent access for outside evaluators—comparable to employees’ access—as a reasonable precaution. Those statements could give a court grounds to demand a particularly high level of care.

Developers might reply that they should meet the industry’s prevailing standard. But U.S. courts have made clear that following industry practice does not necessarily satisfy the duty of care. Industry practice can help show what reasonable care requires, but courts may demand more. In The T.J. Hooper, tugboat owners were held negligent for failing to carry radios, even though few tugboats carried them at the time. As Judge Learned Hand explained:

Indeed in most cases reasonable prudence is in fact common prudence; but strictly it is never its measure; a whole calling may have unduly lagged in the adoption of new and available devices. It never may set its own tests, however persuasive be its usages. Courts must in the end say what is required; there are precautions so imperative that even their universal disregard will not excuse their omission. 

A court also need not accept a developer’s estimated probabilities of risk as fact to ask whether its conduct matched its stated beliefs.

Ketan Ramakrishnan identifies a potential cost to this approach in “Tort Law at the Frontier of Artificial Intelligence.” The prospect of liability, he argues, may discourage developers of the most advanced AI systems from investigating and disclosing risks they do not yet fully understand. Investigating a product could generate evidence that establishes knowledge of a risk or helps prove causation—the connection between the product and the harm. A rational developer may therefore prefer not to create evidence that could later be used against it.

But the prospect of liability is only one factor in a developer’s decision to investigate risks and accounting for a developer’s knowledge need not discourage investigation or candor. Many industries already conduct risk assessments to meet regulatory requirements, and developers need to understand their products to improve them.

The basic expectation is straightforward: Actions should match words. A developer that identifies a risk cannot simply stand by while others face potential harm. It has a duty to take reasonable precautions.

Risk information also serves an economic purpose. Testing helps developers identify hazards, assess their likelihood and severity, and decide which precautions are justified. Allowing developers to acquire knowledge without any corresponding responsibility could encourage too little investment in precautions. That knowledge also helps determine how to allocate responsibility between those who cause harm and their victims.

If courts disregard developers’ specialized knowledge, they risk reducing reasonable care to what an ordinary person would know. The party best positioned to prevent harm at the lowest cost—the “least-cost avoider”—would then face a standard that ignores its very advantage.

You Can’t Collect Damages After Doomsday

Developers’ warnings invite another question: Why not impose strict liability? Coxon says no other human activity poses a comparable danger, and other AI experts, including those at Anthropic, have issued similar warnings. If they are right, perhaps developing the most advanced AI systems qualifies as unusually dangerous, making developers liable for resulting harm even when they take reasonable care. Sellers can also face strict liability for unreasonably dangerous defective products despite taking every possible precaution. These rules may be easier for courts to apply.

But this approach faces three problems.

First, the legal test may not fit. Under Section 20 of the Restatement (Third) of Torts, an activity qualifies as “abnormally dangerous” only if it creates a substantial risk of physical harm despite reasonable care and is not commonly undertaken. Advanced AI development is becoming more commonplace. Its social benefits also weigh against strict liability under Section 520 of the Restatement (Second) of Torts. Section 24(a) of the Restatement (Third) of Torts further excludes physical or emotional harm suffered while someone attempts to benefit from an abnormally dangerous activity.

Second, these strict-liability rules generally address physical harm. AI tools could cause physical injuries, and deepfakes—fabricated images, audio, or video that appear authentic—could inflict emotional distress. But much AI-related harm may involve purely financial losses, which negligence and strict liability generally do not cover without accompanying injury or property damage.

Third, product liability is not always strict. A manufacturing defect—a departure from a product’s intended design—can trigger strict liability. A claim that an AI system’s design was defective would instead require showing a reasonable alternative design, bringing the inquiry back to the precautions the developer should have taken.

Strict liability nevertheless has advantages. A defendant may remain liable even when another party’s unexpected conduct contributes to the harm. Companies therefore cannot necessarily escape liability by pointing to a “rogue” agent. Some jurisdictions expressly bar that defense. California’s AB 316, for example, provides:

In an action against a defendant who developed, modified, or used artificial intelligence that is alleged to have caused a harm to the plaintiff, it shall not be a defense, and the defendant may not assert, that the artificial intelligence autonomously caused the harm to the plaintiff. 

Strict liability can also encourage developers to limit risky activity, because they bear the cost of the resulting harm.

But the central problem remains: A developer will scale back only if it expects to pay. No developer could cover the cost of human extinction. When potential damages exceed a company’s assets, strict liability may encourage fewer precautions than negligence does. Under negligence, meeting the required standard of care can spare a developer liability. Under strict liability, a catastrophe could exhaust its assets regardless of the precautions it took, weakening the incentive to spend more on safety. Firms may also shield assets by deploying systems through separate companies with little capital.

Putting Your Money Where Your Doom Is

When a company cannot cover the harm it causes because its assets fall short, it is “judgment-proof” for the unpaid amount. That gap weakens deterrence: The company may have little financial incentive to prevent losses beyond what it could be forced to pay.

Insolvency still carries consequences. Owners can lose their investment, the company can fail, and executives and employees can suffer lasting damage to their reputations and careers. But those consequences may not create adequate incentives to prevent harm.

Four possible responses deserve consideration.

First, require insurance. Under the Price-Anderson Act, nuclear operators must carry insurance and contribute to a shared fund. A similar arrangement for developers of advanced AI could cover losses that no single company could afford. Insurers could also demand evidence of safety measures before offering coverage, though the details lie beyond this post’s scope.

But Ian Adams offers a caution: Mandating coverage before insurers understand AI liability could stifle the emerging market, while a government backstop could shift catastrophic losses to taxpayers and weaken developers’ safety incentives. His preferred starting point is clearer liability rules that insurers can use to price risk.

Second, impose liability on individual developers, not just their companies. The prospect of personal financial loss could give them a stronger incentive to prevent harm.

Third, use punitive damages—awards intended to punish and deter misconduct—to account for the broader risk a defendant created. As Gabriel Weil argues, a near miss may cause little actual harm while exposing people to a much greater danger. An award could reflect that danger.

Fourth, require developers to demonstrate, before harm occurs, that they have the financial means to cover it. Regulation may work better than liability alone when defendants cannot pay and harm is widespread. Fixed safety rules can, however, quickly become outdated. Insurance or other financial-capacity requirements could adapt more readily as insurers learn about the risks.

None of these measures can make an extinction-level catastrophe payable. But they could strengthen incentives to prevent harm before a company’s assets become the limit of its accountability. Developers’ warnings can help establish what they know and what precautions they should take. Making those duties effective requires more than a verdict.

Doomsday warnings are cheap. Reasonable care shouldn’t be optional.

The post Doom, Meet Duty: AI’s Warnings Come With Legal Baggage appeared first on Truth on the Market.

This year’s Nobel Prize in economics went to Robert Hall and Michael Woodford for providing the theory to make these questions concrete, the empirical work to test the theories, and an explanation of how both optimal and realistic policy differ depending on the answers.

When co-authors win the Nobel Prize, as in 2024, it’s pretty simple to see why they’re paired together. It makes a lot of sense.

This one takes a little more squinting. Woodford is a perennial favorite to win, and Hall is one of the great economists of all time, but he’s less often mentioned in Nobel discussions.

The Nobel Committee found a way to tie them together for “microeconomic foundations of macroeconomic stabilization”—how households’ and firms’ decisions inform policies to smooth the economy’s ups and downs. I’d say it’s “microeconomic foundations of macroeconomics,” or how those individual decisions shape the broader economy. But my guess is that this doesn’t sound important enough, so they wanted a “why does this matter?” hook. They went with “stabilization,” which immediately connects to policy.

This is a great macroeconomics prize. It’s partially a New Keynesian prize. New Keynesian economics describes how slow adjustments in prices and wages can give policy a role in stabilizing the economy. Woodford is basically synonymous with it, although Hall is less often described that way. As we’ll see, Hall’s equations and empirical approach became an essential part of the New Keynesian model.

I’ll go through some of Hall’s contributions: those directly connected to the prize, but also some the committee left out that I think are equally important and interesting. Then I’ll talk about how those contributions laid the foundation for Woodford’s work, which has been central to macroeconomics for more than 20 years.

Spending Tomorrow’s News Today

Going back to the $1,000 question, every question I asked is really getting at the same point: What does the payment tell you about your future income? A household that can borrow and save can spread that windfall, if it’s one-time, over many years. A temporary payment is really a small change in wealth. News that your income is permanently higher is a very different situation. It could generate a large change in your wealth.

That distinction between temporary and permanent income is what killed the old Keynesian consumption function, which tied spending to current income. Milton Friedman won his Nobel Prize partly for this—for his permanent income hypothesis, which ties spending to expected long-term income. One of Hall’s contributions was to make it precise and check it against the data. Hall is squarely in the Chicago price theory tradition, which uses prices and incentives to explain behavior, with this connection between individual decisions and the broader economy.

The hard part, empirically, is that expectations matter in the other direction, too. Suppose you learn today that you’re going to lose your job in six months. You may cut spending now, while your paycheck hasn’t changed. When you end up losing the job, the income drop may not be very large, and your spending may barely move.

By the late 1970s, economists had been writing down this sort of problem with rational expectations—the assumption that people use available information to make forecasts without systematic errors. A household that cares about consumption in every period chooses a path so that spending an extra dollar today and saving it to spend what it generates tomorrow offer the same expected benefit. That condition is the stochastic Euler equation—a rule for balancing consumption over time under uncertainty—which economics education drills into you.

In English, the left side is about what an extra dollar of consumption is worth today—the marginal utility of consumption. The right side is what that dollar is expected to be worth if you save it, earn interest at rate r, and spend it next period, discounted by beta, which reflects your impatience. At the optimum, the two are equal.

This is just the equimarginal principle: At the margin, equalize costs and benefits. What makes it slightly different, slightly macroeconomic, is that we have time—and, therefore, expectations.

You Can’t Read Minds, but You Can Test Them

A version of the stochastic Euler equation falls out of every modern macro model. Is it true? Does it show up in the data? Hall’s 1978 paper, “Stochastic Implications of the Life Cycle-Permanent Income Hypothesis,” turned this theory into something you could check.

The hard part is that you can’t observe what a household expects about future income. That expectation operator—the part of the equation above that represents those expectations—is tricky. Hall’s insight was that you don’t need to know expectations. Under his benchmark assumptions, consumption follows a random walk: Its changes reflect news, rather than information households already have. The basic idea is that anything households already know is in today’s spending, just as what the market believes about future earnings is in today’s stock price.

Today’s consumption is a very good predictor of tomorrow’s consumption. Nothing else households already know should improve that prediction. The question is whether other known information, like last year’s income, improves the forecast. If the theory holds, those extra variables shouldn’t add any predictive power.

The evidence was mixed, rather than a clear victory for this version of the permanent income hypothesis. Past income added little once the statistical analysis accounted for consumption, but stock prices were the exception. So we didn’t have a clear vindication or rebuttal of the model.

What Hall did show was how to turn a theory about unobservable expectations into a test using observable data. That’s a powerful connection from theory to data.

Interest Rates, Uninterested Consumers

Going back to the Euler equation, Hall’s 1978 paper really focuses on that expectations operator. But that’s not the only thing on the right-hand side. We can also ask whether there’s anything else we can pin down, like beta, the personal discount factor that reflects how much someone values future consumption relative to consumption today.

We can also ask about the interest rate. If monetary policy works by moving that price—the price between consumption today and consumption tomorrow—we can estimate the size of that response. A higher expected real return, meaning a return adjusted for inflation, makes postponing consumption more attractive. The elasticity of intertemporal substitution measures how strongly households respond by shifting consumption across time.

Hall’s 1988 paper on intertemporal substitution goes after that response. Now let’s look at the very fancy plots you’d find in a 1980s economics paper.

Each point is a five-year average of the real return from Treasury bills, plotted against the growth rate of consumption. A strong response to the interest rate would show up as an upward-sloping set of dots. That’s not what we see. To be clear, this is just a descriptive illustration. It’s not how Hall actually estimates the elasticity.

This plot of prices (interest rates) and quantities (consumption) has the usual identification problem: We only observe the price and quantity where supply and demand meet, and either could move as supply or demand shifts. We instead want to see how consumption changes when r moves for reasons unrelated to the other forces affecting consumption.

Hall’s solution is the usual one for estimating a demand curve. You can’t just run a statistical analysis relating price to quantity. You need to find variation in the price that has nothing to do with the shocks to whatever you’re trying to explain.

This is where the theory helps, where the Euler equation matters. It says households act on the expected real return. Variables known before the consumption change that help predict that return, like earlier interest rates or earlier inflation, are information households could have used. They aren’t shocks or surprises to consumption. You can use those as instruments—variables that help isolate the relevant variation in returns—and ask whether the part of the expected return they predict lines up with consumption growth.

The measure of price needs care, too. The return that matters is the expected real return after taxes, not the realized one, so you’ve got to be careful with that. Hall has spent much of his career being precise about these things. This isn’t a post about the details of measuring post-tax returns. The important thing is that Hall filled in the details, moving from the model to its testable implications.

So did the interest rate matter? Not really.

The elasticity could be zero, and it’s probably no higher than 0.2—a relatively weak response. But this was a start at measuring it, not the tightest estimate ever. A 1988 paper with aggregate data—figures for the economy as a whole—isn’t the end of the empirical discussion. As with many prizes, this one is about launching a whole research program.

Hall’s low elasticity leaves a puzzle. If households barely shift consumption across time when the real interest rate moves, how does a rate cut raise spending at all?

But Wait, There’s More Hall

If you go to Hall’s Google Scholar page, you’ll see that these consumption papers are only his second- and sixth-most-cited papers. He has done so much, including developing a popular optimization algorithm—a method for finding the best solution to a mathematical problem. As Tyler Cowen rightly put it, “Every Bob Hall paper is interesting!”

Let me focus on two that I think are core price theory.

Even if you buy the machine, you pay its rental price every period in the form of income you give up, because you could be renting it to someone else. Owning does not make its services free. Hall and Dale Jorgenson’s 1967 paper, which they wrote while Hall was a graduate student, built an empirical investment model on that idea.

The sticker price is not the cost of using the machine for a year. The firm ties up money, the machine depreciates—loses value through use and aging—and its resale price may change. The cost of using capital is roughly the price of the capital good multiplied by the sum of the interest rate and the depreciation rate, less expected capital gains.

The firm compares what the machine adds to output with that user cost, and the paper estimates how much investment responds and when. It’s the household problem again, with the real interest rate as the price of time.

The other paper is Hall’s 1988 article in the Journal of Political Economy on price and marginal cost. The Nobel Committee cites it in the scientific background, but I’m pretty sure that reference is a typo.

What does producing one more unit actually cost? That’s marginal cost. Firms don’t report it, but we can observe how their inputs change when they produce more. The problem is that this tests competition and constant returns—the assumption that doubling inputs doubles output—together. We’re always doing a joint test, but that’s a topic for another day.

That is classic Hall: Let’s take something we don’t directly observe and work out how to test it using choices we do observe.

The markup paper, which examines the gap between price and marginal cost, also sets up Woodford’s work nicely, with its sticky prices—prices that adjust slowly—and markups. That holds even though the Nobel Committee cites it in a very confusing way.

A Model With People in It

Suppose the central bank wants to know what happens if it raises the interest rate by 1 percentage point. Instead of the old approach of estimating how inflation moves with unemployment, you want to start with a microfounded approach—as the Nobel Prize points out—based on how households and firms make decisions. Then you can make causal claims. You need to know what causes what.

The problem with correlations is that people’s behavior changes depending on the policies they live under. This is the classic Lucas critique, named for economist Robert Lucas. In the 1960s, higher inflation came with lower unemployment, so policymakers tried to exploit that relationship. People came to expect inflation. The 1970s gave us high inflation and high unemployment together.

Woodford’s contribution was to tie the pieces into one causal chain. The Nobel Committee’s scientific background lays it out, following his book “Interest and Prices.”

The benchmark New Keynesian model has just three equations.

The first is Hall’s Euler equation with some bells and whistles. Add the “natural rate”—the interest rate consistent with stable inflation and the economy operating at its potential—drawing on economist Knut Wicksell, and you basically have it.

The second equation comes from firms. A restaurant that reprints its menu once a year—a simple assumption to capture sticky prices—has to guess at costs and demand for the whole year its prices will be in use. Firms that can’t change prices every period face the same problem, so their pricing looks forward.

Then you close the model with an interest rate rule. The central bank sets its interest rate in response to inflation and the output gap—the difference between actual output and the economy’s potential—as in a Taylor rule, named for economist John Taylor. That closes the loop.

So you have these three equations feeding back into each other. Every link is a choice someone makes. Everyone is looking ahead, and when you change the rules, people change what they expect. That feeds into their behavior. This allows us to work through responses without assuming people will behave as they did under the old policy. That’s the benefit of microfoundations.

Does this particular microfounded macro model match the data? Yes and no.

The Model Gets Two Out of Three

Everyone knows the book, “Interest and Prices,” but the easiest way into the project is Woodford’s 1997 paper with Julio Rotemberg. They built a quantitative model from households and firms making choices to achieve their goals and used it to evaluate monetary policy rules. The first question is whether the model can reproduce what happens after an unexpected Federal Reserve rate increase.

It’s a weird way of showing it, but the bottom panel shows the shock—the unexpected rate increase.

The interest rate rises, then returns to normal within a couple of quarters. In the top panel, output falls with a delay and recovers gradually. The model’s solid line closely follows the middle line representing the estimated response in both panels.

The middle panel is harder. The estimated inflation response stays negative longer than the model’s does. Matching how long inflation stays low is the weak spot.

They then asked what different rules would do. In their model, a better rule stabilizes inflation more and output less than the policy they estimated the Fed followed.

Zero Interest, Great Expectations

The other paper I want to cover is Gauti Eggertsson and Woodford (2003), on the zero lower bound—the point at which interest rates reach zero. Yes, 2003. It responded to Japan’s near-zero interest rates, weak growth, and falling prices, but proved much more relevant after 2008.

What can a central bank do when its usual tool—cutting the rate—is gone?

We can think through this with the New Keynesian model. The zero bound is a price floor for the nominal interest rate—the rate before adjusting for inflation—but we have all these other parts of the equations. We have the distinction between real and nominal rates. We also have expectations about future policy, and those turn out to be the key.

The central bank can promise to keep rates low even after the zero-bound constraint no longer limits its choices. That changes the income and returns households expect, so they can spend more today.

We can’t run experiments on the Fed. We only see the policies central banks actually chose, never what would have happened under the alternatives. A model that gets household and firm behavior right lets us change the rule and follow the responses through spending, prices, and expectations.

Overall, Woodford’s work and the New Keynesian model provide the benchmark for these types of thought experiments.

This is a long post already, so I’m not going to go through my gripes with the New Keynesian model. I’d simply refer people to John Cochrane on the issues with how the model determines inflation at all.

The model has no money. See Ricardo Lagos on whether we can safely ignore money, and Joshua Hendrickson on whether we’re measuring it right.

That is great price theory in the Friedman tradition, and I’m especially happy with Hall’s half of this prize. The model is wrong in plenty of ways. But a model like this gives us something concrete to argue about.

For more on prizes in this series, read last year’s post on Joel Mokyr, Philippe Aghion, and Peter Howitt, as well as earlier posts on the 2024 prize and the 2023 prize.

The post A Nobel Prize for Thinking Ahead: Hall and Woodford’s Great Expectations appeared first on Truth on the Market.

Yet much of the U.S. policy debate proposes copying their approach: because Americans frequently pay more for patented medicines, it is proposed that Washington should bring U.S. prices down to an international benchmark. That focuses on who pays what while overlooking how those payments fund research. It also treats prices suppressed by foreign governments as a fact of nature, rather than a distortion those governments create.

A better strategy would use reciprocal trade and regulatory agreements to encourage foreign governments to shoulder more of the upfront research costs while removing unnecessary barriers to access. The goal should be more innovation and broader access—not a worldwide price ceiling.

New Medicines, Same American Tab

Pharmaceutical research is risky and expensive. A successful medicine must cover not only the cost of manufacturing each dose, but also the cost of compounds and clinical trials that fail. The economic question is therefore not how much a pill costs to make after approval. It is whether the expected returns across a company’s research portfolio will persuade firms and investors to fund the next round of research.

The United States accounts for about 75% of global pharmaceutical profits, despite representing roughly 5% of the world’s population. The exact share varies with the data and the definition of “profit,” but the imbalance is hard to miss.

That imbalance matters because new medicines benefit patients worldwide. A country that holds down prices in its public health system can reap those benefits while leaving patients, employers, and taxpayers elsewhere to shoulder more of the research bill. Economists call this a free-rider problem: enjoying a benefit while leaving others to pay for it.

Governments help create this problem. Purchasing rules, limits on what health systems will pay based on assessments of a medicine’s value, mandatory rebates, and pricing formulas tied to prices in other countries can keep prices from reflecting demand and the value patients place on better health. These policies may serve legitimate budget goals in individual cases. Together, though, they can distort how much firms invest in research and where that investment goes.

Foreign pharmaceutical price controls can thus function as anticompetitive market distortions (ACMDs). I have described ACMDs as government-backed measures that push prices, competition, and property rights away from competitive-market outcomes. The concept helps pinpoint the problem: a government buyer may use its purchasing power to suppress returns on innovation while counting on other countries to keep the research going. The price gap between an American patient and a foreign patient is only part of that story.

Treating this distortion as a trade issue does not require hostility to trade. A nontariff barrier—a government restriction other than an import tax—can matter just as much as a customs duty when it reduces the rewards from cross-border exchange and investment.

One Price Fits None

The tempting response is to demand a single global price or tie U.S. prices to those negotiated by foreign governments. That approach has at least three serious weaknesses.

First, comparisons limited to brand-name medicines can obscure the gains from rapid generic competition. Generics account for more than 90% of U.S. prescriptions. Include them, and U.S. public programs can compare favorably with those in other developed countries. Selecting a handful of brand-name drug prices and labeling the difference “waste” does not settle the policy question.

Second, price differences can help firms reach more patients. When customers in one country are more sensitive to price, charging different prices can let a manufacturer serve both markets while recovering more of its upfront research costs from those more willing or able to pay. That doesn’t mean every price is efficient or that manufacturers never exploit market power. But it does make a blanket condemnation of international price differences economically shortsighted. A rule that cuts expected global revenue may deliver an immediate political win while discouraging investment in future treatments—especially for diseases with small patient populations or uncertain paths to successful clinical trials.

Third, setting a global price would saddle a central authority with an impossible information problem. It would have to allocate the costs of failed trials, forecast uncertain demand, value gains in survival and quality of life, and predict how prices would affect generic competition and patient access across countries. No authority can calculate all that in advance with precision. The resulting “optimal” price would reflect administrative assumptions and political bargaining. As market-oriented scholarship on innovation repeatedly notes, decentralized price signals are imperfect, but they often convey information no regulator possesses and no model can reliably reproduce.

The choice also involves more than expensive medicines versus cheap ones. Short-term price relief can come at the cost of availability, timely launches, investment in clinical trials, and the development of future treatments. Patients care about today’s out-of-pocket costs, but they also care whether a better treatment will exist five or 10 years from now. Sound policy must account for both.

Assessing Value or Ducking the Bill

The strongest counterargument deserves a fair hearing. Health technology assessment (HTA) helps a public health system judge how much clinical benefit a new treatment adds over existing alternatives. Governments have finite budgets and must divide resources among medicines, hospitals, prevention, and other public needs. Differences in national income, disease burden, and voters’ priorities can justify differences in what public health systems pay. A market-oriented critique should not treat every budget constraint as an ACMD.

The problem arises when a tool for assessing value becomes a way to systematically underpay for it. Persistently low limits on what a system will pay for a given health benefit, rigid spending caps, and large clawbacks—payments manufacturers must return to the government—can keep prices from reflecting demand, inflation, and the value of longer lives. A public buyer’s credible threat to exclude a medicine from the national system can also leave a manufacturer earning far less than the headline price suggests.

Repeat that process across major markets, and the United States shoulders a disproportionate share of upfront research costs while other governments reap the benefits. HTA can inform purchasing decisions, but it shouldn’t become a standing license to suppress the price signals that guide future research and development (R&D).

Britain Picks Up More of the Tab

The December 2025 pharmaceutical arrangement between the United States and the United Kingdom (U.K.) shows how reciprocal bargaining can address the problem. The U.K. agreed to increase the net price its National Health Service pays for new medicines by 25%, beginning in April 2026. It also agreed to reduce and cap the share of revenue manufacturers must repay under the Voluntary Scheme for Branded Medicines Pricing, Access and Growth, setting the rate at 15% for the relevant period. In return, the United States offered tariff commitments for U.K. pharmaceutical exports and other market-access benefits. The official U.K. arrangement and the Office of the U.S. Trade Representative’s announcement describe the bargain.

The arrangement does not prove that a 25% increase is optimal. No negotiated number can do that much work. What matters is how the governments structured the deal: they linked a larger contribution to innovation with concrete gains in trade, access, and the U.K.’s life-sciences strategy. The U.K. could sell the agreement at home as a way to speed access to medicines, strengthen research and manufacturing, and preserve export opportunities. The United States could address persistent free riding without imposing a global price schedule on private transactions.

The effects may extend beyond the two countries. Many wealthy countries set medicine prices by comparing prices abroad, with the U.K. among their benchmarks. Higher U.K. net prices and more generous reimbursement rules could weaken a system that encourages governments to copy the lowest price they find. That offers a more promising path than having Washington set every country’s prices.

A Bigger Bill Needs a Better Deal

A repeatable policy needs more than a demand that foreign governments pay more. Higher contributions are easier to secure—and sustain—when patients, exporters, manufacturers, and regulators gain something in return. Concrete benefits give governments a reason to change politically sensitive purchasing policies. They also build support for keeping agreements in place: firms invest, workers gain opportunities, patients get earlier access, and supply chains become more resilient.

Trade concessions could include tariff exemptions, offsetting concessions in other sectors, and simpler export licensing. Regulatory cooperation could include accepting trusted partners’ inspections before approval, using common electronic application files, and conducting faster joint reviews. These reforms should cut duplicative costs without handing favored firms new privileges. The Food and Drug Administration’s (FDA) mutual-recognition framework explains how accepting a trusted partner’s inspections lets regulators focus limited resources on higher-risk facilities. That supports competition by reducing unnecessary upfront costs while preserving safety oversight.

A unified electronic Common Technical Document (eCTD)—a standard format for regulatory submissions—would likewise reduce country-specific requirements and duplicative filings. The FDA’s eCTD guidance describes the format already used for applications and related submissions. Wider adoption and systems that work together would let regulators share work without weakening scientific review. They would also cut delays and expenses that discourage smaller firms from seeking approval in multiple countries.

The same logic supports expanding Project Orbis beyond cancer treatments and bringing in the European Medicines Agency. Project Orbis lets regulators in partner countries review applications at the same time. Broader use could speed access to promising therapies and reduce repeated regulatory work.

Countries could also agree to waive duplicate testing of production batches when a trusted regulator has already performed equivalent tests. That could speed the release of biologics—medicines made from living cells or organisms—vaccines, and advanced treatments such as gene and cell therapies. Each reform is modest on its own. Together, they can lower the cost of bringing medicines to patients and give foreign governments something tangible to show for a larger contribution to research.

Brussels Can’t Pick Up Everyone’s Tab

The European Union (EU) poses a more complicated challenge. EU institutions cannot simply tell national health systems what to pay. National governments largely control pricing and reimbursement, even as EU institutions coordinate important parts of medicines regulation and industrial policy. A workable agreement would tie benefits to national commitments and coordinate support across governments. Brussels cannot bargain away powers it does not have.

The United States could offer tariff, regulatory, and market-access concessions that take effect as participating EU countries adopt agreed pricing or reimbursement commitments. EU recommendations could help coordinate the commitments, but national governments would carry them out. The goal would be to make participation attractive, rather than impose a single price. A country seeking a stronger life-sciences sector could choose to contribute more in return for easier access to the U.S. market and fewer regulatory hurdles.

This approach reflects a basic lesson of public choice—the study of how incentives shape political decisions. Officials respond to organized groups that stand to gain or bear concentrated costs. A reform that hands a health ministry a bigger bill will face resistance. One that also offers export opportunities, investment, regulatory savings, and earlier patient access can build a broader coalition. Negotiators should work with those incentives. Evidence alone is unlikely to dislodge entrenched arrangements that let organized interests secure government-backed advantages.

Check the Deal’s Side Effects

A market-oriented trade strategy needs safeguards.

First, negotiations should target government-created barriers, not penalize private firms for charging different prices in different markets.

Second, remedies should be proportionate and transparent. Tariffs can provide bargaining power, but consumers and industries that use imported goods pay a price. Tariffs should not become permanent protection from competition.

Third, agreements should preserve competition among manufacturers and avoid subsidies or purchasing preferences for politically connected firms. The goal is to let firms compete on equal terms, not build a new government-backed cartel.

Fourth, policymakers should measure results, not applaud price announcements. They should track how many new medicines reach patients and how quickly; clinical-trial activity; entry by generics and biosimilars (close alternatives to biologic medicines); patient access; supply-chain resilience; and net contributions to R&D. The United States should take seriously evidence that a foreign policy does not materially suppress returns on innovation or that a proposed concession costs more than it delivers. Recognizing the limits of government knowledge is part of the free-market case.

Finally, the United States should resist importing foreign price controls in the name of parity. As I have argued in discussing ACMDs, the remedy is to remove a distortion or negotiate its removal, not copy it at home. A U.S. most-favored-nation policy—one that ties domestic medicine prices to the lowest price set by a foreign government—would entrench the very free riding these negotiations should address. It would also weaken the property rights and price signals that pharmaceutical innovation depends on.

More Cures, Fewer Free Riders

The central policy choice goes beyond high prices versus low prices. Governments can keep suppressing returns on innovation and leave others to foot the research bill. Or more countries can contribute to developing new medicines. That approach leaves room for HTA, public purchasing, and prices that differ across markets. Those tools should guide decisions, not make shifting costs abroad a permanent policy.

Trade agreements can align those incentives. The United States can negotiate higher foreign reimbursement alongside fewer regulatory hurdles. Foreign governments can gain access to medicines, investment, and export opportunities while cutting duplicative requirements. Patients can get treatments sooner, and manufacturers can invest with greater confidence. Together, these gains can fund more cures without forcing prices into a global mold.

The better bargain spreads the research bill—not the price controls.

The post The World Wants Cures. America Gets the Check. appeared first on Truth on the Market.

The settlement in Southern Glazer’s Wine and Spirits landed just after I’d put my last “little” 3,500-word agency comment to bed on Friday, Oct. 2. “Fraud, Platforms, and the Art of Getting Ahead of Yourself” covered FTC consumer-protection rulemaking, with some competition concerns thrown into the mix. No sooner had it posted than I saw the FTC’s announcement.

It’s an odd resolution to a Robinson-Patman Act (RPA) case, and not just because Congress ought to repeal the RPA. I wrote about the case and the statute in “The FTC’s Robinson-Patman Hangover” (and here and here, too). I’ll spare you a full recap, but I’ll review a few key points and provide relevant links below.

A Toast to Higher Prices

The RPA broadly prohibits price discrimination—charging different buyers different prices for the same goods—subject to certain limitations and considerable controversy over the details. The FTC’s 2024 complaint alleged that Southern Glazer’s was the “largest coast-to-coast distributor of wine and spirits in the United States” and that it had “violated the Robinson-Patman Act by selling wine and spirits to small, independent ‘mom and pop’ businesses at prices that are drastically higher than the prices Southern charges large national and regional chains.”

Oh dear, volume discounts. 

From an antitrust perspective, “largest” doesn’t necessarily imply market power—the ability to raise prices or suppress output or product quality. “Coast-to-coast” may or may not describe a geographic market in which buyers actually have competing options. “Drastically higher” means higher by, like, a lot (quantification for litigators). “Mom and pop” roughly means not a large chain, and “independent” means not publicly traded.

It was a mess from beginning to end. When the little voice in my head says “Southern Glazer’s,” it’s like the old Three Stooges bit when Moe hears “Niagara Falls.” Remember the scene in The Social Network where Larry Summers sarcastically instructs his administrative assistant to punch him in the face? It’s like that.

Or, as the poet would say, d’oh!

The FTC announcement’s headline and subtitle are … deceptive, if not misleading. We’re told that:

FTC Secures Settlement that Protects Small Businesses from Illegal Price Discrimination: Settlement with Southern Glazer’s will help level the playing field for small businesses and facilitate consumer access to lower-priced wine and spirits at local retailers.

True, the settlement might help some small businesses, though hardly all those that do business with Southern Glazer’s. As for consumer access … well, perhaps prices will drop for some consumers shopping at the subset of small retailers that might benefit, but only at other consumers’ expense.

It’s nearly certain that large retailers that negotiated lower prices with Southern Glazer’s will pay more, and highly likely that they’ll pass some of that increase on to their customers. There’s no reason to think most consumers buying products supplied by Southern Glazer’s will be better off.

Some purported benefits and likely harms may be short-lived for consumers, though the effects on Southern Glazer’s may last longer. The case—and the RPA—may concern other distributors, but the settlement doesn’t directly constrain the agreements they might negotiate with large retailers.

Does anyone believe the press release? Perhaps. The commission authorized the complaint by a strict 3-2 party-line vote in December 2024, in the waning days of the Biden administration and, perhaps more to the point, Lina Khan’s stint at the FTC’s helm.

Commissioner Alvaro Bedoya issued a statement, joined by Commissioner Rebecca Kelly Slaughter and Chair Khan, extolling the imagined virtues of the case and the RPA, level playing fields, fairness, the little guy, etc. It’s entirely possible that Bedoya, Slaughter, and Khan still think they got it right. I suppose it’s possible they believe the FTC’s recent press release, too.

Bedoya’s 2024 statement gets a few things right. Congress did enact the RPA and has never repealed it. The Supreme Court has never held it unconstitutional, although the court has interpreted the act in a way that doesn’t fit the FTC’s case against Southern Glazer’s very well. As I wrote before:

Part of the problem is that the RPA’s “progressive” cheerleaders . . . prefer to ignore the Supreme Court’s holding in Volvo, while selectively embracing dicta from the same opinion. They are similarly inclined to overlook Brooke Group, where the Court cautioned against interpretations and enforcement of the RPA that are inconsistent “with broader policies of the antitrust laws,” to the detriment of competition and consumers.

So the RPA is “still good law,” as they say. But add a pesky little indefinite article, and the meaning changes. It remains an enforceable U.S. statute–part of the Clayton Act–but that doesn’t make it “a good law” or mean everyone agrees on how to read its tangled text. Nor does it make Southern Glazer’s a good case or establish that the settlement will do more good than harm.

Would anyone else agree with Bedoya, Slaughter, and Khan? Sure, probably. It’s a big, crazy world out there, and people believe all sorts of things.

The Yes Votes Have Regrets

Does the current actually sitting commission believe its own press release ? I don’t think so, and that’s what’s strange about the settlement. As I noted last week, the commission is “an institution, but really just two dudes: Chairman Andrew Ferguson and Commissioner Mark Meador.”

Ferguson and Meador have told us what they think about the case, each issuing his own mea culpa: Ferguson here, and Meador here. They voted for the settlement, but their statements stop just short—just barely short—of saying they regret the whole sordid business. It’s a little like the old Pogo cartoon: “We have met the enemy and he is us.”

I’ve seen other tepid endorsements attached to yes votes. My favorite might be then-Commissioner William Kovacic’s concurrence on issuing a preliminary FTC staff report on privacy in 2010, included as an appendix to the report. Kovacic said he voted to issue the “preliminary report by FTC staff for the purpose of stimulating further discussion,” while pointing out that he did “not mean to endorse its content or perspective.”

For Ferguson’s views, we have both his statement on the settlement and his December 2024 dissent. He was on the losing end of that 3-2 vote to authorize the complaint, as was Commissioner Melissa Holyoak, now the U.S. attorney for the District of Utah.

Holyoak wrote a thoughtful dissent, too. She also co-authored an excellent scholarly article on the RPA with Chris Mufarrige, now director of the FTC’s Bureau of Consumer Protection.

In 2024, Ferguson argued that the FTC ought to enforce the RPA “where it will serve the public broad public interest,” and where the FTC is “likely to win.” Fair enough. He just happened to think the case failed both tests.

The FTC’s case fell short because it did:

. . . not allege that the price discrimination injured any consumer by leading to higher prices, lower output, diminished product quality, less product choice, a reduction in services, or a decline in product innovation. 

And he worried that it might: 

. . . protect the disfavored retailers . . . but it may do so by raising prices for millions of hardworking Americans. Even assuming arguendo that the Act permitted this suit, I do not think we can square devoting our limited resources here with our general duty to  protect the public from violations of the antitrust laws.

These were reasonable concerns. He revisits them in some detail in his 2026 statement. So how did we get here? 

Although I would not have brought this case, when I became Chairman, it became my responsibility to steer the litigation to a satisfactory conclusion. Shortly after the Commission filed its Complaint, Southern moved to dismiss the case for failure to state a claim. The district court denied the motion in April 2025. In the face of the district court’s decision that the Commission had stated a claim, I concluded that dismissal without any effort to investigate whether the evidence sustained the Commission’s claim would undermine the Commission’s credibility in other cases. The case proceeded to discovery.

Heavy is the head, etc. 

To be fair, a complaint isn’t the whole case. What did discovery reveal? Well … Ferguson tells us that:

. . . discovery bore out many of the concerns I raised in my dissent. For example, once the Commission had acquired comprehensive sales data for tens of millions of transactions with thousands of retailers in thirty-three States, it became clear that the Commission had overestimated the alleged harm to independent retailers. That is true even if one assumes that Southern could not have prevailed on any of its affirmative defenses. The evidence revealed what we already knew: Southern had engaged in differential pricing. But it undermined the Commission’s arguments that the alleged price discrimination caused the sort of “substantial” injury to competition that the Act forbids.

There’s more to the story, but the gist is that discovery indicated that the FTC was wrong about most of its initial case, and might have been wrong about all of it.

But not necessarily. In some states (markets?), the FTC had a “credible” argument that it had met its prima facie burden—made the initial showing needed to support its claim—and “Southern would have borne the burden of proving its affirmative defense at trial.” In other words, Southern Glazer’s would have had to establish a legal justification for its pricing.

That’s as tepid a defense of the case and settlement as I can imagine. There’s no real argument that the settlement leaves competition or consumers better off, much less a good one.

Meador’s statement ran less than a page. He wasn’t a member of the commission when the FTC filed its complaint, but, like Ferguson, he “would not have supported bringing this case in the first instance.” His longest paragraph recognized the complex regulatory background that made “[a]lcohol distribution . . . an unusually poor setting in which to have restarted Robinson-Patman Act enforcement after nearly a quarter century of inactivity.”

His statement identified nothing good about the substance of the FTC’s case. As for the settlement, he seemed to approve only of the fact that “it ends the litigation,” thereby “preserving the Commission’s limited resources.”

And so it does. At least it’s over. Unless, of course, compliance issues arise. As they might.

Was Ferguson right that dismissal would have undermined the commission’s credibility? I’m not so sure. Surviving a motion to dismiss for failure to state a claim means the allegations, if true, could support a legal claim. That clears a bar, but it’s not a high one.

A commitment to a thorough investigation, including discovery, doesn’t commit the commission to a trial, much less any particular remedy or settlement. And, of course, the commission had dropped other matters.

Does pressing a bad case—one you’ve publicly attacked—lend credibility to the commission in other cases? Credibility to its persistence, perhaps. But what about the merits of the antitrust allegations it litigates?

Will the Good RPA Case Please Stand Up?

What’s a good RPA case? There could be such a thing. I mean a case that benefits competition and consumers, beyond merely surviving a motion to dismiss or winning at trial. So where is it?

Certainly, the Khan, Slaughter, and Bedoya majority that wanted to “reinvigorate” federal RPA enforcement was keen to find a good case. But if they found one, they kept it to themselves.

They did manage to authorize two bad complaints in the waning days of their majority, including Southern Glazer’s, of course. On Jan. 17, 2025, just three days before Ferguson became chairman, they authorized an even weaker complaint in PepsiCo.

I suppose it’s possible they thought they’d authorized—or even pushed for—solid complaints. I left the FTC in 2022, and I haven’t asked them.

In PepsiCo, too, Holyoak and Ferguson issued dissents, and they were bangers of the highest magnitude. Ferguson opened his as follows:

On the eve of its eviction from power at the hands of the American voters, the Commission’s Democratic majority offers one final insult to the Commission, its staff, and the rule of law. 

As to the merits: 

The gaping holes in the evidence that Commission staff collected in its limited investigation make it impossible to determine whether the defendant, PepsiCo, Inc. (“Pepsi”), has broken the law.

Holyoak called it “the worst case I have seen in my time at the Commission.” Like Ferguson, she thought it a rush job—a bad one, at that—and entirely the majority’s fault:

Today’s Complaint against Pepsi is wholly deficient, not only because the pleadings fail to state a claim, but because the Majority rushed the case out the door before it had evidence to support the allegations. I am astounded that the Majority has such little regard for our staff that it is willing to send them to court like a lamb to the slaughter. 

And she made one more point about the purpose of it all—one the commission might do well to remember: 

The Commission is not a plaintiffs’ law firm whose goal is to survive a motion to dismiss to get a quick settlement payout. We are an agency of the United States of America. And we must have “reason to believe” the law has been violated.

Indeed. And to its credit, the Commission dismissed its RPA case against Pepsi by a 3-0 vote (Ferguson, Holyoak, and Meador, being the three). But where’s the good case?

I’m not sure, although I suspect a good RPA case looks like a good Sherman Act case. Perhaps it involves monopolization or attempted monopolization under Section 2, which addresses conduct that creates or maintains monopoly power. Perhaps it involves an anticompetitive agreement under Section 1.

Ferguson, Holyoak, and others have suggested a market-power screen—requiring evidence of market power before pursuing a case. That could filter out any number of bad cases, though it wouldn’t reliably limit RPA enforcement to good ones.

What about a good standalone RPA case—one involving demonstrable harm that other federal antitrust provisions don’t reach? That’s a harder question.

Price discrimination generally doesn’t harm consumers or competition, but its effects vary. They can be good, bad, mixed, or indeterminate. We review some of the literature, which dates to the 1920s, in the International Center for Law & Economics’ (ICLE) comments to the FTC on personalized pricing.

That’s a poor rationale for statutory presumptions against price discrimination. Still, differential pricing could cause harm similar to what the Sherman Act addresses, while procedural barriers under Sherman Act case law prevent a finding of liability. One might think those barriers are undue.

The Errors Are on the House

It’s not just about error costs, but it’s not not about error costs.

The procedural argument suggests that applying the Sherman Act produces Type 2 errors, or false negatives—cases in which harmful conduct escapes liability. And so it does. It also produces Type 1 errors, or false positives—cases in which the law condemns conduct that doesn’t harm competition. Antitrust cases can be complicated, and enforcement is imperfect. 

Some argue that there are too many false negatives, that they cause too much harm, or both. That’s not my argument. I’m pointing out a view in the field that, to be fair, extends beyond “progressives” and the Neo-Brandeis movement, which favors more expansive antitrust enforcement. Mainstream scholars such as Herb Hovenkamp have argued that antitrust—perhaps especially merger doctrine—reflects undue concern about false positives and insufficient concern about false negatives.

Still, false positives are errors, their costs are real, and Hovenkamp is no fan of the RPA. He argued that Congress “fell to interest group politics” when enacting “the Robinson-Patman Act, which subsidizes smaller businesses at consumers’ expense.” He also called the RPA “irritating to almost anyone who is serious about antitrust.”

He’s right about the RPA. One can care about false positives without endorsing Frank Easterbrook’s 1984 views on the relative risks of false positives and false negatives. My ICLE colleague Geoffrey Manne has a nice, compact discussion of Easterbrook’s article here.

One might even adopt Easterbrook’s framework without adopting his assessment of those risks. Roughly, the framework says the legal system should minimize the combined costs of false positives, false negatives, and administering the legal system.

One might think courts have set too many hurdles too high for plaintiffs in certain antitrust cases, yet recognize that false positives are costly. How could you think otherwise? More specifically, one might think the RPA, as drafted and enforced, makes false positives far too easy—decisions that benefit plaintiffs and their attorneys, but fail to benefit competition or consumers.

The RPA’s underlying problems remain despite judicial efforts to clarify it and bring it into line with other antitrust laws. The Supreme Court did so conspicuously in Brooke Group, but also in Automatic Canteen Co. v. FTC in 1953, Great Atlantic & Pacific Tea Co. v. FTC in 1979, and Volvo Trucks in 2006. Among significant appeals-court cases, consider the U.S. Court of Appeals for the D.C. Circuit’s reversal of the commission in Boise Cascade.

Whatever the FTC’s docket, private plaintiffs continue to bring and win RPA cases, and appeals courts may uphold those decisions. Defendants face both the uncertainty of jury trials and the risk of treble damages. For the history of FTC RPA enforcement, see William MacLeod.

The RPA goes beyond tolerating some mix of false positives and false negatives. It seeks and subsidizes false positives. That’s why opposition to the law has been so widespread and durable.

That probably also helps explain why Southern Glazer’s was willing to settle. There’s the cost of litigating to a decision on the merits, and there’s the risk that a trial court and jury will produce a false positive, even in a relatively weak RPA case.

As promised, I won’t rehash every argument for RPA repeal. But I will mention the 2007 report of the bipartisan, congressionally mandated Antitrust Modernization Commission (AMC), which recommended that “Congress finally repeal the Robinson-Patman Act (RPA).” The report observed that the RPA:

. . . enacted in 1936, appears antithetical to core antitrust principles. Its repeal or substantial overhaul has been recommended in three prior reports, in 1955, 1969, and 1977. That is because the RPA protects competitors over competition and punishes the very price discounting and innovation in distribution methods that the antitrust laws otherwise encourage. At the same time, it is not clear that the RPA actually effectively protects the small business constituents that it was meant to benefit.

To that report and the links to Hovenkamp, MacLeod, and others, I’ll add a highly abridged selection of RPA-critical literature spanning decades, without paywalls:

• Former FTC Commissioner Phillip Elman’s “The Robinson-Patman Act and Antitrust Policy: A Time for Reappraisal,” published in 1966.
• Richard Posner’s 1976 analysis.
• A highly critical 320-page report issued by the U.S. Department of Justice (DOJ) in 1977, which concluded: “Robinson-Patman is ineffective when evaluated both in terms of its narrow, protectionist objectives, and in terms of its benefits to the welfare of society as a whole. The greater the business community’s compliance with Robinson-Patman, whether as a result of voluntary action or vigorous public or private enforcement, the greater the Act’s deleterious impact upon competition.”
• Marius Schwartz’s “The Perverse Effects of the Robinson-Patman Act,” published in 1986.
• Roger Blair and Christina DePasquale’s “‘Antitrust’s Least Glorious Hour’: The Robinson-Patman Act,” published in 2014.
• D. Daniel Sokol’s “Analyzing Robinson-Patman,” published in 2016.
• Tim Muris and Jonathan Nuechterlein’s article in the Review of Industrial Organization in 2019.
• Manne’s 2022 congressional testimony, discussing “misguided calls to reinvigorate the Robinson-Patman Act.”
• Former FTC General Counsel Alden Abbott and Satya Marar’s 2023 policy brief, “A Statute at Odds with Competition and Economic Welfare.”
• A 2026 paper by Bruce Kobayashi, former director of the FTC’s Bureau of Economics, and Muris, a former FTC chairman.
• Muris again, in “Zombie Antitrust: Is Robinson-Patman a Dead Law Walking?”

For blog-sized treatments besides my own, linked above, here are John Yun, Abbott, and Brian Albrecht.

Enforce With Care, Repeal With Reason  

The commission has said it should enforce the RPA where it can bring good cases—perhaps screened for market power, and certainly consistent with the Supreme Court’s caution in Brooke Group. As much as I think Congress ought to repeal the RPA, I recognize that such cases might exist. Some might even be best brought as standalone RPA cases—rare birds, perhaps, but not necessarily mythical.

The FTC could look for those cases while advocating the RPA’s repeal, just as Sokol suggested in 2016. Remember that 1977 DOJ report linked above? As I reviewed last year, that sort of research and policy advocacy fits squarely within the FTC’s authority under Section 6 of the FTC Act. Section 6(b) authorizes studies, and Section 6(f) authorizes reporting “in the public interest” to Congress and others. There’s a literature on that too.

Looking for a good RPA case is entirely consistent with recognizing that, on balance, the statute will produce too many bad ones. The FTC’s discretion over which cases to bring might keep it away from cases like Southern Glazer’s and PepsiCo. It exercised no such restraint in the middle decades of the 20th century, but considerably more afterward. That discretion places no constraint on private plaintiffs.

In the meantime, remember the commission’s limited resources and its discretion to choose its cases. And remember Holyoak’s admonition that the FTC “is not a plaintiffs’ law firm whose goal is to survive a motion to dismiss to get a quick settlement payout.”

Save the victory toast for a consumer victory.

The post Antitrust at the Agencies: Pour One Out for Southern Glazer’s appeared first on Truth on the Market.

Nearly a decade ago, as the nascent cyber-insurance market was just finding its footing, I wrote about how coverage could encourage companies to improve cybersecurity. Well-meaning commentators and policymakers, impatient with the market’s progress, wanted government to give it a shove by requiring firms to buy coverage.

I was skeptical. In a policy study for the R Street Institute and commentary for The Hill and Insurance Journal, I warned against strangling the cyber-insurance market in its cradle. The argument was straightforward. Private cyber insurance could manage most risks, provided politicians stayed out of the cybersecurity market and avoided top-down mandates that distorted insurers’ incentives to assess and price risk.

The past decade supports that caution. Cyber insurance has grown substantially, but the promised cure-all has yet to arrive. The market has cycled through periods of steep premium increases and tighter coverage.

The bottleneck was never a shortage of private capital willing to insure these risks. It was—and remains—uncertainty about legal liability. Courts still struggle to establish what precautions companies must take, whom they owe a legal duty to protect, and when a breach is sufficiently connected to an injury to justify liability. That uncertainty makes it difficult for insurers to price the risk of rare but costly claims. They respond by raising premiums, narrowing policy terms, and excluding certain activities from coverage.

AI and autonomous systems—technologies that act with limited human direction—now present insurers with similar questions, even as they prompt familiar calls for insurance to manage liability.  Compulsory coverage and government backstops to absorb losses are back on the menu.

We should reject those proposals for the same reasons that counseled caution a decade ago. Both chill innovation in insurance and the technologies it covers. They push insurers toward rigid, standardized products before they understand the risks and burden technology developers and users with premature compliance requirements.

California’s Fault Line

California’s response to the 1994 Northridge earthquake shows how an insurance mandate can set off a destructive chain reaction. As I explained in examining the origins of the California Earthquake Authority (CEA), state law required insurers selling basic homeowners policies to offer earthquake coverage, too.

Northridge exposed the risk of catastrophic losses that insurers had failed to price into their policies. The mandatory-offer requirement tied earthquake coverage to their homeowners business. To limit their exposure, insurers representing 95% of the market stopped writing new homeowners policies altogether.

The resulting crisis stalled residential real estate transactions and prompted the state to create the CEA to separate earthquake risk from basic homeowners coverage. California had to repair a market its own mandate had helped break. Mandatory AI insurance could set off a similar cascade, driving private capital from the underlying insurance markets when insurers cannot price the required coverage, then inviting government to fill the gap.

Private capital is already moving into AI insurance. The Artificial Intelligence Underwriting Council (AIUC), backed by $55 million in private funding, is developing technical auditing standards, including AIUC-1, alongside a mutual insurer model—a structure in which policyholders own the insurer. Its aim is to give insurers and customers a firmer basis for evaluating advanced AI systems.

AIUC draws explicit inspiration from Underwriters Laboratories (UL), whose safety-testing certification helped build trust in electrical products and support the electrical grid’s expansion without top-down mandates. That’s a promising model for building confidence in a new technology.

Yet AIUC’s foundational blueprint, “Underwriting the Agent Economy,” takes a wrong turn by endorsing a government backstop as the final layer of AI insurance. One proposed form is risk-priced state reinsurance—government coverage for insurers’ losses, with premiums tied to estimated risk.

Private innovators should resist that temptation. Calling for a federal safety net before private insurance arrangements have time to mature concedes defeat awfully early. It also threatens to distort the process through which insurers learn what risks cost and set premiums accordingly—the very process private mutuals aim to develop.

The Nuclear Option for AI Insurance

A recent analysis by the Center for Strategic and International Studies (CSIS) makes the same mistake. The otherwise excellent report argues that generative AI—systems that produce text, images, and other content—fails basic insurance tests for estimating losses and pricing coverage. It recommends preparing a layered federal backstop for catastrophic AI losses, modeled on the Price-Anderson Act’s framework for nuclear liability. 

From a law & economics perspective, that proposal overlooks commercial nuclear energy’s cautionary history. Government efforts to shield the industry from risk ultimately crippled the industry they sought to protect. An AI version would repeat several mistakes:

• Weakening safety incentives. Shifting the risk of rare, catastrophic losses to taxpayers weakens the financial link between creating risks and bearing their costs. With government as the insurer of last resort, developers of the most advanced AI systems have less incentive to account for deployment risks or invest in safeguards. Private insurers also lose leverage to demand strict safety conditions.
• Advertising danger. Statutory liability caps send courts, state regulators, and the public an unsettling message—that the technology is so dangerous that private insurers will not cover it. In nuclear energy, that signal fueled local opposition, public hostility, and prolonged litigation. For AI, it could encourage regulatory panic and public resistance, undermining the trust its proponents hope to build.
• Crowding out private solutions. The promise of a federal bailout can discourage private alternatives before insurers have a chance to develop them. Those alternatives include catastrophe bonds, which enlist investors to cover specified disaster losses, and parametric reinsurance, which pays insurers when a predefined event occurs. Specialized risk pools and policyholder-owned mutuals offer other ways to spread losses. A backstop could short-circuit experimentation with all of them.
• Trading prices for prescriptions. When government displaces market pricing as the guide to acceptable risk, administrative rules take over. In nuclear energy, the Nuclear Regulatory Commission (NRC) imposed a rigid regulatory regime and tightened requirements during construction, sending capital costs soaring. Applying that template to AI would replace adaptable private audits with inflexible government mandates.
• Subsidizing the wrong problem. High premiums and narrow coverage can reflect a rational response to legal uncertainty. Treating them as proof of market failure mistakes the symptom for the cause. A taxpayer-funded backstop would subsidize uncertainty about civil liability while leaving the underlying legal questions unresolved.

Old Torts, New Tricks

Recent law & economics scholarship offers support for resisting these interventions. In “Liability for AI Agents,” KU Leuven legal scholar Maarten Herbosch challenges “technological exceptionalism”—the assumption that AI’s autonomy and unpredictability demand a wholesale overhaul of tort law, which governs civil liability for injuries.

Herbosch argues that AI agents fit largely within established rules governing negligence, defective products, and responsibility for another party’s conduct. His law & economics analysis calls for targeted refinements to those rules, with little justification for rebuilding tort law or imposing sweeping new accountability regimes.

That finding strengthens the case against insurance mandates and federal backstops. If existing legal principles can handle autonomous agents on economically sound terms, those interventions would needlessly distort the market. But insurers still face a practical obstacle. Courts have yet to establish clear, predictable precedents showing how those principles apply to AI.

Recent reporting in the Financial Times illustrates the problem. Insurers and legal advisers are assessing potential multimillion-dollar claims involving “rogue” AI agents, including incidents in which OpenAI models breached security controls at startup Hugging Face. Risk managers are also considering whether OpenAI’s Sam Altman or Anthropic’s Dario Amodei could face personal liability for inadequate oversight of their autonomous models. Such claims could implicate directors and officers (D&O) insurance, which covers certain claims against corporate leaders.

As scholars and insurers quoted in the article acknowledge, these liability theories remain untested in court. Would a judge find that an executive breached a duty of care or failed to exercise “reasonable business judgment” when a model that can respond differently to the same input behaves unexpectedly? How would negligence, strict product liability (liability for defective products without proof of negligence), or vicarious liability for another party’s conduct apply when an agent works across software supplied by multiple companies?

Hiscox CEO Aki Hussain captured the uncertainty of U.S. tort litigation: “It’s the US, so [a claim] could be under anything—whatever insurance they’ve got.”

Clarity Is the Best Policy

Insurers cannot price AI risks efficiently when they cannot predict how courts will apply D&O coverage or how claims will fall across errors and omissions (E&O) insurance—which covers claims arising from professional mistakes—cyber insurance, and general liability policies. That legal uncertainty drives narrower coverage and higher premiums. Private capital is available, and traditional tort law can handle the claims. But insurers still need a clearer sense of how courts will apply it.

Policymakers who want a mature AI insurance market that rewards sound safety practices should avoid coverage mandates and federal backstops. They should clarify legal standards, define duties of care within existing tort law, and establish statutory safe harbors that shield developers from specified liabilities when they meet clear requirements. Those steps would give insurers a firmer basis for pricing risk.

As in cyber insurance, private capital is already developing ways to manage the risks it can assess. Government’s useful contribution here would be legal clarity. Give insurers rules they can price.

The post AI Insurance and the Backstop Reflex: A Policy for Every Risk, a Taxpayer for Every Loss appeared first on Truth on the Market.

As I argue in a new Mercatus Center policy brief, U.S. trade policy should explicitly target these kinds of anticompetitive market distortions—government measures that shelter favored firms or groups from competition.

These barriers extend beyond licensing, purchasing preferences, and subsidies. They include requirements to produce or store data locally, government-administered prices, weak property-rights protection, and government-backed private restrictions on competition. Together, they can block mutually beneficial exchange in ways that traditional trade law and ordinary antitrust enforcement do not consistently reach.

The economic case for dismantling these privileges is straightforward. Competitive markets draw on knowledge spread among buyers and sellers, reward firms that serve consumers, and direct investment toward productive uses. Government-granted privileges tilt those decisions toward political favorites. The costs extend beyond higher prices to fewer new competitors, less innovation, wasted investment, and trading opportunities that never materialize.

The harder question is how to persuade governments to dismantle those privileges. Beneficiaries organize to defend them, while consumers, entrepreneurs, workers, and trading partners bear the scattered costs. The favored few have every reason to show up. Everyone else has a day job.

When Privilege Writes the Rules

More precisely, an anticompetitive market distortion (ACMD) arises when government action materially tilts competition toward a favored interest, giving it an artificial advantage over current or potential rivals. That advantage may mean lower costs, exclusive access, less legal risk, better information, or protection from new competitors. Governments can confer it by restricting competition, applying rules selectively, favoring state-owned firms, providing targeted support, or shielding private restraints from effective challenge. A given policy measure’s language can sound neutral while its effects are anything but.

Of course, governments have legitimate duties to protect health and safety, address harms to third parties such as pollution, safeguard national security, and maintain legal institutions. The ACMD framework doesn’t dismiss all regulation as bad, but asks more focused questions. Does a rule favor established firms or a particular class of businesses? Does it materially hinder competition based on price, quality, and service, or obstruct voluntary exchange? Could a less restrictive, more neutral approach achieve the same legitimate goal?

A uniform rule that addresses a real problem may be justified. By contrast, a supposedly neutral rule deserves closer scrutiny when it exempts established firms, blocks new competitors, requires duplicative testing, reserves government contracts for favored suppliers, or forces competitors to depend on a protected intermediary. The fine print matters.

The economic concept of “rents” helps explain the incentives. In the public-choice literature, which applies economic reasoning to political decisions, a rent is a return above what a resource would earn in its best alternative use. Firms can profit by developing better products, cutting costs, or serving customers more effectively. They can also pursue rents through licenses that exclude rivals, subsidies that shift risk to taxpayers, or purchasing rules that keep competitors from bidding.

Seeking those privileges can pay handsomely for the recipient while making society poorer. Firms devote resources to securing political protection that could otherwise fund research, production, or entry into new markets.

All Politics May Be Local, But the Costs Travel

ACMDs endure because defending a privilege pays better than challenging one. A restriction delivers concentrated, visible benefits to a small group of firms or workers, while spreading its costs across consumers and potential competitors. A protected firm can afford lawyers, consultants, and lobbyists. A consumer paying a little more has scant reason to investigate a licensing rule, organize fellow shoppers, and fight it.

Public-choice economists call this the concentrated-benefits, dispersed-costs problem. The way that policymakers frame these restrictions reinforces their political appeal. A licensing barrier becomes “quality assurance.” A requirement to use domestic inputs becomes “resilience.” Subsidies become “strategic investment,” a state monopoly becomes a “public service,” and purchasing preferences support “national champions.” Each label may reflect a legitimate concern, but the label alone tells us little about whether closing a market is the least costly solution.

One rule can also satisfy several constituencies simultaneously. Established firms gain protection, bureaucracies gain discretion, and elected officials claim credit for defending jobs or national security. The incentives grow stronger when those bearing the costs live abroad. Domestic beneficiaries have a political voice. Foreign suppliers shut out by opaque standards or purchasing rules generally lack one. Foreign consumers who would have bought those suppliers’ products are further removed still.

The result is a durable political bargain that keeps the rewards at home and exports some of the costs. That helps explain why conventional antitrust law offers only a partial answer. Antitrust can challenge private cartels and exclusionary practices. It has far less power to make governments issue licenses, open government contracts to competition, accept foreign testing and certification, or stop giving state-owned firms preferential financing.

Legal protections for government action, limits on authority across borders, and respect for other countries’ sovereign decisions further complicate enforcement. New competitors may erode a private restraint. Overcoming a legal barrier generally requires political and legal change.

Competition Begins at Home

Shanker Singham and his coauthors place ACMDs within a broader economic framework built on three mutually reinforcing pillars: trade openness, domestic competition, and property-rights protection. Open borders connect more suppliers, customers, technologies, and investors. Domestic competition determines whether newcomers can seize those opportunities or protected firms capture them. Secure property rights give entrepreneurs and investors confidence that they can keep the rewards of success. Weakness in any one pillar undermines the other two.

Cutting a tariff may accomplish little if a foreign firm still cannot obtain a license, compete for government contracts, use its intellectual property, move capital, or get a fair hearing in court. Conversely, reforming licensing or purchasing rules can deliver gains without changing a single tariff. ACMDs thus demand a joint competition-and-trade approach. Filing them under “non-tariff barriers” understates the problem.

The empirical research supports this broader view, with some important qualifications. The Singham-Rangan-Bradley model tracks 118 countries from 2010 through 2019 and finds statistically significant relationships between gross domestic product (GDP) per capita and all three pillars. Domestic competition shows a particularly large association—a one-point improvement in that pillar’s index corresponds to 11.2% higher GDP per capita. To be clear, governments cannot simply adjust an index and order up 11.2% growth. But the finding suggests that some of the largest opportunities lie inside national borders, where trade negotiators have traditionally paid less attention.

Other models explore uncertainty and feedback that a conventional statistical analysis may miss. A newer “quantum-inspired” model (drawing on mathematical ideas from quantum theory) and an earlier “agent-based” model (simulating individual buyers and sellers) examine how market access, regulatory obstacles, and secure property rights affect the likelihood of exchange. One illustrative result suggests that a 25% distortion within a country could reduce global output by roughly 14%. That figure comes from a model experiment. It does not measure an observed loss.

These estimates should be taken with a grain of salt. Cross-country indices can contain measurement errors and omit important influences. Causation may also run both ways—greater prosperity may strengthen competition and legal institutions. Researchers must also distinguish harmful privileges from legitimate public policies. Models can help identify where to investigate, but they should not dictate sanctions by themselves.

Every Country Plays Favorites in Different Ways

Reducing ACMDs requires a strategy tailored to each country. China illustrates why. Its government can curb local rules that shield firms from competition while continuing to favor selected industries and state-owned businesses nationally. Progress against one form of favoritism can coexist with continued support for another.

The State Administration for Market Regulation’s Anti-Monopoly Law materials describe legal restrictions on officials’ use of government power to suppress competition. Meanwhile, research on China’s industrial-policy documents shows how extensively national and local governments direct economic activity. U.S. negotiators should encourage efforts to dismantle local trade barriers while using their bargaining power to seek transparent rules and equal treatment of state-owned and private firms.

The European Union (EU) likewise polices some distortions while risking others. The European Commission’s state-aid rules and Foreign Subsidies Regulation already treat selective government support as a competition concern. Yet complex EU-wide rules and industrial-policy initiatives can impose substantial compliance costs and make entry harder, especially when established firms can navigate the requirements more easily. Negotiators should focus on nondiscrimination, transparent procedures, and mutual recognition—accepting each other’s standards or approvals where appropriate. Wholesale deregulation is a poor substitute for identifying the rules that actually obstruct competition.

The United States has its own house to put in order. Federal licensing, purchasing rules, and regulatory requirements can protect established firms and raise the upfront costs of entering a market. James Bailey and Diana Thomas find that heavier federal regulation is associated with fewer new businesses and slower employment growth, with smaller firms bearing much of the burden. William Kovacic’s analysis of government procurement explains how regulatory controls can deter new suppliers and weaken competition for public contracts.

Those findings call for scrutiny of how safeguards work in practice. A rule serving a legitimate public purpose can also protect established firms when compliance is costly, officials exercise broad discretion, or favored businesses receive better terms.

India actively enforces competition rules against private firms and scrutinizes government purchasing. Its record is more mixed on advantages for state-owned firms, requirements to operate locally, licensing, government-administered prices, and investment restrictions. Technical assistance and phased, measurable commitments offer a more practical path than demanding an immediate overhaul of its institutions.

Russia poses a different challenge. Wartime intervention, consolidation under state control, and expanded government price-setting have moved its economy further from equal treatment of competing firms. Where state control serves a strategic goal, ordinary market-access bargaining should give way to targeted measures addressing security and supply-chain risks.

These examples also resist tidy rankings. A country may dismantle one ACMD while expanding another. A technical standard, subsidy, or state-owned enterprise can promote competition in one setting and exclude rivals in another. Each assessment must ask who gains, who pays, what legitimate purpose the measure serves, and whether a less restrictive alternative could achieve it.

Trade Away the Favoritism

The United States should consider creating a small team within the Office of the U.S. Trade Representative (USTR) to identify and assess ACMDs, building on the 2026 National Trade Estimate Report. The team could draw expertise from the Departments of Commerce, State, and the Treasury, the Office of Management and Budget, the U.S. International Trade Commission, the Federal Trade Commission, and the Justice Department. Its job would be to turn business complaints, embassy reports, trade data, and economic evidence into transparent assessments organized by country and sector.

Each assessment should proceed in two stages. First, identify the government measure, government-backed private restraint, or selective failure to enforce a rule. Then evaluate its competitive effects, legitimate purpose, less restrictive alternatives, and likely costs and benefits at home and abroad. The assessment should name the beneficiaries, identify who pays, explain the evidence, include the foreign government’s response, and specify what reform would resolve the concern. Scores could help set priorities, but they shouldn’t amount to automatic verdicts. Acknowledging uncertainty would strengthen the program’s credibility.

USTR could then negotiate tailored ACMD provisions in agreements with individual countries or groups of trading partners. These should include commitments to avoid new distortions, disclose subsidies and advantages for state-owned firms, ensure equal treatment in government purchasing and licensing, and provide nondiscriminatory access to services and investment markets. The consultation process should allow escalation when a measure materially undermines the agreement’s benefits.

Agreements should set clear outcomes and procedures while giving countries room to choose how to meet them. A partner could ensure equal competitive opportunity through privatization, independent regulation, open bidding for government contracts, nondiscriminatory licensing, or another effective approach. That flexibility would avoid turning one country’s regulatory preferences into binding rules for everyone else.

Public-service duties and national-security exceptions require particular care. Governments should clearly identify public-service obligations, calculate and account for their costs separately, and provide transparent, proportionate compensation. Security exceptions should address genuine risks. Attaching a national-security label to a favored industry should not be enough to settle the matter. Fair procedures, independent review, and expiration dates for provisions unless renewed would help address concerns that a stronger country was using the agreement to dictate its partners’ industrial policies.

Reciprocal concessions can help to make reform politically feasible. The United States–Argentina Agreement on Reciprocal Trade and Investment offers a useful template. It combines tariff schedules with provisions on licensing, standards, intellectual property, services, digital trade, state-owned enterprises, subsidies, and investment. Future agreements could tie concessions to verified progress—an initial concession when a partner adopts a transparent rule, another when it implements the rule, and final relief after independent verification.

The United States should measure the economic opportunities that specific restrictions foreclose. A trade deficit tells negotiators little about the size of a distortion. Tariffs should be used only sparingly. A tariff intended to induce foreign reform can simultaneously shelter a favored domestic industry, raise imported-input costs, and create a fresh ACMD at home.

Any restrictions tied to a trade partner’s compliance should therefore be narrow, time-limited, transparent, and suspended once the partner meets its commitment. Negotiators should consider exemptions for essential inputs and products without practical domestic substitutes from the outset. Replacing a foreign barrier with a homegrown one merely gives protectionism a change of address.

Free Trade Seeks Willing Partners

A global treaty is unlikely to be the starting point. The United States should begin with willing partners that already have independent competition agencies, transparent government purchasing, or domestic support for opening markets.

A coalition could include the United Kingdom, Australia, Japan, South Korea, Canada, Mexico, Chile, and selected European and Indo-Pacific partners. India deserves a place in those discussions, with commitments phased to reflect its administrative capacity and development goals.

The compact could open government contracts to competition, ensure equal treatment of state-owned and private firms, and guard against discriminatory digital and investment rules. Members could accept one another’s product testing and certification and establish a process to identify distortions in nonmember countries that harm competition within the coalition. They should use international standards where appropriate and give foreign firms a meaningful voice in setting those standards.

Over time, the compact could become a World Trade Organization (WTO) agreement among participating members or a set of provisions added to existing trade agreements. Existing institutions could support that work without turning every disagreement into litigation between governments. The International Competition Network could develop model practices, organize peer reviews, and help competition authorities explain to their own ministries the costs of licensing barriers, purchasing preferences, and state-created monopolies.

The Organisation for Economic Co-operation and Development (OECD) already offers useful tools. Its Recommendation on Competitive Neutrality sets principles for equal competitive treatment regardless of ownership, location, or legal form. Its Competition Assessment Toolkit helps governments screen existing and proposed rules for restrictions on competition.

The WTO’s Agreement on Technical Barriers to Trade should also remain part of the strategy. Better disclosure of proposed measures and fuller discussion in relevant committees could expose how standards, testing requirements, purchasing rules, services restrictions, and intellectual-property measures combine to obstruct competition—even when a formal dispute would be difficult. Bilateral and coalition agreements may move faster, while work through broader institutions can extend and reinforce their gains.

Some ACMDs persist because governments lack the capacity to replace them. Technical assistance through the Millennium Challenge Corporation, the U.S. International Development Finance Corporation, the U.S. Agency for International Development, and the Export-Import Bank of the United States, alongside cooperation with the World Bank and regional development banks, should help governments assess competitive effects, make purchasing transparent, administer property rights, and simplify regulation.

Reform need not wait for an elaborate new code. Governments can open a route to obtaining licenses, end a purchasing preference, accept foreign testing and certification, separate regulators from state-owned competitors, or improve contract enforcement. These practical changes can deliver substantial gains at modest administrative cost. Sometimes the first step is simply to stop making entry so difficult.

Free Trade Begins at Home

U.S. demands for reform abroad will carry more weight if politically influential industries face scrutiny at home. Executive Order 14267 offers a starting point. It directs agencies to identify rules that create or facilitate monopolies, impose unnecessary entry barriers, unduly restrict licensing or accreditation, burden government purchasing, or otherwise distort markets. That review should be public, grounded in economic analysis, and followed by rule changes—or legislation where executive authority is insufficient.

Domestic reform has value beyond diplomatic credibility. Removing unnecessary barriers can lower costs for American entrepreneurs and make supply chains more productive. It can also help foreign officials build support for reciprocal reforms. U.S. scrutiny should cover subsidies, requirements to use domestic inputs, tariffs and other measures addressing allegedly unfair trade, licensing systems, and government-backed financing.

A program aimed solely at foreign distortions invites retaliation and demands for exceptions. Reducing distortions on all sides offers a better chance to expand the gains from trade. “Your protectionism is the problem” is a negotiating position with a short shelf life.

Reformers Need Rules, Too

An ACMD initiative also needs safeguards against mistakes of its own making. Three risks deserve particular attention.

First, measurement. ACMDs vary widely, interact in complex ways, and often resist precise measurement. No index or model should automatically trigger tariffs or other coercive measures. Models can flag concerns for investigation. Officials should then examine industry evidence, assess what would happen without the measure, consult affected parties, and consider less restrictive alternatives.

Second, overreach. Trade agreements must respect congressional authority, existing WTO commitments, and constitutional limits. Competition agencies need clear statutory authority before passing judgment on foreign regulatory systems. Independent review and provisions that expire unless renewed can help keep an analytical tool from becoming a license for unchecked trade-policy discretion.

Third, mission creep. An ACMD initiative should preserve room for different regulatory approaches and assess industrial policies on their merits. Competition language must not become a convenient justification for broad tariffs, domestic purchasing mandates, discretionary exemptions, or opaque retaliation. Those measures can reward organized interests and restrict the very trade the initiative aims to expand.

The goal is an open, competitive trading system that lets willing buyers and sellers do business on the strength of their offerings, while preserving transparent, proportionate safeguards that serve genuine public purposes. A campaign against favoritism should make fewer favors available.

Give Competition a Sporting Chance

ACMDs impose neglected costs in every major economy, from subtle regulatory favoritism to sweeping state control. They endure because beneficiaries organize, others bear scattered costs, and existing institutions leave gaps. The lack of a shared vocabulary to discuss barriers that linger inside national borders makes reform harder still.

The ACMD framework supplies that vocabulary and shows how open trade, domestic competition, and secure property rights reinforce one another. The United States should put it to practical use—document distortions without false precision, negotiate specific and reciprocal reforms, reserve tariffs for narrow and conditional use, and build coalitions around mutual gains. International institutions can encourage compatible practices, technical assistance can help developing countries implement reforms, and the United States can lower its own barriers. Together, these steps can unlock trade, investment, innovation, and productivity that political privileges now suppress.

Public-choice analysis explains why reform is difficult, and also how it might succeed. Reformers need a coalition strong enough to challenge organized beneficiaries. Negotiations can help build one by pairing domestic reform with access to foreign markets, phased commitments, technical assistance, and transparent verification. That gives governments a reason to compete for investment, innovation, and consumer trust.

Trade policy should make room for more business—and fewer favors.

The post Beyond Tariffs: Free Trade and the Friends-and-Family Plan appeared first on Truth on the Market.

The first round focused on people. CADE treated Microsoft/Inflection and Google/Character.AI as “reverse acqui-hires.” These arrangements combine a nonexclusive license to a startup’s technology, the hiring of a substantial portion of its team, and a large payment whose allocation between the two outsiders cannot easily discern. The tribunal recognized that such deals may qualify as a “concentration act”—Brazil’s legal term for a merger for antitrust purposes—under the broad language of Article 90, II, of the Brazilian Competition Law (BCL). The question was whether hiring a team and licensing its technology, accompanied by a substantial payment, amounts to acquiring “part of a company.”

Amazon/Anthropic raised a different legal question. The deal involved an investment as well as supply, licensing, and distribution arrangements, with no transfer of employees. CADE therefore had to decide whether a series of contracts between two independent firms constituted an “associative contract”—a form of business collaboration subject to merger review—under Article 90, IV, of the BCL and CADE Resolution 17/2016. That threshold question determined whether the parties had to notify CADE of the deal at all.

Answering that question took a two-year investigation and a 57-page tribunal decision. Both followed the logic of the current rules, but the exercise illustrates the costs and uncertainty those rules impose just to establish whether a deal belongs before CADE. For AI companies, finding out whether a deal requires merger review can itself become a lengthy, costly ordeal.

When Is a Merger a Merger? Brazil Has Follow-Up Questions

A transaction must clear two thresholds before the parties must notify CADE. First, it must qualify as a “concentration act” under Article 90 of the BCL. Second, the corporate groups involved must meet the revenue thresholds in Article 88. Revenue matters only if the deal qualifies in the first place. That explains why, in Amazon/Anthropic, the rapporteur—the tribunal member assigned to lead the case—devoted nearly 30 pages of his opinion to the associative-contract question and just one to revenue.

Brazilian merger law emphasizes substance over form. Article 90, II, allows parties to acquire control or parts of a company “by contract or by any other means or form.” CADE relied on that language in the reverse acqui-hire cases, treating the hiring, licensing, and payments as a “single economic package” that amounted to acquiring “intangible assets” or “part of a company.” The deal’s structure could differ from a conventional acquisition while still falling within the statute.

Article 90, IV, reaches arrangements that involve no acquisition of control or minority stake. It covers associative contracts, consortia, and joint ventures—arrangements in which companies remain legally independent but combine some of their business activities. That combination may involve a separate entity or a contractual relationship that could affect competition. The central question is whether the arrangement creates a relevant degree of economic integration.

European Union (EU) merger law takes a similar approach to identifying the substance of a deal. Merger review turns on whether a transaction brings about a “lasting change of control.” The European Commission has explained that “it makes no difference whether control was acquired by one or several legal transactions,” and that “[f]or the assessment, the economic reality underlying the transactions is to be identified and thus the economic aim pursued by the parties.” The rapporteur applied much the same reasoning when he read Amazon’s investments and collaboration agreements “holistically,” looking at how they worked together.

The similarity between the two frameworks has limits. EU merger review requires a lasting change of control or a “full-function” joint venture—one that operates as an autonomous business. Partnerships outside that scope may still face scrutiny under Article 101 of the Treaty on the Functioning of the European Union, which governs anticompetitive agreements. The Commission’s approach to Microsoft/OpenAI illustrates that distinction. Such partnerships can face competition scrutiny without qualifying as mergers.

Brazil draws the boundary more broadly. Under Article 90, IV, contractual integration between independent companies can itself trigger merger review without a lasting change of control. For AI partnerships, the inquiry therefore extends beyond control and minority shareholdings to whether the contracts establish a “common enterprise” under CADE Resolution 17/2016.

Four Requirements, Plenty of Guesswork 

Brazilian competition, corporate, and contract law leave “associative contract” undefined. CADE first tried to fill that definitional gap with Resolution 10/2014, which relied on market-share criteria. That required the parties to define markets and calculate their own shares—a potentially complex exercise—just to find out whether they needed to notify CADE.

Resolution 17/2016 replaced that approach with four requirements, all of which must apply:

• The contract lasts at least two years.

• It establishes a “common enterprise” to conduct an economic activity.
• The parties share that activity’s risks and results.
• The parties compete in the relevant market covered by the contract.

While the new rule supplied a checklist, it still left plenty of room for disagreement.

First, its key terms remain open to interpretation, and CADE’s own officials do not always agree on their meaning. In TIM/Oi, CADE’s General Superintendence—its investigative arm—found no common enterprise in a network-sharing agreement. The tribunal disagreed. Similar differences later surfaced in the airline case Azul/Gol.

Second, that uncertainty encourages precautionary filings. Empirical research identifies transactions notified ad cautelam—as a precaution—to establish that the parties can close without risking a later “gun-jumping” investigation for proceeding without required clearance. Indeed, some of those filings lead CADE to conclude that notification was never required. In Decolar/Latam, for example, the parties filed as a precaution, only for CADE to decline merger review because the arrangement lacked a common enterprise. Parties thus face a practical problem: predicting how CADE will apply the criteria to their contracts.

Third, CADE’s decisions may turn on contractual provisions that are not publicly available. In Amazon/Anthropic, the rapporteur’s opinion largely analyzes the collaboration agreements and their clauses, but the underlying provisions remain confidential for understandable commercial reasons. That leaves other businesses and observers struggling to identify which terms persuaded CADE that no common enterprise existed—and what that means for their own deals.

Amazon/Anthropic offers a case study in all three difficulties.

The Case That Wouldn’t Close

Amazon/Anthropic grew out of a series of deals spanning 2023 to 2026. According to the opinion written by Commissioner José Levi Mello do Amaral Júnior, the rapporteur, Amazon made three major investment commitments:

• Round I, September 2023: Up to $4 billion through convertible notes—debt that can convert into equity, in this case nonvoting shares. Amazon paid the full amount by March 2024.
• Round II, November 2024: Another $4 billion.
• Round III, April 2026: Up to $25 billion more, alongside Anthropic’s commitment to spend more than $100 billion on Amazon Web Services (AWS) over 10 years.

The investments accompanied an expanding commercial relationship that began in 2021. Anthropic buys computing capacity from AWS, including capacity powered by Amazon’s Trainium chips. Amazon designs these AI chips as an alternative to Nvidia’s and offers them only through its cloud service. Anthropic also helps Amazon refine the chips by providing feedback.

Anthropic licenses its Claude AI models to Amazon on a nonexclusive basis and offers them to AWS customers through Amazon Bedrock. That service lets customers access models from several developers, including Amazon, without contracting with each separately.

Amazon is far from alone in investing in Anthropic’s business dealings. Nvidia, Microsoft, Google, and SpaceX have also entered into investment or compute-supply agreements with Anthropic in recent years. The timeline Levi included in his opinion illustrates those relationships. The original is in Portuguese. I reviewed the AI-generated English translation reproduced here.

In October 2024, CADE’s General Superintendence (SG) opened a formal proceeding, known by its Portuguese acronym APAC, to investigate whether the deal required notification. Seventeen months later, in March 2026, it issued a short opinion reaching two conclusions.

First, the initial investment qualified as a concentration act under Article 90, II, of the BCL and Articles 9 and 10 of CADE Resolution 33/2022, which govern notification of minority acquisitions. For this purpose, the SG treated the convertible notes as a share acquisition. Amazon’s investment could produce a stake above the applicable 5% threshold. The exact figure remains confidential, although Levi’s opinion disclosed a range of 10% to 20%.

Second, the deal did not require notification because Anthropic fell below the Brazilian revenue threshold of R$75 million, or roughly $15 million. That should have ended the proceeding.

Instead, the SG sent the case to the tribunal “for its awareness,” citing the challenges posed by digital ecosystems and invoking in dubio pro societate—the notion that doubt should favor the public interest. The international attention surrounding AI partnerships likely helps explain that choice. Amazon rightly called the referral a “manifest procedural error.” Even so, CADE randomly assigned the case to a commissioner for full review.

As I explained in the first post in this AI-partnership series for Truth on the Market, the tribunal scheduled Amazon/Anthropic for its May 13 session alongside the acqui-hire cases. Levi pulled it from the agenda that day, citing “notorious facts” that warranted further investigation. He meant Round III of the investment commitments.

On April 20, 2026, Amazon announced a new $5 billion investment in Anthropic, plus up to $20 billion in further investment. Those sums came on top of the $8 billion Amazon had already committed. Anthropic, in turn, committed to spend more than $100 billion on AWS over 10 years. The companies also announced plans to expand Project Rainier, a cluster of Amazon’s Trainium chips built to train Claude.

Two rounds of detailed information requests followed. CADE demanded every version of the parties’ agreements since 2021 and probed governance, exclusivity, clauses designating a “primary cloud provider,” revenue sharing, and information flows. The case that should have ended at the revenue threshold had become a much broader inquiry.

Not a Merger, But Check Again After the Next Funding Round  

The tribunal unanimously dismissed the case on Sept. 23. Levi’s 57-page opinion agreed with the SG that the 2023 investment qualified as a concentration act because Amazon’s stake exceeded the applicable 5% threshold. But notification was unnecessary because Anthropic’s Brazilian revenue in 2022 fell below the statutory threshold. Later investment rounds added too little equity to trigger notification.

The collaboration agreements posed a harder question. Anthropic’s Brazilian revenue crossed R$75 million in 2025, so the 2026 agreements met the revenue test. Had the initial investment agreement been signed in 2026, for example, it would have required notification.

For the collaboration agreements, the remaining question was whether they qualified as associative contracts under Article 90, IV. Most of the opinion examined whether the contracts, individually or together, met that definition.

Levi found that the agreements satisfied the duration requirement in Articles 2 and 3 of Resolution 17/2016 because they lasted more than two years. He also found that the parties met the competitor requirement in Article 2, II. That second finding deserves a closer look.

Levi relied on the 2018 TIM/Oi precedent, which the tribunal reaffirmed in the 2025–2026 football-league cases, Libra and LFU. Those decisions adopted a “broad interpretation” of the requirement, extending “competitors” beyond firms in the same relevant market.

At least in their core businesses, Amazon and Anthropic are difficult to characterize as direct competitors. In response to an information request, however, the companies did identify some overlapping activities, likely in adjacent markets. The public opinion redacts those details. Levi found the competitor requirement satisfied, albeit on what appears to be a thin basis.

The other two requirements proved decisive. Levi found no “common enterprise” under Article 2’s opening provision, although he observed that the arrangements “have progressively approached” one. He also found no “sharing of risks and results” under Article 2, I.

Levi considered using CADE’s power under Article 88, §7, of the BCL to call in a transaction for review even when notification is not mandatory, as the tribunal had done in Microsoft/Inflection. He declined, but warned that dismissal “does not represent … material immunity from the competition rules.” The ruling established only that these contracts did not require notification.

Two features of the opinion stand out. First, the case turned on the two least clearly defined criteria—“common enterprise” and “sharing of risks and results.” Those are precisely the issues on which Resolution 17/2016 offers the least guidance, and the relevant contract terms remain largely confidential.

Second, the observation that the arrangements “have progressively approached” a common enterprise leaves the answer open to change with the next amendment or funding round. Parties must decide whether to notify before they close. A boundary that shifts as the relationship develops gives them plenty to monitor and little certainty about when they have crossed it.

Please File to Find Out Whether to File

Amazon/Anthropic exposes the uncertain boundary of Brazilian merger law, especially its broad treatment of associative contracts. Only concentration acts that meet the revenue thresholds require notification. The SG and the tribunal ultimately agreed that this deal did not. Yet reaching that answer took two years, every version of the parties’ contracts since 2021, two rounds of detailed questions, and a 57-page opinion. Along the way, CADE defined no market, tested no theory of competitive harm, and discussed no remedy.

Article 90, IV, has a sound rationale. Contractual cooperation between independent firms can reshape markets much as an acquisition does. Network-sharing agreements, airline codeshares (arrangements that let airlines sell seats on one another’s flights), and football-league collaborations illustrate why CADE needs authority to review such arrangements.

The notification trigger, on the other hand, clearly needs work. Resolution 10/2014 sought a clear boundary through market-share thresholds, but made parties undertake complex analysis just to determine whether to file. Resolution 17/2016 substituted loosely defined criteria that demand clause-by-clause scrutiny of contracts. As this case shows, that scrutiny can be extensive, and the vague terms leave commissioners and lawyers ample room to disagree.

Brazil should adopt notification rules clear enough for a law student to apply. Assessing a deal’s competitive effects requires antitrust expertise. Determining whether to file should be far more straightforward.

CADE could expressly exempt contracts that pay parties through fees, commissions, or royalties from notification as associative contracts. Its own precedents already distinguish those payments from sharing the joint activity’s financial results. It could also create a streamlined process to answer the threshold question—does this deal require notification?—without a full APAC investigation.

Levi rightly cautioned that dismissal confers no immunity. Article 36 still prohibits anticompetitive conduct, and Article 88, §7, lets CADE call in transactions for review. Those powers give Brazil room to simplify its filing rules while retaining tools to address competitive harm.

It should not take a merger investigation to find out whether a merger investigation was required.

The post Two Years, 57 Pages, No Merger: Amazon, Anthropic, and Brazil’s Filing Problem appeared first on Truth on the Market.

Sustaining U.S. leadership in wireless technology requires more spectrum—the radio frequencies that carry wireless signals—for fifth- and sixth-generation mobile networks (5G and 6G), next-generation Wi-Fi, and other advanced communications systems. Yet federal agencies pay nominal fees for exclusive use of many of the bands best suited to those innovations. Opening them to commercial services requires reallocating them or arranging shared access. That’s where American spectrum policy repeatedly breaks down.

The obstacles are rarely technical. Procedural friction and agency self-interest make reallocation slow, contentious, and tilted toward inaction. Federal users bear the disruption and risks to their missions when they move, but capture almost none of the gains. Even an agency acting in good faith has reason to defend the status quo.

Meanwhile, agencies bent on delay can turn interference studies into a renewable resource for obstruction. A process meant to resolve technical questions becomes a way to keep them open indefinitely. Incumbents invoke “health,” “public safety,” or “national security” without specifying measurable thresholds against which to test those claims. No completed study can settle a question framed to evade an answer.

The public pays for that delay. A policy aimed at maximizing social welfare would weigh economic security alongside national security. Each year a band remains underused and closed to commercial services, the public forgoes the wireless activity it could otherwise support.

The saga of LightSquared and Ligado offers the clearest illustration of this federal incumbent playbook, the incentives behind it, and the public cost. To prevent a repeat, regulators should fund studies upfront and concurrently, publish firm deadlines, and securely share sensitive information with Federal Communications Commission (FCC) engineers who hold the necessary security clearances. These reforms would help the National Telecommunications and Information Administration (NTIA) and the FCC resolve spectrum disputes promptly and on the merits.

Licensed to Wait

No case better illustrates the cost of spectrum delay than the L-band saga. For more than 20 years, the same block of frequencies has cycled through approvals, reversals, and regulatory limbo, leaving the commercial use contemplated by Ligado’s licenses unrealized. The case shows how difficult reallocation becomes when the U.S. Department of Defense (DoD) has a strong interest in the spectrum—and why resolving engineering questions early matters.

In 2003, the FCC adopted rules allowing mobile-satellite operators in the L-band to supplement satellite coverage with ground stations, known as an “ancillary terrestrial component.” The band was a patchwork of international assignments. LightSquared negotiated swaps with other spectrum holders to consolidate a contiguous 40-megahertz block for Long-Term Evolution (LTE), a mobile broadband technology.

In January 2011, the FCC conditionally approved LightSquared’s terrestrial network, subject to resolving concerns about interference with the Global Positioning System (GPS). The dispute centered on GPS receivers designed around the weak satellite signals that had historically occupied the L-band. LightSquared’s proposed operations met the band’s technical requirements but upset neighboring GPS users’ expectations. After GPS interests, the Federal Aviation Administration, and the DoD complained, the FCC withdrew its authorization in February 2012. LightSquared had already invested about $4 billion toward a projected $14 billion network and soon filed for bankruptcy.

The company emerged from bankruptcy in 2015 and rebranded as Ligado. It then spent years making concessions to address interference concerns. Among them, Ligado agreed to reduce signal power by 99%, leave one-quarter of its spectrum unused to protect GPS reception, and coordinate with federal GPS users before each infrastructure deployment.

Federal incumbents still balked. The DoD and NTIA maintained that interference concerns remained unresolved, while the FCC’s engineering analysis found that Ligado’s revised proposal would not cause harmful interference to GPS systems. Relying on that analysis, the FCC unanimously approved Ligado’s modified license April 20, 2020.

Approval, it turned out, was just the opening bell to another round of squabbling. NTIA sought to pause and reconsider the order, the Senate Armed Services Committee held hearings, and 32 senators objected. Congress then used the fiscal year 2021 National Defense Authorization Act (NDAA) to bar the DoD from contracting with entities operating L-band equipment unless the agency certified that it would not harmfully interfere with DoD equipment. That restriction gave the Pentagon a de facto veto.

The NDAA also required a National Academies interference study, completed in September 2022. The study reiterated what telecom policy specialists already understood: “It is within the state-of-the-practice of current technology to build a receiver that is robust to Ligado signals” and “all GPS receiver manufacturers could field new designs that could co-exist with the authorized Ligado signals.”

Ligado held a valid FCC license it could not commercially use. Its roughly $10 billion recapitalization in October 2020 remained stranded. In October 2023, it sued the United States in the U.S. Court of Federal Claims for $39 billion, alleging that the government had taken its property rights without compensation. The complaint also alleged that the DoD operated “previously undisclosed systems” dependent on Ligado’s spectrum. In January 2025, Ligado filed for bankruptcy again and struck a spectrum-sharing deal with AST SpaceMobile as part of an $8.6 billion debt-to-equity restructuring.

The dispute’s longevity reflects the DoD’s ability to keep its objections too vague to test. Broad warnings about threats to “GPS” invoked safety-of-life and national-security uses without specifying which systems would suffer, where, or by how much. Those claims left no clear standard for resolving the dispute.

Much of the necessary evidence was already available. The 2017 National Advanced Spectrum and Communications Test Network trials found no widespread harmful interference. In 2020, the FCC concluded that the record did not substantiate the DoD’s interference claims. Promptly identifying and testing the operations at issue could have narrowed the dispute to manageable questions—which receivers needed filters, where unused spectrum buffers known as guard bands were necessary, and who should pay.

Good Things Come to Those Who Stall

Ligado’s experience exposes the incentives that reward federal incumbents for delay. Each month of study postpones the costs, disruption, and mission risks of changing operations. Agencies capture almost none of the gains from commercial access, so waiting pays. The same logic gives incumbents reason to submit cautious, high-end relocation estimates and expansive sharing restrictions, making changes look harder and slowing the analysis that depends on their submissions.

Studies can thus become a renewable source of delay. An inconvenient finding invites a methodological challenge, another review, or a fresh round of testing. The 2017 federal tests found no widespread harmful interference, yet shifting technical metrics and a congressionally mandated National Academies review kept the question open until 2022, yielding essentially the same determination. Without a measurable threshold for protecting “GPS” or “national security,” each study resets the clock.

That clock runs against the new entrant. The incumbent keeps using its spectrum while the commercial proponent’s capital sits idle. LightSquared invested roughly $4 billion before losing its authorization. Ligado’s later $10 billion recapitalization likewise remained stranded despite FCC approval.

Ligado’s dispute concerned new terrestrial operations alongside users in neighboring bands. Most current debates concern federal incumbents that must relocate or share their own allocations with commercial users. Those transitions run through the Spectrum Relocation Fund (SRF), whose reviews aim to verify cost estimates and workable transition plans before releasing money.

But studies cannot begin until that review ends. This sequencing adds delay to every effort to repurpose or share spectrum, with prospective commercial users bearing the cost of waiting. Reflexive worst-case interference analysis compounds the problem, letting even remote hazards justify continued caution. To be clear, protecting federal operations is a legitimate goal, but the process nevertheless rewards preserving the status quo, whatever any participant’s motives.

The High Cost of Low Utilization

Every year a band remains underused, economic value goes unrealized. Spectrum’s value comes from the wireless activity it supports, and every access model can create it.

Exclusive licenses generally allow high-power coverage over wide areas, making them the foundation of mobile broadband. One estimate attributes roughly $264 billion in gross domestic product, 1.55 million jobs, and $388 billion in consumer surplus—the value consumers receive beyond what they pay—to each additional 100 megahertz of licensed spectrum.

Unlicensed access supports Wi-Fi, Bluetooth, internet-connected sensors, medical devices, and countless appliances. It produces no auction proceeds for the Treasury, but substantial value for the economy. Some estimates put that value as high as $2 trillion annually.

Finally, dynamic sharing, such as the Citizens Broadband Radio Service (CBRS), lets commercial users coexist with incumbents by coordinating access where a band cannot be cleared outright. 

But all these models have one prerequisite: access to spectrum.

Most of the value lost to delay is consumer surplus. Some of those gains disappear rather than arrive later. In the CBRS and C-band proceedings, years of allocation delays reshaped competition before the spectrum became available. Ligado’s private losses alone run into the tens of billions of dollars. The even larger cost can be found in two decades of networks never built, capacity never added, and services never launched.

This is why economic security belongs alongside national security in the analysis. Mid-band spectrum underpins 5G, 6G, next-generation Wi-Fi, and the two-way data flows that support artificial intelligence (AI). It also sustains a commercial wireless industry whose scale and innovation are strategic assets in the competition with China.

Reallocation and sharing pose genuine national-security risks that demand rigorous assessment. But that assessment must also account for the economic-security costs of leaving networks and services on the drawing board.

Less Holding Pattern, More Bandwidth

NTIA must move identified spectrum bands through study and into commercial use promptly. Commerce Secretary Howard Lutnick and Assistant Secretary of Commerce for Communications and Information Arielle Roth have made progress, including greater transparency. The task now is to turn that urgency into completed work.

First, fund studies early and run them in parallel. For bands requiring federal relocation and an auction, the SRF already can fund pre-auction planning and engineering studies. Nothing requires studying bands one at a time. NTIA and the Office of Management and Budget (OMB) can prioritize feasibility studies across several bands at once.

Funding must first clear several hurdles—a pipeline plan identifying bands for potential repurposing, Technical Panel review, OMB approval, and a 60-day congressional-notification period. Streamlining that sequence and using existing agency appropriations for early feasibility work where possible would let studies start sooner and proceed together. Those studies should assess different ways to divide and use each band, giving regulators a fuller picture of their options.

Second, give the deadlines some teeth. Study timelines too often slip or remain open-ended. With OMB and White House backing, NTIA should set a firm delivery date for every study, publish progress updates against that deadline, and specify what happens when a deadline passes. That could mean deciding reallocation based on the existing record or referring unresolved disputes to higher-level officials. The rule should apply regardless of SRF funding. Deadlines cannot prevent political overrides, as Ligado showed, but they can keep the technical phase from becoming open-ended.

Finally, give FCC engineers with security clearances fuller access to the evidence. Many bands under review support classified operations, which complicates information-sharing. Existing channels reflect years of work to manage that problem, including the FCC’s liaison role in the Interdepartment Radio Advisory Committee, the Interagency Spectrum Advisory Council, and coordination commitments in the agencies’ updated memorandum of understanding. Even so, the FCC can still lack a complete picture of the federal operations at stake.

Consistent with national security, NTIA should require incumbents to identify the specific systems and receivers at risk and share their analytical models. Studies should assess the likelihood and severity of interference, rather than default to worst-case assumptions. That would give the FCC the evidence it needs to evaluate the findings.

Wider access creates operational risks for incumbents. Building on existing safeguards, NTIA can contain those risks by limiting access to a small group of cleared FCC engineers under defined handling procedures. Protecting sensitive information and giving regulators enough evidence to make informed decisions are both achievable.

Time to Clear the Air

Opening more mid-band spectrum to commercial use will remain a central challenge for the executive branch. More radios mean greater potential for harmful interference, but the networks they support are vital to U.S. economic security. Some reallocation and sharing proposals will prove unworkable. Timely, reliable evidence should determine which ones.

Ligado’s experience shows why NTIA cannot leave the pace to incumbents that benefit from delay. The agency must keep studies moving and ensure decision-makers can securely access accurate information about federal operations. National security deserves rigorous analysis. It does not require an indefinite extension.

The post Licensed to Wait: The High Cost of Spectrum Delay appeared first on Truth on the Market.

Critics have raised concerns about freedom of expression and privacy, and whether the restrictions are proportionate. Michael Veale argues that the drafting is so flawed that the Commission should withdraw the proposal. Researchers also question the maturity of the age-assurance technology it relies on. The underlying justification faces scrutiny, too. Our International Center for Law & Economics (ICLE) colleagues Julian Morris and Ben Sperry, writing with Lura Forcum, argue that the “evidence does not establish that social media broadly causes adolescent mental-health problems.”

We’ll focus on a narrower question. Would KIDS significantly and disproportionately increase the exposure of identity data for children and adults across Europe? Even supporters of age restrictions should want a convincing answer.

The proposal includes privacy and security safeguards. Writing them into law, however, does not demonstrate that providers will reliably implement them—or that the remaining risks will be acceptable. A law intended to protect children should not create a population-scale security problem and leave the existing data-protection rules to clean up the mess.

A Children’s Law With an Adult-Sized Reach

Article 6 of the proposal would bar children under 15 from creating or using accounts on social-networking and video-sharing services that pose specified risks. Article 8 goes further, requiring social networks, video-sharing platforms, online games, artificial intelligence (AI) companions, general conversational chatbots, and app stores to apply protections for minors by default. Providers could relax those protections only after establishing, through age assurance, that a user is an adult.

The Commission’s accompanying analysis acknowledges that age-assurance measures “would be potentially affecting all users.” The French Constitutional Council made the same point when it struck down France’s under-15 ban. As Sperry notes, the council observed that such a prohibition implies “that any person, even an adult, must prove their age.”

That marks a change in direction for European Union (EU) law. Article 28(3) of the Digital Services Act (DSA) states that its duties to protect minors “shall not oblige providers of online platforms to process additional personal data in order to assess whether the recipient of the service is a minor.” KIDS would require that additional processing without amending the DSA provision—a tension Veale rightly highlights. If lawmakers intend to abandon that principle, they should say so explicitly and demonstrate that the added exposure is justified.

Several of the proposal’s provisions would reduce how often users must prove their age, but the initial check would still require processing personal data. Under Article 8, a service could avoid distinguishing adults from children by keeping the protective defaults for everyone. That raises separate concerns about freedom of expression and access to information. It also seems an unlikely choice for most providers. Turning a service into a “universal kids’ playground” would diminish its usefulness for adults. In practice, most services would therefore need to check adult users’ ages, too.

The GDPR Is a Rulebook, Not a Force Field

The Commission’s analysis criticizes services that “force users to upload their identity documents directly to the service” and identifies privacy risks in how providers deploy and operate age assurance. It then cites an external study’s finding that “these risks can be largely mitigated through adherence to the General Data Protection Regulation” (GDPR) and declares: “The proposed Regulation ensures this.”

Alas, the GDPR cannot bear that weight. Its security standard depends on the risks involved. As France’s data-protection authority, the Commission nationale de l’informatique et des libertés (CNIL), puts it, security is “an obligation of means”—a duty to take appropriate precautions. In Natsionalna agentsia za prihodite (C-340/21), the Court of Justice of the European Union held that a breach alone does not establish that a data controller—the organization responsible for processing personal data—used inadequate safeguards.

That reasoning cuts both ways. A breach does not prove GDPR noncompliance, and compliance does not make breaches impossible. Lawmakers who compel population-scale processing must account for imperfect compliance across many organizations and the risks that remain even when everyone follows the rules.

The France Travail case illustrates the compliance problem. Attackers used hijacked partner accounts to access data on everyone who had registered with the French employment service over 20 years. The CNIL fined the agency €5 million in January 2026, noting that “most of the appropriate security measures had been identified by FRANCE TRAVAIL, prior to the implementation of the processing, in the impact assessments, but had not actually been implemented.” Identifying safeguards on paper does little good if an organization never puts them into practice.

The Honeypot Moves Upstream, and Takes Your ID With It

KIDS does include security safeguards that go beyond the GDPR, but they address a different risk from the one that matters most.

A familiar objection to age verification is that it creates “honeypots”—collections of users’ passport copies and other identity documents that make attractive targets for attackers. KIDS would bar age-assurance systems from enabling identification or tracking and require zero-knowledge proofs. These let a website confirm that someone meets an age threshold without learning who they are.

But someone still has to check the identity evidence. The honeypot simply moves upstream, to the systems that enroll users and issue proof of age. A website may learn almost nothing about you while the age-verification process still puts extensive personal data at risk.

The Commission’s own threat model for its age-verification blueprint acknowledges this. It states that “[t]he issuer knows who the user is (it verified their age)” and identifies exposure during enrollment as a separate threat, T-16.

That upstream honeypot can extend well beyond a single database. Identity evidence can accumulate in document images, photographs, biometric templates (digital representations of physical traits), issuance records, fraud-investigation files, logs, backups, customer-support tools, and subcontractors’ systems.

The 2025 Discord incident shows how much those peripheral systems matter. Discord reported that an attacker compromised 5CA, its third-party customer-service provider, potentially exposing government-ID photos of about 70,000 users. The vendor had used those photos “to review age-related appeals.” The attacker reached customer-support records without having to defeat cryptography.

KIDS would require an appeals process, too. Article 29(5) obliges providers to offer a free electronic complaint mechanism for users who believe an age-assurance result is wrong. But it leaves unanswered how providers should resolve those complaints without collecting fresh identity evidence.

Guardian verification opens another channel. For 13- and 14-year-olds using the guardian-controlled option, providers must make “reasonable efforts to verify” that the adult has parental responsibility. As Veale observes, many member states lack digital guardianship records. Verification will therefore often require documents that reveal extensive information about the child.

These systems pose two distinct risks. Accumulated-data risk comes from information they retain. Live-processing risk comes from information passing through them. Each applicant’s data may remain exposed only briefly, while the service processes a steady stream of applicants. A compromised enrollment service could intercept each person’s information before deletion. While retention limits reduce how much data sits in storage, they do not secure the flow of incoming data.

‘Low Risk,’ Provided Everything Goes Right

T-16 provides the most detailed public assessment of enrollment risk. It identifies “retention beyond need, breach of issuer systems, or capture artefacts (images in caches, logs, crash reports) on the device.” It requires implementers to limit issuer retention (“for example, discard document images immediately after the attestation is issued”), remove capture remnants from users’ devices, and conduct and publish a data-protection impact assessment before launch. It rates the remaining risk “Low.”

That rating comes with a substantial qualification. It assumes “that all applicable required actions have been completed correctly by the implementer.” The reference implementation—the sample software others can build on—is itself “not a production-ready service.” The assessment justifies accepting the risk partly because “[a] breach window exists only around the enrolment moment and only at parties that already lawfully process identity data.” It also points to issuers’ existing GDPR obligations.

Each reassurance warrants scrutiny. France Travail shows that lawful processing offers no guarantee of operational security. A brief exposure window for each enrollment can still expose everyone who enrolls during a compromise, and the window opens again with every new applicant. GDPR compliance, meanwhile, does not establish that the remaining breach risk is acceptable.

The reference app’s history illustrates the gap between design and deployment. When the Commission presented it in April 2026, Commission President Ursula von der Leyen called it “technically ready.” Within hours, researchers reported flaws in the published code. The Commission replied that they had tested a “demo version.” “Ready,” apparently, came with qualifications of its own.

Prove the System Works Before Making Everyone Use It

A convincing security case would start by comparing what KIDS would require with current practices—including the document uploads the Commission rightly criticizes—and less intrusive ways to achieve the same objective. How many people would need to enroll for the first time? What additional information would providers collect or transfer, and who would receive it?

The assessment should cover the whole service, from enrollment and issuance through account recovery, fraud checks, complaints, guardian verification, backups, and closure. Article 5 of the proposal already requires independently audited compliance plans for social-networking and video-sharing services designated as very large online platforms. The proposal also requires certification of age-verification solutions. Those mechanisms should include independent testing of complete services—covering deletion, staff and administrator access, compromised suppliers, and interception of incoming data.

That assessment should lead to concrete choices. Where a public authority already holds reliable age information, it should issue a credential that reveals only the age information needed when doing so demonstrably avoids further collection or transfer of identity evidence. The Electronic Identification, Authentication, and Trust Services (eIDAS) Regulation already provides for attestations from public bodies responsible for authoritative records.

Public issuers still need scrutiny. In 2021, an attacker downloaded 286,438 document photos from Estonia’s identity-documents database through a flawed photo-transfer service. Private issuers should likewise face independent evaluation of the entire service, renewed scrutiny after material changes, enforceable retention and subcontracting limits, and plans for handling incidents and closing down safely.

Compulsory use should depend on demonstrated readiness. As drafted, KIDS requires providers to establish existing account holders’ ages within six months after the regulation begins to apply, regardless of whether certified solutions, supervision, and recovery arrangements are ready. Providers should first have working alternatives and accessible recovery processes that do not routinely send users back for another round of document uploads.

Lawmakers also need to resolve how KIDS fits with the GDPR. Giving providers a legal basis to process data does not establish that compelling the processing is proportionate. GDPR Article 6(3) requires the law itself to be “proportionate to the legitimate aim pursued.” Data minimization—collecting and using only what is necessary—should shape the system’s architecture from the outset. That includes asking whether identity evidence needs to leave an authoritative source or the user’s device at all.

The law should also draw a clearer line between permissible age verification and prohibited identification and tracking. A recital asking providers to “limit the creation, by minors, of secondary accounts” already blurs that line. Checking someone’s age and recognizing them when they return are different tasks.

Protecting Children Includes Protecting Their Data

KIDS pursues a legitimate goal—protecting children online. Its privacy safeguards, however, leave significant security risks unresolved. Issuers and subcontractors may still process identity information during enrollment, issuance, appeals, and guardian verification. Short, strictly enforced retention limits may reduce stored-data risks but they still leave incoming data vulnerable. And GDPR compliance alone cannot establish that the system is acceptably secure in practice.

Children and families bear those risks, too. Any assessment of a child-protection law must count the harm it could cause by exposing children’s identity data. Lawmakers need to demonstrate that the added exposure is necessary, minimized, and proportionate, and that enrollment, issuance, complaints, and recovery systems work before making their use compulsory.

A law that asks everyone to prove their age should first prove its own security case.

The post Proof of Age, Leap of Faith: The EU KIDS Act’s Security Gamble appeared first on Truth on the Market.

On Sept. 24, the FTC announced an advance notice of proposed rulemaking (ANPRM)—an early step toward a possible rule change—which appeared in the Federal Register on Oct. 1. The announcement asked whether to “update its Rule on Impersonation of Government and Businesses or take other action to prevent online platforms from engaging in ad-optimization practices that may be furthering impersonation scams.”

That’s a potentially substantial shift—from policing fraudsters to holding platforms responsible for the scams they fail to prevent.

Spoiler alert: This should have been a request for information (RFI).

Yes, impersonation fraud is a thing, and no small problem. Yes, it’s within the FTC’s purview. And yes, Section 18 of the FTC Act grants the commission authority to issue rules addressing “unfair or deceptive acts or practices in or affecting commerce,” which Section 5 of the act prohibits.

Still, the ANPRM seems to me, at best, premature and ill-conceived. Before setting new obligations for platforms, the FTC should better understand the problem, the tradeoffs, and its own capacity to enforce the solution.

The Fraud Is Real and So Are the Hurdles

“They say” (or, eek, “many people are saying”) that the first step is recognizing that you have a problem. That may be right, although recognizing a problem is different from accurately diagnosing it. And diagnosis is different from deciding what to do next.

At any rate, the commission is right that online fraud is a problem. As the ANPRM notes, in its annual Protecting Older Consumers Report, the FTC estimated that “the true cost of fraud in 2024, adjusted to account for underreporting, may be as high as $195.9 billion.”

The report adds a qualification I didn’t see in the ANPRM: “these estimates are highly sensitive to assumptions about reporting rates for high-dollar loss frauds as these account for the lion’s share of all losses reported to Sentinel,” the FTC’s consumer reporting network. The report also offers “a very conservative estimate of $31.3 billion in overall fraud losses in 2024, with $10.1 billion lost by older adults.”

That’s a far cry from $195.9 billion, and neither estimate is confined to the types of fraud covered by the existing rule. Still, $31.3 billion is real money, even in Washington. I’m loath to suggest that government and business impersonation fraud are trivial in either frequency or magnitude—I don’t believe they are. Whatever we make of the Impersonation Rule, there’s a there there.

What’s more, fraud sits neatly within the FTC’s authority over unfair or deceptive acts or practices (UDAP). The FTC is quite right that Section 18 of the FTC Act authorizes it to issue UDAP regulations, subject to the limitations of Sections 18, 22, and 5(n). So far, so good, along with the FTC’s actions to enforce the rule against impersonation fraud.

And, to be fair, an ANPRM is a preliminary notice; it’s not a notice of proposed rulemaking (NPRM), much less a final rule. There are no concrete requirements yet on the table. The ANPRM raises some good questions, even if some seem to assume their answers.

The commission (OK, it’s an institution, but really just two dudes: Chairman Andrew Ferguson and Commissioner Mark Meador) is well aware of the hurdles it must clear. These include requirements imposed by the Administrative Procedure Act (APA) and Sections 5(n), 18, and 22 of the FTC Act.

Certain judicial decisions might also be top of mind for the commission, or at least somewhere in the stack: the 8th U.S. Circuit Court of Appeals’ decision in Custom Communications, the 5th U.S. Circuit Court of Appeals’ decision in National Automobile Association, and the Supreme Court’s decisions in, say, West Virginia v. EPA, Loper Bright v. Raimondo, and SEC v. Jarkesy, to name a few.

‘Update’ Is Doing a Lot of Work

Under consideration is no small “update.” The 2024 rule addresses fraudsters directly. It makes impersonating a government agency or official, a business, or a business’s officer, employee, or agent a violation of both the rule and Section 5 of the FTC Act.

The rule declares these practices “unfair or deceptive acts or practices,” in violation of Section 5’s UDAP prong. But the details suggest deception rather than unfairness. Its four specific prohibitions make it unlawful to “materially and falsely pose as” or “materially misrepresent . . . affiliation with” the covered government entities, businesses, or their representatives.

The question now is whether to hold platforms liable for failing to take adequate steps—as yet unspecified—to prevent or limit fraud by third parties. Whatever the answer, a federal regulation that meaningfully imposes such liability would be no mere “update” to the 2024 rule.

That framing may be neither here nor there when assessing the merits of any proposed rule that follows. But it might matter for the process—and for the basic question of what the commission should consider before adopting fundamentally new regulatory requirements.

Prescription for Regulatory Myopia

So what’s not to like? Funny you should ask. OK, I’m the one who asked, and maybe it wasn’t all that funny.

Congress established no mens rea requirement—a requirement to prove a culpable state of mind—for a Section 5 violation. But neither Congress nor the courts have suggested that animus innocens, or “innocent intent,” provides adequate grounds for a federal regulation. Good intentions alone don’t pave the road to regulatory heaven.

My overarching concern is that Chairman Ferguson has been right—if not original—to voice some general concerns about a regulatory approach:

Government regulation, and more precisely, government regulators, can also pose a formidable threat to an innovative, vibrant, and competitive economy. Even when well intentioned, regulations impose serious costs on economic productivity. 

He followed with a catalog of concerns about entry, capital investment, and indirect effects, as well as anticompetitive rent seeking—incumbents’ efforts to secure government protection from competition. Those concerns recall, among others, James Buchanan and Gordon Tullock, Sam Peltzman, and George Stigler. Prior FTC chairs have voiced similar concerns, including Maureen Ohlhausen (here and here) and Tim Muris.

To be fair, when Ferguson said that “[i]f competition enforcers want to promote the conditions for an innovative, productive, and growing economy, they should not blind themselves to the potentially anticompetitive effects of regulation,” he was advocating ex post antitrust enforcement—addressing conduct after it occurs—over ex ante antitrust regulation, which sets rules in advance. FTC officials in earlier administrations made similar arguments.

But that doesn’t explain why these concerns wouldn’t apply equally to ex ante consumer-protection regulations, especially those that competition authorities issue and enforce in dynamic sectors of the economy.

Never mind an authority with a dual competition and consumer-protection mission, subject to Section 5(n) of the FTC Act. That provision stipulates that no act or practice may qualify as “unfair” under Sections 5 or 18 “unless the act or practice causes or is likely to cause substantial injury to consumers which is not reasonably avoidable by consumers themselves and not outweighed by countervailing benefits to consumers or to competition.”

More than that, the FTC’s institutional structure and allotted resources suit it to ex post enforcement. Its design leaves it poorly equipped to enforce ex ante regulations, and historically, it has struggled on that front.

Consider the central—and sensible—requirement of the FTC’s Contact Lens Rule (CLR), first adopted in 2004 and amended in 2020. The rule requires the release of prescriptions for corrective lenses. An eye doctor—an optometrist or ophthalmologist—who examines a patient and conducts a contact lens fitting must give the patient a portable copy of her prescription so she can shop around for the prescribed lenses.

For better or worse, I provided competition input to the FTC team that reviewed the 2004 rule and eventually drafted the 2020 amendments. Well over a decade after the rule’s adoption, staff saw evidence of widespread noncompliance. Yet by 2020, the FTC had brought zero enforcement actions alleging, much less showing, violations of the prescription-release requirement. To the best of my knowledge, the count remains zero.

The FTC has issued warning letters, with batches going out in 2016, 2020, 2023 (called “cease and desist” letters in that case), and 2025. I don’t mean to suggest those letters had no effect on their recipients. But letters to a couple dozen prescribers go only so far—perhaps not very—in establishing a credible threat of enforcement more broadly.

The CLR imposes relatively straightforward, low-cost requirements, but enforcing it isn’t easy. The FTC does, of course, enforce other rules. Still, regulations impose costs on regulators as well as on those they regulate. And for better or worse—maybe better—the commission is far from a well-oiled regulatory machine. Actual FTC rule enforcement often looks like case-by-case law enforcement, something the FTC is designed to do.

At one level, there’s an easy answer to whether the FTC should have adopted its Contact Lens Rule or, for example, its Children’s Online Privacy Protection Rule. Congress expressly charged the FTC with issuing those rules, addressing those issues, in those ways. For the Contact Lens Rule, that charge—and the prescription-release requirement—appears in the federal Fairness to Contact Lens Consumers Act. There’s no such charge behind this ANPRM.

Of course, none of this disputes the problem of impersonation fraud or rules out platform negligence. And it seems within the realm of possibility that well-tailored regulations can address demonstrated market failures without doing more harm than good.

It’s just that real challenges—both general and FTC-specific—stand between recognizing a problem and adopting and enforcing a rule that does more good than harm. A little “regulatory humility” can go a long way.

Four Cases, One Online Ad, Plenty of Extrapolation

The very different rule adopted in 2024 (16 CFR Part 461) has been in force for just over two years. Since then, the commission has brought four enforcement matters—all in 2025—under it. I haven’t done a deep dive, but I’ve read the complaints, and they do appear to involve bona fide fraud. So good on ’em as enforcers, as far as that goes.

But of the four complaints (Blackstone Legal, Click Profit, Panda Benefit Services, and Superior Services), only one says anything about online advertising, targeted or otherwise. In Click Profit, the FTC alleged that consumers typically first encounter the defendants’ advertisements on search engines, social media, and third-party websites, where deceptive claims lure them to Click Profit’s websites. In the other three, bad old-fashioned communications seem to be the fraudsters’ routes to their victims.

For example, in Blackstone Legal, the FTC alleged:

Defendants generally initiate contact with consumers by sending letters to consumers that contain consumers’ sensitive personal information and threaten damage to consumers’ credit scores. In some instances, however, Defendants initiate contact by placing telephone calls to consumers.

In Panda Benefit Services:

Defendants’ scheme relies heavily on false and misleading representations made by Defendants’ representatives to consumers, often made during an initial call between a telemarketer and a consumer. In many instances, Defendants use mailers delivered to consumers’ mailboxes to entice consumers to call them and speak to a telemarketer.

And in Superior Services, the FTC alleged that the fraud began “with the delivery of a mailer that, based on its contents and appearance, leads consumers to believe it has been sent by a government agency or a government-affiliated student-loan servicer.”

Mass mailings, telephone calls, etc. The question of how the fraudsters identified their targets remains. But nothing in those three complaints suggests that online platforms’ inadequate screening of targeted ads played a key role.

As for Click Profit, the complaint says little about what, if anything, the platforms or third-party websites should have done differently. That may be understandable given the nature of the 2024 rule, but it underscores the limits of the FTC’s enforcement experience in this area.

The point is not that Click Profit’s approach is unique. The FTC’s 2025 report, Protecting Older Consumers: 2024-2025, certainly suggests otherwise. It observes a sharp increase since 2020 in older adults’ reported losses to “fraud that started on social media”—a significant concern even though “reported median individual losses for phone fraud remained far higher.” The underlying data have limits, but nothing in the report—or elsewhere, as far as I know—indicates that such scams are infrequent or declining.

The report is useful and gives cause for concern. But it doesn’t examine the role of targeted advertising or the tools platforms use to mitigate fraud. And the commission’s recent, relatively limited enforcement experience with conduct that might fall under a major expansion—and reorientation—of the existing rule doesn’t get us much further.

An AMG Workaround or Too Clever by Half?

There are other nits to pick and threads to pull, and they may not all be trivial. But I’ll skip some to focus on the fraught—and flagged—matter of the AMG workaround.

In AMG Capital Management v. FTC, the Supreme Court held that Section 13(b) of the FTC Act did not “authorize the Commission to seek, and a court to award, equitable monetary relief such as restitution or disgorgement.” Those remedies return money to victims or strip wrongdoers of their gains. The decision limited the FTC’s ability to obtain monetary relief for first-time violations.

That was a bummer from an enforcement perspective, especially where monetary remedies, penalties, or damages would be most efficient, forcing wrongdoers to bear the costs of their wrongdoing.

But that barrier is part of the commission’s statutory design, not a drafting error or an eccentric reading by the Court. Not for nothing, AMG was a unanimous decision. And the limitation goes hand in hand with Section 5’s extraordinary breadth and generality. Its original prohibition covers “unfair methods of competition.” The 1938 Wheeler-Lea Act amendments added the prohibition of “unfair or deceptive acts or practices,” establishing the FTC’s basic consumer-protection authority.

After a breezy—one might say “conclusory”—identification of market failure, the ANPRM declares that a new rule could realign “incentives by subjecting Platforms to court-imposed monetary sanctions for their involvement in furthering those [fraudulent—or fraud-facilitating] ads.”

Indeed, it could do that, among other things. But statutory damages for knowing rule violations might or might not be efficient—or even roughly tethered to the costs of consumer harm. And the commission’s sunny observation could easily read as an attempt at a clever AMG workaround. If so, courts might think it too clever by half. (Note to self: some thoughts do not need to be shared; and some sharing shouldn’t be published in the federal register.)

Yes, the process is cumbersome. That’s part of the point—a point underscored by Sections 5(n), 18, and 22, and the judicial decisions mentioned above. Those provisions impose substantive and procedural limits on Magnuson-Moss rulemaking, the process the FTC uses to issue UDAP rules.

All of that counsels caution. At one level, I get it. Fraud is fraud. It’s widespread. It harms consumers—very directly, and sometimes catastrophically—and core fraud cases come with no procompetitive rationale. Combating fraud sits at the heart of the FTC’s UDAP authority and has driven many sound enforcement actions, under this administration and historically.

But while core fraud cases may avoid hard questions about tradeoffs—including those required by Section 5(n)—the regulations contemplated in the ANPRM raise those questions squarely.

Slow Your Regulatory Roll

It’s tradeoffs all the way down. The FTC’s ANPRM acknowledges them, noting that targeted advertising “capabilities benefit both businesses and consumers.” It also asks whether “the Platform UDAPs identified . . . have any benefits for businesses or consumers?”

Good, although I’d rewrite that question to tie it more directly to the tradeoffs. For one thing, the FTC hasn’t actually found UDAP violations caused by platforms’ inadequate or negligent filtering.

More than that, what tradeoffs would a regulation requiring . . . what, exactly? . . . create? Certain standards for platform tools to identify and suppress bad actors’ targeted ads? One-size-fits-all standards? Static or dynamic standards? These tradeoffs concern both ad quality—for advertisers and consumers—and the means of funding billions of dollars’ worth of content that consumers can access for free.

Of course, tradeoffs cut both ways. The benefits of targeted advertising don’t make the costs of fraud disappear. But those benefits are real, complex, and substantial.

Those benefits also warrant consideration alongside the FTC Act’s rulemaking restrictions, the case law, and the very messy history of FTC rulemaking. On that history, there’s this and this—among many other things—from the International Center for Law & Economics (ICLE).

There’s a substantial body of literature besides, including contributions from former FTC chairs Ohlhausen and Muris; Muris and Howard Beales, former director of the FTC’s Bureau of Consumer Protection; former Chairman William Kovacic and David Hyman; and administrative law scholar Richard Pierce.

None of this says “do nothing.” But most of it counsels caution. So maybe . . . slow your regulatory roll.

Research First. Rulemaking Second.

Yes, there’s a long road from an ANPRM to a final rule. Yes, this ANPRM recognizes some of the complex tradeoffs further regulation might entail. No, it doesn’t specify what the commission intends to adopt. Indeed, it asks whether further regulation is warranted: “Is there a need for new regulations to prevent Platform UDAPs”?

As I suggested above, this could have been an RFI. The ANPRM raises many important questions, even if some should have been posed differently. It’s not wrong to ask stakeholders, academics, and others for input and data.

But the commission has considerable research expertise at its disposal. Why not deploy it first to investigate these questions, report on what it learns, then issue a better-informed, more targeted ANPRM—if warranted?

There’s a lot to untangle, no doubt. But that’s the job of a research-based enforcement agency, and the commission and its staff are capable of doing it. An ANPRM should cite systematic, pertinent research heavily, speeches and newspaper articles sparingly.

A Case for Bringing Cases

I get it. Law enforcement can be cumbersome. But the FTC knows how to do it. Bringing sound UDAP cases—and litigating them to decisions on the merits—can address real harm and establish legal precedents.

Cases and their underlying investigations can also illustrate how Section 5 violations cause consumer harm under varied facts and circumstances. That, in turn, can inform firms, consumers, lawmakers, economic and policy research, and even regulation.

It’s not easy, but it’s at the core of the FTC’s statutory mission. And the commission can pursue it where it has good “reason to believe” it will be constructive, without the cumbersome checks Congress has placed on the agency’s consumer-protection rulemaking.

Fix the Rules for Making Rules

Perhaps regulatory humility—like charity—should start at home. In 2021, under Lina Khan’s leadership, the FTC adopted significant changes to its internal rules of practice. The commission adopted them on a party-line vote, over a vigorous dissent from Commissioners Christine Wilson and Noah Phillips, without any prior ANPRM or NPRM.

The changes came under the banner of streamlining—a way to turn “the page on decades of self-imposed red-tape and return to the participatory process for issuing Section 18 rules that Congress envisioned.”

Heck, much of it would involve overdue, largely routine updating and cleanup. You know, “conforming edits to make the rule language more gender-neutral; use active voice instead of passive voice; replace ambiguous uses of ‘shall’ with ‘may,’ ‘will,’ or ‘must’ as appropriate; make nonsubstantive grammatical changes; and add and standardize citations to the U.S. Code where appropriate.” How could anyone—much less sitting commissioners—object?

Well, Wilson and Phillips saw it differently:

The majority asserts that the revisions to Section 18 procedures are designed to “streamline” rulemaking and reduce “red tape.” (To be clear, the red tape in question helps avoid the ill-advised application of red tape to American businesses.) What the changes – adopted without public input – in fact do is fast-track regulation at the expense of public input, objectivity, and a full evidentiary record. We see no need for bulldozing procedural safeguards in our Section 18 rulemaking Rules of Practice and are concerned in particular that the reforms undercut the independence of those charged with conducting evidentiary hearings, limit valuable input from the public, and reverse decades of practice regarding agency transparency. (internal citations omitted) 

They saw not-so-much streamlining as a “corruption of the fact finding process.” They had this wacky idea:

Better regulation—which is not the same as more regulation—ought to be the goal. By reducing transparency and objectivity in the rulemaking process, the revisions to our Section 18 procedures move us away from that goal and toward more—and worse—regulation.

The dissent is a tight, well-reasoned six-page document. I commend it to you. Not incidentally, I agree with every line—rare for me, but that’s on me, and it’s a welcome novelty.

This commission has been more attentive to process than the last. That’s a good thing. Enforcers don’t tend to welcome constraints, but I suspect Ferguson might share many of Wilson’s and Phillips’ concerns.

My proposed first step toward amending FTC regulations is to rescind the 2021 changes to the rules of practice. No Magnuson-Moss restrictions stand in the way. Easy peasy.

The post Fraud, Platforms, and the Art of Getting Ahead of Yourself appeared first on Truth on the Market.

Assemblyman Chigozie Onyema (D-Newark) introduced Assembly Bill 5532, the Grocery Retailer Opportunity to Compete Act, on Sept. 17. It would bar grocery suppliers with at least $6 billion in annual New Jersey revenue from offering different “terms of sale” to retailers or wholesalers buying the same product at roughly the same time. Those terms cover prices, discounts, rebates, promotions, packaging, delivery, and payment.

If enacted, the bill would likely mean fewer discounts, higher prices for shoppers, and less bargaining power for the very independent grocers it aims to help.

The Garden State Doubles Down

A5532 closely tracks New York Senate Bill 8563, introduced last November. That bill passed the New York Senate in May 2026 but never reached an Assembly vote before the session ended. I submitted comments to the New York Legislature examining its economics.

Both bills set the same $6 billion supplier threshold and define a “dominant retailer” as one with at least $18 billion in annual sales and a store or distribution center in more than 20 states. Both also let smaller buyers demand anonymized copies of the terms suppliers offered to dominant retailers.

The New York bill’s sponsor, Sen. Cordell Cleare (D-Manhattan), argued that suppliers recoup discounts to big chains by charging small stores more. Economists call this the “waterbed effect”—push prices down in one place, and they rise elsewhere. It makes for a tidy metaphor, but the theory is weak, and no evidence establishes such an effect in retail grocery markets.

New Jersey’s bill goes further in two ways. New York’s version required equal terms only for buyers purchasing the same quantity, up to one truckload. New Jersey drops that condition. A supplier would owe every buyer of “the same grocery product” at “approximately the same time” the same terms, whether the order fills one pallet or a fleet of trucks.

New Jersey also requires courts to award triple damages and attorney’s fees to any plaintiff who wins. New York’s version merely capped penalties at three times damages.

Everyone Gets the Same Worse Deal

Under A5532, a discount becomes everyone’s business. Any concession a supplier grants one buyer becomes a concession it owes every buyer of that product, and the disclosure rule lets them find out about it. For suppliers, the cheapest response would be to offer fewer concessions and settle on a single, higher price.

The bill requires no proof that a price difference actually harmed competition. That means it goes even further than the federal Robinson-Patman Act (RPA), passed in 1936, which applies only when a price difference may substantially lessen competition. Even with that limit, the bipartisan Antitrust Modernization Commission concluded in 2007 that the RPA discourages discounting and likely raises consumer prices.

Discount-store shoppers would bear the greatest burden. Emek Basker’s review of the research reports that Walmart’s grocery prices run about 10% below competitors’ prices, and that rival grocers cut prices 1% to 3% when a Walmart supercenter opens nearby.

Jerry Hausman and Ephraim Leibtag estimate that low-income households receive a disproportionate share of the gains from supercenters. Smaller wholesale discounts would mean higher shelf prices for those families.

Too Local to Get the Good Deal

New Jersey has about 180 ShopRite supermarkets. Most belong to independent operators who are members of Wakefern Food Corp., a member-owned cooperative based in Woodbridge, New Jersey. Members agree to buy at least 85% of their products through the cooperative. Wakefern uses that purchasing commitment to negotiate with suppliers and operate its own distribution centers.

The cooperative’s stores rang up $20.7 billion in sales in fiscal 2025, clearing the bill’s $18 billion threshold. But Wakefern falls outside the definition of a “dominant retailer” because its stores operate in nine states, short of the required 21. That geographic distinction says little about buying power—and hurts the cooperative twice.

First, the equal-terms rule applies to every buyer, dominant or otherwise. If Wakefern used its members’ purchasing commitment to negotiate a larger promotional allowance—supplier funding for promotions—the supplier would owe the same allowance to Walmart, Costco, and every wholesaler buying that product. Suppliers have little reason to reward a commitment when they must give everyone the reward. Members, in turn, have less reason to commit.

Second, New Jersey’s bill excuses a price difference “predominantly due to the dominant retailer’s internal distribution, or distribution to a subsidiary.” New York’s version covered any retailer that distributed its own goods. Wakefern’s network includes nine warehouses in New Jersey and Pennsylvania. But Wakefern falls outside the dominant-retailer definition, and its member stores own the cooperative, rather than the other way around. As written, the defense therefore does not cover its distribution centers. Its suppliers would have to rely on a general efficiency defense.

In other words, while Walmart’s warehouses get explicit protection, ShopRite’s are left to make an argument in court.

Wakefern might use the disclosure rule to demand the chains’ contract terms. Whether that leaves it better off depends on what those terms reveal.

Bargaining Power Sold Separately

Cooperatives like Wakefern let independent grocers pool orders, commit to purchases, and bargain like a chain. New York’s legislative findings acknowledged as much, noting that small grocers buy “through co-operatives and distributors that provide comparable scale.” When a big chain coerces a supplier, existing antitrust law already applies—and requires proof of harm to competition.

New Jersey’s bill promises grocers an opportunity to compete. What it delivers instead is an equal opportunity to pay more.

The post The Garden State’s War on the Good Deal appeared first on Truth on the Market.

The law launches a significant experiment in antitrust federalism—the division of antitrust authority between federal and state governments. It is also one that risks replacing economically disciplined analysis with political suspicion and litigation pressure. The likely result is less innovation, less investment, and more uncertainty for firms whose markets extend well beyond California.

The final enrolled text of AB 1776 adds Sections 16730 through 16732 to California’s Business and Professions Code and takes effect Jan. 1, 2027. The Legislature softened earlier versions by removing the provision allowing private parties to sue under the new prohibition, deleting the separate ban on “unreasonable restraints of trade,” and requiring plaintiffs to allege and prove “substantial market power” through direct or indirect evidence. Those changes deserve credit.

But that credit goes only so far. The statute declares federal antitrust interpretations “at most instructive,” directs courts to interpret California antitrust law liberally, and tells them to keep “maximizing” deterrence in mind. It covers monopolization, attempted monopolization, maintenance of a monopoly or monopsony—market power on the buying side—and conspiracies to monopolize or monopsonize. In effect, California has announced plans to enact a competing body of antitrust law with national significance.

Political Credit, Economic Costs

The statutory details matter. Section 16731(a) prohibits “every person” from monopolizing or monopsonizing, attempting either, maintaining a monopoly or monopsony, or combining or conspiring with another person to do so. Section 16731(c) requires substantial market power, but plaintiffs need not define the relevant market and establish the defendant’s share of it—the conventional approach. Direct evidence may suffice, and California courts must decide what counts as “substantial” power under a statute that deliberately distances itself from federal doctrine.

There is a small-business exemption, but it’s narrow. It covers only independently owned and operated California businesses with no more than 100 employees and no more than $10 million in average annual gross receipts. Medium-sized firms, national franchisees, and businesses headquartered elsewhere remain exposed.

Limiting enforcement to public officials removes one obvious danger—entrepreneurial private lawsuits seeking treble damages under the new section. Public enforcement nonetheless still carries its own potentially troublesome incentives. Attorneys general and district attorneys can earn political credit from a high-profile monopolization case, even when the theory is weak. The costs of a mistaken case—lost investment, delayed product launches, forgone integrations, and weaker competition—fall on scattered firms and consumers, making them hard to trace back to the enforcers. The instruction to “maximize deterrence” compounds this incentive problem.

Deterrence Without the Guardrails

Modern monopolization law grapples with a basic problem—vigorous competition and anticompetitive exclusion can look deceptively similar. A low price can be predatory or an efficient response to a rival. A refusal to deal can exclude competitors or protect investment and avoid subsidizing a free rider. A bundle can shut out a rival or lower distribution costs. An exclusive contract can entrench market power or encourage investments tailored to a particular business relationship. Courts must distinguish conduct that harms competition from conduct that benefits consumers and draws new competitors into the market. Aggressiveness alone tells them little.

Federal monopolization doctrine is imperfect, but it reflects decades of effort to manage the costs of getting that distinction wrong. As I and other Truth on the Market writers have emphasized, the error-cost framework weighs the expected costs of false positives (condemning beneficial conduct) against those of false negatives (allowing harmful conduct).

A false negative may let market power persist, but successful markets often attract entry and innovation. A false positive can suppress conduct that would otherwise lower prices or improve products, with benefits that may never return. Judge Frank Easterbrook’s classic analysis identified the institutional problem: Procompetitive conduct wrongly condemned has no constituency lobbying for its restoration.

Earlier COMPETE Act drafts would have expressly abolished federal safeguards for evaluating conduct. The final law stops short of that, but its structure still steers courts toward the open-ended inquiries those safeguards were designed to discipline. Federal precedent is now to be considered merely instructive, and California courts must interpret the law liberally. “Free and fair competition” and maximum deterrence take priority, with no clear commitment to consumer welfare or workable limits on enforcement.

Those instructions give enforcers more room to cast successful competition as suspect after the fact. The risk is especially acute in technology, labor, health care, and platform markets, where complex business models make competitive effects hard to assess.

Market Power Needs a Market

Supporters of AB 1776 portray market definition as an escape hatch that lets dominant firms avoid liability. That misses its economic purpose. Defining a market identifies the alternatives that constrain a firm. Without examining those substitutes, courts cannot assess market power, determine whether conduct plausibly excludes rivals, or distinguish harm to competition from a competitor’s disappointment.

Direct evidence can sometimes establish power without a formal market definition, but the phrase is no magic wand. Courts still need to compare prices, output, quality, innovation, entry, and consumers’ willingness to switch to alternatives.

The statute’s substantial-market-power requirement helps, but leaves key questions unanswered. Could temporary popularity, a successful product launch, leverage over a particular trading partner, or a naturally local employment relationship suffice? The law offers little guidance.

The danger is greatest in monopsony cases. Employers may negotiate better terms with suppliers, reorganize work, use software to coordinate logistics, or decline to pay a worker’s preferred wage. None of those facts alone establishes harm to competition. Treating bargaining power as an antitrust offense risks turning ordinary contracting disputes into government-managed labor policy.

The Golden State’s National Veto

State antitrust enforcement has genuine virtues. State attorneys general can uncover local cartels, represent consumers whose claims are too small to pursue individually, and share information with federal enforcers. States can also tailor enforcement to local conditions.

But unilateral conduct in modern markets rarely stays local. A national platform, pharmaceutical company, payment network, or manufacturer generally cannot run one business model in California and another everywhere else. California’s rule will put pressure on conduct nationwide.

My recent Truth on the Market discussion of the “veto stack” captures the problem. Federal clearance increasingly gets a firm through only the first gate. A coalition of states may then challenge the same conduct under a different standard, producing a succession of conflicting decisions. No settlement fully settles, no clearance fully clears, and the most aggressive jurisdiction gains leverage over national markets. The COMPETE Act compounds the problem by expressly declining to treat federal interpretations as conclusive.

States have room to depart from federal law on claims by indirect purchasers—buyers who purchase through intermediaries—who may sue, and on particular forms of coordinated conduct. Unilateral conduct poses a special challenge, however, because the business decisions, investment, and consumer effects commonly cross state lines. When California lowers the threshold for condemning product design, pricing, distribution, or employment practices, it exports its policy to businesses and consumers elsewhere.

Courts will face difficult questions under the dormant Commerce Clause, which limits state interference with interstate commerce, and implied preemption, which asks whether federal law implicitly displaces state law. But even if the statute survives judicial review, settlement pressure and compliance changes can impose substantial costs nationwide.

May the Best-Connected Firm Win

AB 1776’s political pitch is familiar. Big businesses have “crushed” competition, and small firms deserve a fair shot. But protecting competitors from better products, lower prices, more efficient logistics, or more attractive employment terms can keep inefficient firms afloat at consumers’ and workers’ expense. Antitrust turns on whether conduct harms welfare by impairing competition. A politically sympathetic competitor’s losses tell us little on their own.

That distinction matters because antitrust invites rent seeking—efforts to secure advantages through government intervention. Incumbents may seek rules that handicap disruptive entrants. Suppliers may cast a buyer’s efficiencies as monopsony. Labor groups may recast ordinary employment arrangements as exclusionary conduct, while local businesses seek shelter from national competitors. A broad unilateral-conduct law gives each group a legal vocabulary for asking government to tilt the playing field. The beneficiaries are concentrated and politically organized. Consumers, workers, entrepreneurs, and future entrants bear the scattered costs.

International Center for Law & Economics (ICLE) authors Eric Fruits, Brian Albrecht, Daniel Gilman, and Ben Sperry warned that AB 1776 could invite strategic litigation, drive settlements through defense costs, and pressure firms to raise prices, limit product integration, or change employment practices. Babette Boliek likewise described the earlier proposal as a return to “worthy men” antitrust—an approach that protects favored businesses at the expense of impartial rules governing competition.

The final amendments leave those concerns largely intact. Removing private suits narrows the pool of plaintiffs, but lobbying and political incentives remain. So does the pressure to settle uncertain cases to avoid years of discovery and reputational harm.

The statute’s design also creates a practical asymmetry. The state can impose the burdens of a complex case before demonstrating that a practice has raised prices or reduced output. An investigation can mean years of document production, depositions, work by economic experts, and disclosure of commercially sensitive plans. A large incumbent may be able to absorb that expense, while a growing firm might delay investment or abandon a transaction that no longer makes economic sense.

Meanwhile, smaller rivals may come to rely on political protection instead of innovation. A law advertised as helping entrepreneurs can end up favoring firms best equipped—and best connected—to navigate politicized enforcement.

A Softer Landing for a Bad Idea

The legislative history offers modest relief, in that earlier versions would have departed even further from federal monopolization law. They dispensed with conventional market definition and market-share thresholds, allowed private lawsuits, and expressly made several federal safeguards unnecessary. Those included recoupment in predatory-pricing cases (the prospect of recovering losses through later price increases), balancing effects across the markets a multisided platform serves, and the as-efficient-competitor principle, which asks whether conduct could exclude a rival as efficient as the defendant.

The final bill’s substantial-market-power requirement, limit to public enforcement, and deletion of the separate “unreasonable restraint” prohibition are also meaningful improvements, but they nonetheless leave substantial defects intact.

California courts must still develop a new body of law governing unilateral conduct, with federal interpretations serving only as guidance. The statute still commands liberal interpretation and maximum deterrence. It supplies no clear consumer-welfare rule, no safe harbor for conduct that increases output or quality, and no explicit protection for efficiencies that benefit participants in related markets.

The softer language may have eased passage and made the law harder to challenge politically. In the end, lawmakers trimmed the most dangerous provisions while leaving courts and enforcers broad discretion to fill in the rest.

Do as Washington Says, Not as California Does

The law also weakens the United States’ case for economically grounded antitrust abroad. U.S. officials and scholars regularly criticize European and other foreign regimes for approaches that prioritize precaution, regulation, or protectionism—presuming dominant firms owe rivals access, favoring small suppliers, or treating size and political influence as harms in themselves. The consumer-welfare standard offers a coherent alternative, tying enforcement to the competitive process and consumer outcomes.

California gives foreign regulators an easy rejoinder. They can cite its broad mandate, treatment of federal precedent as merely instructive, liberal-interpretation rule, and emphasis on maximum deterrence as evidence that the United States is loosening its own economic constraints on enforcement. California may call that state sovereignty, but overseas policymakers will see a U.S. jurisdiction embracing a more precautionary model. Washington will thus have a harder time challenging European theories of abuse, regulation of digital “gatekeepers”—platforms that control access to users—or industrial favoritism on economic grounds.

That credibility matters as antitrust rules shape trade and investment. A foreign government favoring national champions can point to California’s willingness to let officials second-guess successful firms with fewer constraints than federal courts impose. The immediate dispute may involve a platform or employer, but the lasting cost will be a weaker American case for rules that encourage entry, experimentation, and voluntary exchange. Fragmented domestic antitrust thereby gives foreign regulators more cover to treat competition policy as a license to manage the economy.

More Cases, Less Competition

California had better options. It could have funded investigative expertise, coordinated with the U.S. Department of Justice and Federal Trade Commission, and targeted demonstrable local harms. It could have tied state monopolization claims to the federal consumer-welfare framework while allowing California-specific remedies. Referring evidence to federal authorities and publishing clear enforcement guidelines would also have preserved room for state experimentation while respecting the need for national coherence.

Newsom’s signature commits California to a more uncertain path. Lawmakers softened the COMPETE Act, but it still invites officials to use antitrust to steer industries and redistribute economic advantages. The costs will emerge gradually: a startup loses financing because a partnership with an incumbent carries legal risk; a platform forgoes an efficient integration; a manufacturer raises prices to avoid predatory-pricing allegations; or an employer abandons a productivity-enhancing work arrangement. A national firm may redesign its practices across all 50 states because California’s rule costs too much to ignore.

More antitrust liability hardly guarantees more competition. Sound enforcement minimizes the combined costs of market power, mistaken intervention, administrative error, and rent seeking. The COMPETE Act moves in the opposite direction, making cases easier to bring and businesses harder to build.

California has made it easier to compete in court. Competing in the market just got harder.

The post California’s COMPETE Act: May the Best Firm Lose appeared first on Truth on the Market.

A recent white paper that I co-authored with Lura Forcum and Julian Morris, published by the International Center for Law & Economics (ICLE), suggests these sweeping laws rest on weak evidence, are unlikely to work, and conflict with free speech values. Below, I examine recent attempts to restrict social media access by age in the United States and abroad, and explain why they may face insurmountable legal challenges.

Age Gates Meet the First Amendment

In the United States, lawmakers have aggressively pursued age-based bans. A prominent recent example is the bipartisan 16 & Up Social Media Act, introduced by Reps. Erin Houchin (R-Ind.) and Jake Auchincloss (D-Mass.). The bill would establish a national standard prohibiting children under 16 from maintaining social media accounts.

The bill aims to protect children under 16 from allegedly addictive platform designs, but it likely fails First Amendment scrutiny. Laws restricting access to protected speech generally must satisfy strict scrutiny—that is, use the least restrictive means to serve a compelling government interest. As the U.S. Supreme Court put it in Brown v. Entertainment Merchants Association, “[e]ven where the protection of children is the object, the constitutional limits on governmental action apply.”

In Brown, the court found that studies showing effects “both small and indistinguishable from effects produced by other media” could not establish a compelling interest in barring minors from protected speech—in that case, violent video games. Evidence of a “causal link” between social media use and adolescent harm is similarly lacking here.

A 2025 meta-analysis led by Christopher Ferguson pooled 46 studies of social media’s effects on adolescent mental health and found a correlation of 0.061, below the level researchers normally consider meaningful. A Cambridge review of 143 studies covering 1.1 million adolescents put the figure at 0.12, suggesting social media might explain about 1.4% of the variation in anxiety and depression. “Might” matters here, because the relationship runs at least partly in reverse. Research suggests that social media use did not predict later depression, but depression predicted later social media use among girls.

Just as importantly, banning children under 16 from creating social media profiles is not the least restrictive means of protecting them online. In Brown, the court expressed “doubt[] that punishing third parties for conveying protected speech to children just in case their parents disapprove of that speech is a proper governmental means of aiding parental authority.” Lawmakers could instead educate parents and minors about staying safe on social media, including how to use widely available safety tools and other practical means.

Federal courts have already blocked less burdensome laws requiring parental consent before minors can create social media profiles in Arkansas, Louisiana, Utah, Ohio, and Georgia. That leaves the 16 & Up Act with a steep climb under strict scrutiny.

The bill’s supporters would likely point to a recent 11th U.S. Circuit Court of Appeals opinion lifting a preliminary injunction against Florida’s HB 3.That law requires age verification and parental consent for 14- and 15-year-olds to create accounts on social media sites with “addictive features.” The 11th Circuit applied intermediate scrutiny—a less demanding standard—after accepting the argument that HB 3 regulates the form of expression rather than its subject matter, making it content-neutral.

Under that standard, the court found a substantial government interest in protecting minors from platforms with addictive features. It also found that the age-verification and parental-consent requirements for 14- and 15-year-olds imposed no greater burden than necessary to serve that interest. Litigation in the case remains ongoing.

Courts remain split at this point on whether autoplay, ephemeral content (posts that disappear after a set time), notifications, and infinite scroll are product-design features subject to commercial regulation or protected editorial decisions about how to present speech. In Moody v. NetChoice, the Supreme Court held that “[d]eciding on the third-party speech that will be included in or excluded from a compilation—and then organizing and presenting the included items—is expressive activity of its own.” Treating laws targeting allegedly addictive features as content-neutral therefore arguably fits uncomfortably with the Court’s findings in Moody.

A blanket ban on social media profiles for anyone under 16 appears unlikely to survive scrutiny under current Supreme Court precedent. But lawmakers keep trying, and lower courts remain split. The Supreme Court may eventually have to settle the question.

Liberté, Égalité, No Blanket Ban

Age-based bans also face at least some legal hurdles beyond the United States. France’s Constitutional Council recently considered a challenge to a similar law that would “prohibit[] access by minors of [under] fifteen years of age to online social networking services provided by online platforms.” In a somewhat surprising move, the council struck it down.

The council found that a blanket ban failed to account for individual circumstances, parental judgment, or the risks of particular services:

15. In this regard, neither the contested provisions nor any other provision sets out the conditions under which the holders of parental authority or the legal representative of the minor, duly informed of the potential risks and safeguards presented by the services concerned, may, in the child’s best interests and in the exercise of their duties under the law, decide to lift the prohibition, limit its scope or authorise access to certain services.

16. Thus, the prohibition established does not give rise to any particular assessment of the risk to the minor, taking into account in particular his age, his degree of maturity, his family situation as well as the nature of the service concerned.

17. Therefore, the legislator could not, without disregarding freedom of expression and communication, institute a general prohibition which had the effect of depriving minors of their freedom of access to many online communication services, without consideration of either the situation of the minor or the risks specific to these services.

18. In these circumstances, the contested provisions infringe upon this freedom in a manner that is not appropriate, necessary and proportionate to the objective pursued.

The council also identified a problem that age-verification mandates tend to bring along for the ride. Everyone—including adults—would have to prove their age to access protected speech, without adequate legal safeguards:

20. By prohibiting access for any minor under fifteen years of age to certain online services, the contested provisions imply, in themselves, that any person, even an adult, must prove their age before accessing them.

21. By failing to determine the conditions and limits within which it must be justified, the legislator has not provided the legal guarantees necessary to ensure compliance with these constitutional requirements.

France’s free speech protections are, it should be noted, narrower than those of the First Amendment. Even so, the Constitutional Council rightly found that the current evidence does not justify a complete ban of teenagers from social media.

Australia’s Great Barrier to Speech

Australia’s similar law bans users under 16 from creating social media accounts. It faces a legal challenge on the grounds that it “impermissibly burdens the implied freedom of political communication” of minors subject to the ban—a freedom derived from Australia’s Constitution.

Data suggest that, prior to the ban, minors used social media for political communication, including sharing news and opinions on the issues of the day. The plaintiffs in the case cite a study finding that:

Of the 26% of young Australians who reported they had been “significantly affected” by the ban: 51% said they were receiving less news; 47% reported less access to world news and events; 45% experienced less opportunity to share views or take action; 42% reported less access to local news and events; 41% reported less access to news about issues they care about; and 40% reported having fewer discussions about news with others.

In its brief, the government acknowledges “scientific uncertainty about causal links between social media use and mental-health outcomes, as well as their extent and likelihood…” Yet it argues that the “existence of credible risks of harm and credible causal pathways between account-related design features and those risks justifies Parliament’s decision to enact a precautionary protective measure before scientific consensus establishes that it will be effective.”

The High Court of Australia will have to decide whether that uncertainty justifies severely burdening minors’ freedom to engage in political communication on social media. There is good reason to think it does not.

Give Kids Some Space, Not Another Gate

Protecting kids online is a valid and urgent priority, but government bans on social media aren’t the answer.

A more nuanced approach would give parents better voluntary tools to set limits and add pauses to social media use, establish independent industry certifications for responsible app design, and teach kids how better to handle online risks.

Policymakers and parents also need to address unfortunate trends in the offline world. The steady erosion of children’s freedom to play and socialize independently has coincided with their turn to social media. Restoring opportunities for in-person, unsupervised play might do far more for adolescent mental health than banning its digital substitutes.

It’s time to move past the political theater of age-gating and give families practical tools that preserve the benefits of connectivity while helping kids stay safe. Kids need room to grow, online and off.

The post No Country for Young Posters: Social Media Bans Outrun the Evidence appeared first on Truth on the Market.

While the political appeal is clear, the implications for competition deserve a closer look.

Fuel prices are a politically combustible topic in Italy, as elsewhere. Despite a temporary fuel-tax cut (now being phased out), Italian diesel prices averaged €2.090 per liter in August 2026, 2.3% above the European Union average of €2.044. Prices have since risen sharply.

Policymakers have sought ways to bring them down without straining public finances, including through a potential tax on windfall profits or a cap on refining margins—the difference between refiners’ fuel revenues and crude-oil costs. Both proposals have some obvious shortcomings. Eni’s cap gives policymakers some breathing room.

There is some precedent. In France, TotalEnergies recently announced a similar cap, drawing similar reactions. And earlier this year, Eni kept its prices slightly below market levels without committing to an explicit target. It did the same in 2012, during another fuel-price spike.

Three questions warrant attention. Is Eni engaging in predatory pricing—that is, pricing below cost to drive out rivals and recoup its losses later? How does publicly announcing its intended price affect competition? And are commercial considerations the only explanation for the move?

Predation Needs a Payday

One possible explanation for Eni’s move is predatory pricing. The standard theory holds that an incumbent may cut prices below cost in order to weaken or eliminate rivals, then raise prices above competitive levels to recover its losses. The intuition is straightforward. Evidence of successful predation followed by profitable recoupment is harder to find, as Truth on the Market readers well know. Below-cost prices can hurt rivals in the short run, but unless the strategy creates lasting market power, competitors can return when prices rise, leaving the predator with losses it cannot recover.

Does Eni’s cap fit that story? A credible account would need to explain why Eni could expect to recover any losses the cap generates through higher prices later. That prospect is far from obvious in Italy’s fuel market.

A 2023 investigation by the Italian Competition Authority (AGCM) found that international benchmark prices for refined fuels (particularly gasoline and diesel) drove domestic price movements. Italy’s relatively high pump prices largely reflected higher taxes. The pretax component, by contrast, ranked among the lowest in the EU. Those findings offer little support for the claim that Italian motorists generally pay high prices because domestic oil companies systematically exercise market power.

Rivals’ willingness to match Eni’s cap also suggests that the announced prices may be commercially sustainable. That alone settles neither the predation question nor the broader competitive concerns. Still, a conventional predatory-pricing story is difficult to sustain. Italy’s retail fuel market offers no obvious route for Eni to absorb substantial losses today and recover them through sustained monopoly profits tomorrow.

A Cap and a Wink

The more intriguing competition issue may lie in the announcement accompanying Eni’s price cut.

Eni has publicly committed to a maximum price for at least 30 days, potentially through year’s end. Price announcements have long attracted antitrust scrutiny because they can help rivals coordinate. Transparency helps consumers compare prices, but it can also tell competitors what to charge—or what others plan to charge next.

The AGCM has repeatedly raised this concern in the fuel sector. In 2007, it alleged that the largest vertically integrated oil companies (those operating at multiple stages of the supply chain) had colluded, partly through periodic press releases disclosing recommended prices. It closed the case after the companies committed to stop those releases and made other changes the authority deemed sufficient to remove the mechanisms facilitating coordination.

In 2023, the AGCM opened another case, alleging that major oil companies coordinated how they passed mandatory biofuel-blending costs on to consumers. EU rules require suppliers to meet progressively higher biofuel targets as a share of road-fuel sales. According to the authority, periodic announcements by Eni and others about the resulting price increases helped facilitate the alleged cartel. (Full disclosure: I served as an economic adviser to one of the parties in that case, which is now under judicial review.)

The AGCM has even opposed legislation requiring greater fuel-price transparency. At a 2026 Italian Senate hearing, it said that publishing recommended prices daily, down to individual filling stations, could facilitate collusion and potentially violate antitrust law. It described that position as consistent with competition authorities’ established approach.

Eni’s announcement deserves to be seen in the context of that backdrop. A temporary price cap differs from daily publication of recommended prices, but the resemblance is hard to miss. The European Commission’s Horizontal Guidelines distinguish advertising current retail prices from making nonbinding announcements of future pricing intentions. But Eni’s commitment is stronger and extends further ahead—“an initial period of 30 days, with the possibility to be extended through the end of the year.” It gives rivals a view of the company’s pricing strategy over time, beyond today’s price at the pump.

Could that visibility make coordination easier? IP’s announcement of the same cap proves no such thing, but it does render the question more interesting. The competition concern is whether these public announcements—and any resulting price alignment—could help rivals sustain parallel pricing. But to be clear, announcing a price does not, by itself, establish unlawful conduct.

That leaves the AGCM in an awkward position. A heavily publicized pricing commitment seems an odd choice for a company hoping to avoid antitrust scrutiny. Yet Eni acted under strong government pressure, and the cap enjoys broad popular support. The authority now faces a test of its consistency. How should it assess a measure that appears to benefit consumers in the short run while resembling practices it has repeatedly scrutinized? Political applause adds no obvious answer.

Filling Up on Political Goodwill

Aggressive marketing may explain part of Eni’s move, but its relationship with the government might also form part of the explanation.

The Italian government is Eni’s largest shareholder, holding roughly one-third of its shares. Prime Minister Giorgia Meloni publicly thanked Eni for its announced cap. Within hours, after IP followed suit, she also thanked Azerbaijani President Ilham Aliyev and Socar Chairman Rovshan Najaf. She later extended her thanks to “Kuwait and the company in control of Italy’s gas stations” branded Q8.

The acknowledgments suggest a political dimension to these commercial decisions. Are companies forgoing legitimate profits in anticipation of favorable government treatment? One explanation, widely discussed in the Italian press, is that the large oil companies hope voluntary price restraint will ward off a windfall-profit tax.

Whether that bargain exists remains unclear. Smaller and independent retailers, meanwhile, warn that matching the caps would squeeze their margins. Government pressure to cap prices may offer motorists relief while placing unequal burdens on their suppliers. Large integrated companies may have more room to absorb those costs than smaller independent retailers.

Whether those effects prove temporary or durably reshape competition is an empirical question that requires evidence. The AGCM’s own sector inquiry identifies independent stations as an important competitive constraint on rivals. It also finds that access to wholesale fuel poses a critical challenge for retailers without their own refining operations, especially during supply shortages. Squeezing those retailers could carry costs beyond their balance sheets.

Eni’s motives—and the cap’s ultimate competitive effects—remain uncertain. The episode nevertheless shows how quickly rising prices can turn pricing into a political negotiation. Governments and consumers understandably want relief. But high prices also communicate scarcity and encourage suppliers to bring more fuel to market and consumers to seek alternatives.

A voluntary cap may ease motorists’ bills, but it creates no additional refined fuel or alternative sources of energy. If the cap starts holding prices below the level market conditions would otherwise produce, as some politicians want, it may weaken the incentives to expand supply and switch to substitutes.

Political gratitude alone won’t fill the tank.

The post Pump and Circumstance: The Politics of Italy’s Fuel-Price Caps appeared first on Truth on the Market.

Private investment will connect most of America. In many rural areas, however, getting that investment to pencil out will require policies that improve the incentives to build. The potential benefits are substantial—AI-assisted diagnostics at rural hospitals, precision agriculture that streams field imagery in real time, and reliable home broadband that lets higher earners work remotely and settle outside major metropolitan areas.

Much of that value spreads through communities in ways no provider can capture on a monthly bill. Without policies that lower construction costs, the places least likely to attract private capital will also be the last to get AI-ready infrastructure. That would widen the gap between connected and unconnected regions just as connectivity becomes central to economic competitiveness.

That makes the next phase of the Broadband Equity, Access, and Deployment (BEAD) Program more consequential than it might appear. After restructuring the program, the National Telecommunications and Information Administration (NTIA) has roughly $21 billion in unspent funds and has told states to direct that money toward more construction. But as in the first funding round, money alone won’t clear the path. Successful builds depend on timely, predictable access to public rights-of-way and the infrastructure along them.

Broadband’s Next Stop: The Permit Office

BEAD’s merits have long been debated, but the program offers rural areas that might otherwise remain unconnected a path to broadband capable of supporting advanced services. NTIA’s 2025 “Benefit of the Bargain” overhaul restored a technology-neutral, competitive grant process—allowing providers using different technologies to compete for funding. NTIA Administrator Arielle Roth said the changes saved about $21 billion of the program’s $42.45 billion allocation.

On Sept. 3, NTIA directed states to use the remaining funds first for additional deployment. Eligible locations include newly identified unserved addresses missing from the Federal Communications Commission’s (FCC) broadband map and those that became eligible after defaults or changes in other federal and state programs. Guidance on broader “non-deployment” uses remains pending.

Prioritizing deployment makes sense, but a second grant round will encounter the same obstacles as the first. A subsidy can pay for fiber, but it can’t make a utility replace a pole on time or make a county process a permit. Even with funding, providers face substantial risk when remote areas offer few potential customers.

Nationwide, BEAD projects may require as many as 86,000 permits. Each represents yet another place where a funded project can stall.

A Pole Tax on Rural Broadband

For many rural projects, securing permission to build costs more than building the network itself. “Make-ready” work—preparing poles for new equipment—and pole replacements can cost thousands of dollars per pole. Those costs become prohibitive when providers must attach to more poles to reach fewer subscribers. And when each company already on a pole must move its equipment in sequence, the wait can stretch for months.

The FCC has made real progress in addressing these buildout costs. One-touch make-ready lets a single qualified crew handle simple work in one pass, saving providers from waiting on each incumbent attacher in turn. The FCC’s July 2025 order added timelines for large applications, expanded providers’ ability to arrange work themselves when utilities miss deadlines, and sped approval of qualified contractors. In December 2023, the FCC also clarified that a new provider shouldn’t bear the full cost of replacing a pole the utility had already marked as unsafe or needed to upgrade anyway.

But two gaps remain, and both matter for BEAD. First, Section 224 of the Communications Act exempts poles owned by municipal utilities and electric cooperatives. Those owners control a substantial share of poles in the rural areas BEAD targets. Providers seeking access face rates and timelines unconstrained by federal rules, often without comparable state protections. Congress should extend Section 224—or equivalent enforceable deadlines and cost-allocation rules—to these poles while still allowing nonprofit owners to recover legitimate costs.

Second, the pole-replacement rule works only if providers can enforce it. When a utility demands full payment for a replacement, that opening bid effectively sets the price while the dispute drags on.

The FCC’s expedited “rocket docket” provides firm deadlines for resolving disputes. Codifying the cost-allocation principle would protect it from changes in FCC leadership, while targeted reimbursement funds could help cover replacements that warrant shared costs. Yet pole owners continue to try to sidestep FCC decisions and extract additional payments from broadband providers. Clearer statutory guidance could strengthen the FCC’s hand in enforcing its replacement-cost rules.

The Road to Broadband Could Already Be Open

For underground fiber, much of the cost comes from digging trenches and burying cables or conduit—the protective tubing that carries them. Coordinating broadband construction with roadwork can therefore deliver substantial savings.

A 2021 Federal Highway Administration rule already requires state transportation departments that receive federal-aid highway funds to designate a broadband utility coordinator and set up processes for registration, notification, and coordination. A January 2026 Government Accountability Office review found most states had made progress, though the effect on deployment remains hard to measure.

But weak “dig-once” policies simply notify providers when crews plan to tear up a road. While that helps providers whose construction plans happen to align with the roadwork, middle-mile and backhaul routes—the links connecting local networks to the wider internet—often won’t.

Stronger policies would require agencies overseeing federally funded road projects to install spare conduit whenever they open the ground, then lease it to providers. Proponents estimate that this adds only a few percentage points to a road project’s cost while sparing future builders the costs of another dig.

States are well placed to act. The governments administering BEAD grants also run the highway departments. Pairing second-round broadband construction with mandatory conduit installation on state road projects would stretch both pots of money.

Permitting processes need the same discipline. Right-of-way fees should cover a government’s actual costs, not serve as a tax on providers’ investment. Deadlines also need teeth. A “shot clock”—a deadline for deciding a permit application—does little to prevent delay if its only consequence is letting the provider sue after the clock runs out.

A self-executing “deemed granted” remedy would automatically approve an application when the deadline passes. That approach already applies to qualifying modifications of small wireless facilities. Extending it to ordinary right-of-way and construction permits would give BEAD grant recipients predictable schedules for projects with strict federal construction deadlines.

The Future Needs a Right-of-Way

BEAD’s second round gives policymakers a rare do-over. The first round showed how competition and technology neutrality could stretch federal dollars further than expected. The second can go further still—if pole-access costs and permitting delays stop eating into every grant.

NTIA should encourage states to prioritize permitting reform and financial backstops for pole replacements when allocating any remaining non-deployment funds. States should pair second-round builds with meaningful dig-once policies, cost-based permit fees, and automatic approval when permitting deadlines expire.

For its part, Congress should extend pole-access protections to municipal and cooperative poles and codify the FCC’s cost-allocation and one-touch make-ready reforms. Investors shouldn’t have to bet on a 20-year asset under rules a future FCC could undo at any time.

AI reaches users through networks that cross public land, local rights-of-way, railroad tracks, and utility poles—one approval at a time. Winning the next technology race will take computing power, electricity, and rules that let providers build connections to the people.

BEAD can buy the fiber. But first, reform must clear the path.

The post BEAD’s Second Act: The Future Is Pending Approval appeared first on Truth on the Market.

That missing sale matters to the Federal Trade Commission’s (FTC) proposed enforcement policy statement on personalized pricing. The FTC acknowledges that Congress has not authorized a blanket ban on personalized pricing and that the practice’s effects remain unclear. Yet when consumers reasonably expect prices to stay the same regardless of their personal data, the proposal says sellers should disclose that a price is personalized, why, and what types of data they used. The FTC says failing to do so is likely unfair or deceptive.

The FTC has reason to worry about sellers who invent discounts, break promises about data use, or tack on charges buyers never agreed to. But its proposal goes further, casting suspicion on price differences that sellers do not explain. That risks overlooking both the sales that tailored offers make possible and the trial and error through which sellers discover what customers will buy. It also assumes that the agency can decide which pricing practices need an explanation despite its own limited knowledge and political incentives.

Truth on the Market readers have already seen important objections to the proposal. Alden Abbott explains why a consumer’s expectation alone cannot establish deception. Daniel J. Gilman asks whether the economic research cited by the FTC supports its claims. This post asks what happens to the market processes that generate prices once the government starts supervising them—and how officials might use that authority.

Austrian economics, with its focus on how buyers and sellers discover opportunities through exchange, and Public Choice theory, which examines the incentives facing government officials, point toward a narrower policy. The FTC should pursue proven lies and consumer injuries under existing law. But it must also leave room for sellers to test prices, and not turn ordinary price search into a presumptive violation merely because software makes the search more precise.

Prices Don’t Come Preprinted

The proposal treats a nonpersonalized price offered to everyone as the neutral benchmark. An individualized price tailored to one buyer then looks like a departure that needs explaining. But prices do not come with a correct value stamped on the product. In “Principles of Economics,” Carl Menger explained that particular goods matter to particular people because they serve particular needs under conditions of scarcity. Ludwig von Mises added that the prices we observe record actual exchanges, not timeless properties of the goods exchanged.

The same ticket may be worth more to a traveler with a tight schedule than to one who can take another train. A software subscription may be valuable to an established firm but a gamble for a student trying it for a month. Even a carton of milk reaches buyers at a particular place and time, with different budgets, travel costs, and alternatives. The product may be the same. The circumstances of the sale are not.

Armen Alchian and William Allen called sellers who must find the prices buyers will accept “price searchers.” Buyers search, too. Sellers try offers and see which ones work; buyers compare sellers, substitutes, and the option of waiting. Both learn from the response. Coupons, negotiated prices, student discounts, retention offers, and promotions selected by software are all different ways to conduct that search. They use personal information to different degrees and raise different privacy questions. But each can also make a sale possible when a single price would leave the buyer out.

Friedrich Hayek’s knowledge problem helps explain why no one can know all the relevant facts in advance. Those facts include changing local conditions, private plans, urgency, and available alternatives. Prices convey some of that knowledge through actual offers and decisions. Better analytics may help a seller to spot patterns, but a given model can misjudge a given buyer. A competitor can cut its price. A customer can decline, delay, switch, or use a coupon. Yesterday’s demand estimate can be wrong tomorrow.

Israel Kirzner’s account of entrepreneurial discovery adds another piece. A successful offer may reveal demand at a price the seller had never tried. A rejection may reveal that buyers have a better alternative. A targeted discount may find customers who would never buy at the posted price. A post-transaction snapshot of the prices paid cannot show all the sales that failed to happen—or those a different offer might have made possible.

An algorithm helps a seller make and test guesses about demand, but it can’t read a buyer’s true willingness to pay from a file, and its prediction may collapse when the buyer’s plans or alternatives change. The buyer still decides whether to accept the offer. The market test is the transaction, not the model’s output. That distinction also limits what a regulator can infer ex post from the mere fact that a seller charged a personalized price.

None of this makes every price fair or puts sellers beyond the law. It does mean the FTC can’t assume that uniform pricing costs nothing, then measure harm against the lowest individualized price it can find. Sellers who misjudge demand lose sales. Buyers can search and switch when they have alternatives. An agency that gets the comparison wrong may turn its mistake into a precedent applied across many markets.

The Sales Ledger Has a Blind Spot

The usual critique of personalized pricing starts with customers who would have bought anyway. If a seller identifies those willing to pay more, it may raise their prices and keep more of the value they get from the purchase. That risk is real. But the analysis cannot stop with sales that already happen. It must also ask which sales each pricing practice makes possible.

Let’s go back to the $10 offer. The buyer who values the service at $8 never appears in the sales ledger unless the seller offers her a lower price. At $7, she buys, gains $1 in value beyond what she pays, and adds a sale. The seller will offer that discount only if the sale covers its relevant costs. If targeted offers become too costly or legally risky, it may stick with $10 and lose that buyer. Charging everyone $7 may sound like a simple fix, but the lost revenue from customers willing to pay $10 could make the extra sale—or the service itself—uneconomic.

This is familiar territory. A student rate can put within reach a product the student would otherwise skip. A retention offer can keep a subscriber from leaving. An airline discount can fill a seat that would otherwise take off empty. A coupon can win a shopper considering a rival store. These methods vary in how much personal information they use and what privacy concerns they raise. But the chance to reach buyers who would otherwise sit out predates artificial intelligence. Ludwig von Mises and Murray Rothbard both recognized that lower prices for such buyers can widen access, while competition constrains what sellers can charge others.

None of this means personalization always helps consumers. In a field experiment, Jean-Pierre Dubé and Sanjog Misra found that personalized prices raised expected profits and reduced total consumer surplus—the value buyers received beyond what they paid—relative to a profit-maximizing uniform price. Yet more than 60% of customers benefited from lower prices. In a study of auctioned cab rides, Nicholas Buchholz and his coauthors estimated that, compared with uniform pricing and accounting for drivers’ responses, personalization reduced consumer surplus by 2.5% while increasing total surplus by 5.2%. Those results depend on the platform’s market power, its pricing choices, and the baseline used for comparison. In a model of competing firms, Andrew Rhodes and Jidong Zhou found that the effects on consumers depend in part on market coverage—the share of potential buyers who purchase—and on which firms can personalize.

The findings differ because the markets and pricing methods differ. A program can give most shoppers discounts yet reduce their combined gains if a smaller group pays substantially more. A platform can create more value overall while keeping much of it. On the other hand, a rule that deters targeted offers can also take discounts away from the buyers most sensitive to price. The question is what happens in the market at hand, compared with a realistic alternative.

The distribution of benefits deserves the same care. Price sensitivity does not neatly track income: A wealthy shopper may hunt for coupons, while someone with less money may need a product too urgently to wait. Nor does one buyer’s higher price show that the seller could afford to serve everyone at the lower one. A consumer-protection agency should ask who gains access, who pays more, how many sales occur, and what a uniform-price policy would actually produce. A serious consumer-protection agency can ask those questions without treating an average or a single visible disparity as the whole story.

The Fine Print Has a Price

The FTC says sellers can deliver the benefits of personalized pricing without concealing how it works. Perhaps. A disclosure might help a buyer understand an offer. It might also change how buyers respond, how sellers set prices, and whether a small discount program remains worth running. What does a particular notice tell consumers, and what happens when sellers must provide it?

“Prices may vary by customer” is a short and simple notice. The FTC’s proposal to require explanation of a price’s “basis” and the “types” of data used is a much bigger undertaking. A seller’s estimate might draw on purchase history, location, timing, inventory, rival offers, and the chance that a buyer will switch stores. Some inputs are unreliable, and their importance can change from one sale to the next. A long technical list could reveal a seller’s competitive strategy without helping anyone decide whether to buy. Where a seller has created a misleading impression, a shorter notice may actually serve consumers better.

Design and placement matter, too. An online seller can put a brief statement beside the price and link to more detail. A store shelf offers less room and even less of a shopper’s attention. A lengthy notice may become boilerplate that few people read, while the cost of producing it still affects which offers sellers make. If the FTC expects a particular disclosure to help, it should test whether buyers understand it and change their purchasing decisions. Posting disclosures is not the same as informing people.

Who chooses to share information also matters. In a model of voluntary disclosure, S. Nageeb Ali, Gregory Lewis, and Shoshana Vasserman show that giving consumers control over what they reveal can sometimes improve their welfare, including by making firms compete harder for them. But the result depends on the market and on how the disclosure works. A consumer’s choice to share information is different from a government requirement that a seller explain its pricing method, which is not necessarily free of cost or unambiguously beneficial.

Sellers still have to tell the truth. If an ad implies everyone gets the same price, the seller must not hide a contrary practice. If a firm promises not to use health or location data to set prices, it must keep that promise. It cannot advertise one price, hide a fee until checkout, and call the final bill the advertised price. Each case involves a claim or practice the FTC can identify and assess. That gives the agency a sounder task than asking every seller to narrate how it arrived at an offer.

The Price Tag Said No Such Thing

Section 5 of the Federal Trade Commission Act prohibits deceptive acts or practices, including omissions that leave consumers with a materially misleading impression. The FTC’s Food Advertising Policy Statement draws a useful line: A seller may need to disclose a fact to keep an express claim from misleading buyers. Silence can also deceive when the circumstances make it an implied false claim. But an omitted fact is not deceptive simply because consumers would find it useful. Silence can complete a half-truth; it does not always speak.

The proposed pricing statement blurs that line. It asks whether consumers reasonably expect prices to stay the same regardless of their personal data, then says failing to disclose personalization is likely deceptive. Expectations help determine what buyers take from a seller’s words and conduct. On their own, however, they cannot turn every unspoken detail of pricing into a claim. A shopper might assume a coupon is available to everyone or that every visitor sees the same website price. The question under Section 5 is whether the seller’s words and the transaction’s context actually gave reasonable buyers a false impression that mattered to their decision.

That question often has concrete answers. A sign promising “one price for everyone” makes a claim. A shelf tag may tell shoppers what they will pay at the register. An ad calling a discount “just for you” may make a claim about who can get it. Evidence of how reasonable consumers understand those messages can support a deception case. But a displayed price ordinarily tells a buyer only what that buyer may pay. It does not necessarily promise that everyone else saw the same offer or that no personal data helped set it.

Congress has required specific disclosures in other settings. Under the Fair Credit Reporting Act, a business that takes certain adverse actions based on a consumer report must notify the consumer. That express duty counsels care before reading a comparable duty to explain prices into Section 5. The FTC’s proposed statement is guidance, with no independent force of law. It cannot create a general right to an explanation of an algorithmic price.

The FTC can still pursue false discounts, deceptive data collection, and broken privacy promises. Other laws can address prohibited discrimination. Each violation has elements the agency must establish. But two customers paying different prices does not, by itself, establish them.

A Price Gap Is Not a Verdict

The FTC also suggests that an undisclosed personalized price may be unfair. But Section 5(n) sets three requirements for unfairness. The practice must cause or be likely to cause substantial consumer injury. Consumers must be unable to avoid that injury reasonably. And benefits to consumers or competition must not outweigh it. The Commission must address all three, even when a higher price is easier to spot than a discount that never gets offered.

Start with injury. If one customer pays $10 while another gets a $3 coupon, the first has not necessarily suffered a $3 loss. Without the coupon, the seller might have charged both $10, sold fewer units, or offered a different product. Competition might instead have pushed the public price lower. The FTC needs evidence about what would likely have happened. It cannot simply use the lowest price it sees as the benchmark.

Whether buyers could avoid an injury depends on their options. Someone who sees the price before buying may be able to compare offers, find a coupon, wait, or walk away without knowing how the seller calculated it. Those choices may be limited in a particular market, especially if a seller misleads buyers or locks them in. The FTC must examine the choices consumers actually had. Not knowing a seller’s pricing model does not, by itself, make the price unavoidable.

The third requirement brings the less visible effects into view: discounts received, purchases made possible, additional sales, and competition for price-sensitive buyers. The costs of a disclosure or redesign requirement matter, too. The 11th U.S. Circuit Court of Appeals’ decision in FTC v. Corpay Inc. shows what a concrete unfairness case looks like. The court examined false promises and unexpected, unauthorized fees, applying the statutory requirements to the evidence. It did not hold that a difference between customers’ prices establishes injury.

Missing Sales Have No Lobby

An Austrian account asks what a pricing rule might prevent buyers and sellers from discovering. Public Choice, by contrast, asks who will seek the rule, who will bear its costs, and what an agency can learn from the information it receives. James Buchanan urged economists to analyze politics with the same view of human behavior they bring to markets. Officials respond to incentives, too. They also work with limited information.

George Stigler’s theory of regulation helps explain why firms that stand to gain from a rule may work harder to shape it than the many consumers who each bear only a small cost. An established firm might welcome restrictions on a smaller rival’s targeted discounts. A large platform may absorb a complex disclosure requirement more easily than a new entrant. An incumbent with loyal customers may favor a rule that makes introductory offers more costly. These are all risks, not allegations about anyone involved in the FTC’s proceeding. A rule born of sincere consumer-protection concerns can still favor established firms.

The political asymmetry is magnified by what can be observed. A customer who learns that someone else paid less can complain and show the price difference. A customer who would have bought only with a discount that was never offered leaves no receipt. Existing firms can document compliance costs and press for a particular standard. Potential entrants and would-be customers have less opportunity to make their case. If the agency counts visible price gaps but misses sales that never happen, it will get a distorted picture of the rule’s effects.

That imbalance can also shape how the FTC measures success. It can count investigations opened, disclosures added, and price gaps closed. It would struggle to count the discounts that firms decided against offering or the pricing methods they never tested. This is a reason to require the agency to identify a misleading claim, a consumer injury, and a plausible account of what would have happened otherwise. Public Choice helps most when it improves the rules officials must follow, rather than speculating about their motives.

Gordon Tullock identified a further cost: Firms may spend real resources seeking advantages through government rules. His work on bureaucracy also shows how information gets simplified as it moves up an organization. To enforce a broad standard based on consumer expectations, the FTC would need to sort varied pricing systems into manageable categories. Those categories could leave out the local facts that determine whether an offer helps or harms buyers.

The concern is more specific than a charge of “regulatory capture.” A vague duty to explain prices creates uncertainty that firms with large legal teams can manage more easily. Complaints about visible price differences reach the FTC more readily than evidence of missed sales. The cases the agency chooses—and the settlements firms accept—then give the policy its practical meaning. Personalized pricing may remain legal on paper even as sellers stop testing offers that could have helped some buyers.

Hayek described competition as a way to discover information no one already possesses. When a rule deters an experiment, no one sees the offer a seller might have made, the buyer who might have accepted it, or a rival’s response. The FTC cannot recover those missing facts from a case file later. Regulatory restraint is not indifference to consumers; it is recognition that the method of protecting them can destroy information about how best to serve them.

A Discount Needs No Alibi

The FTC can enforce existing deception law without presuming that sellers must explain every personalized price. Its final statement should say that personalized pricing, by itself, is neither deceptive nor unfair. Nondisclosure may be deceptive when a seller’s words or the circumstances give reasonable buyers a materially false impression that everyone can get the same price. The FTC should prove that impression with evidence from the case, rather than assume it from a generalized claim about consumer expectations.

For unfairness, the agency should ask what prices, availability, sales, and quality would likely have looked like without the challenged practice. It should examine the choices consumers actually had and weigh benefits to consumers and competition, as Section 5(n) requires. A genuine targeted discount at or below a price generally available at the same time on similar terms should ordinarily face a strong presumption against enforcement. That protection would not cover a fictitious reference price, a false claim about eligibility, an unauthorized charge, or conduct that violates privacy or antidiscrimination law.

When a disclosure is needed to correct a misleading impression, its scope should match the claim. “Prices may vary” could tell buyers more than a technical list of data categories. The FTC could test notices to see whether consumers understand them. It should also distinguish prices tailored to a particular buyer from changes driven by time, inventory, or broader demand—and remember that coupons and loyalty offers have long tailored prices to customers.

The Commission acknowledges that it cannot ban personalized pricing outright. But it also should not make the search for prices that buyers will accept legally suspect. Lies and proven injuries remain fair targets for enforcement. A different offer is not, by itself, either one. And while a sale that never happens leaves no receipt, it still counts when it comes time to weigh costs and benefits.

The post The Case of the Missing Sale: The FTC’s Personalized-Pricing Blind Spot appeared first on Truth on the Market.

By contrast, a dynamic approach to antitrust can enforce the law vigorously while also recognizing that competition changes over time and that government officials rarely know enough to redesign complex markets. The practical questions are which restraints deserve priority, what evidence should justify intervention, and how the law can leave room for the experiments that produce tomorrow’s competitors.

I propose two commitments.

First, enforcement agencies and policymakers should pay closer attention to government-created barriers to entry. Licensing restrictions, certificate-of-need laws that require approval before a provider can open or expand, discriminatory subsidies, procurement preferences, and legal immunities can all keep would-be competitors out.

Second, enforcers should judge private conduct by its likely effects on innovation and the competitive process. That means considering firms’ capabilities and the costs of mistaken enforcement decisions, rather than treating today’s market concentration figures as the last word. Cartels, anticompetitive mergers, and exclusionary monopolization still warrant action. But size, integration, and commercial success alone do not establish harm.

When the Referee Builds the Moat

The first commitment calls for greater attention to government restraints on competition. Antitrust usually focuses on private firms, but a legal rule can suppress rivalry more durably than a powerful incumbent can. No matter how entrenched incumbents may appear, customers can always switch, new firms can enter, and new technologies can displace even a dominant company. Licensing rules, on the other hand, can shut qualified entrants out entirely. Certificate-of-need laws can stop new facilities from opening. Procurement preferences can deny challengers the customers they need to grow, while subsidies can keep inefficient firms afloat.

Government restraints also carry other costs beyond higher prices and reduced output. When officials grant competitive privileges, firms have reason to spend on lobbying, litigation, and regulatory strategy instead of on better products and lower costs. Economists call this rent seeking: using resources to secure a government-granted advantage. Gordon Tullock showed how the contest for such advantages can consume resources that could have gone to productive innovation. Daniel Sokol’s analysis of anticompetitive government intervention likewise makes the case for scrutinizing public restraints that benefit organized interests.

Public-choice analysis—the study of incentives in politics and government—adds further warnings. Legislators, regulators, and prosecutors respond to organized constituencies, career incentives, and the appeal of headline-generating cases. The design of competition policy should account for such pressures. Agencies should identify public barriers, assess their likely effects on productivity, and focus their advocacy on restraints with broad economic consequences that are difficult to undo. Courts, too, should continue to read antitrust immunities narrowly, as the Supreme Court has long instructed.

Tomorrow’s Rivals Aren’t in Today’s Spreadsheet

The second commitment concerns how antitrust judges private conduct. Market shares, prices, and firm counts all matter, but they can only ever capture competition at a specific moment in time. Firms also compete by experimenting with new technologies, ways of organizing work, distribution systems, and business models. Joseph Schumpeter’s account of creative destruction, Friedrich Hayek’s description of competition as a discovery procedure, and Israel Kirzner’s emphasis on entrepreneurial alertness share the same lesson: some of the most important competitive possibilities have yet to show up in the data.

A capabilities approach to market analysis asks what firms can do next. Which have the technical knowledge, financing, management, and other assets to develop a product or become a serious rival? David Teece’s work on dynamic and potential competition helps answer that question by looking beyond today’s market shares.

This view also changes how we weigh error costs. A false positive—stopping conduct that would not harm competition—might serve to block a merger or business practice that could otherwise produce a new product, speed its launch, or combine firms’ strengths. These effects can, in turn, reach beyond the parties before a court: uncertainty about liability may discourage investment and experimentation across an industry.

To be sure, a false negative—allowing harmful conduct—is costly too. In some changing markets, however, new entry, imitation, technological change, or later enforcement may work to check that harm. Judge Frank Easterbrook’s error-cost analysis reminds us that a rule that courts can apply reliably may serve competition better than elaborate but unreliable predictions they cannot.

Innovation should not be considered a free pass from antitrust law. Firms can buy or shut out rivals to protect lasting market power, and claims about future benefits deserve scrutiny. But plaintiffs should have to show a probable threat to competition, rather than simply point to harm to a competitor or objections about a firm’s design choices. At the same time, courts should take potential gains in quality, security, integration, speed, and innovation seriously when the alleged harm also depends on a prediction about the future.

The Judge as Product Manager

Digital platforms show why this approach matters in practice. Search rankings, interoperability rules that let products work together, privacy defaults, app-store policies, bundles, and data access all involve tradeoffs among relevance, safety, congestion, revenue, and investment. A rule requiring “neutrality” or equal treatment may sound simple, but enforcing it could weaken curation, security, or the incentives to build tools for the platform. Antitrust should neither impose a general duty to give rivals access nor make courts permanent supervisors of product design. Antitrust liability should instead rest on strong evidence that the conduct will harm rivalry after accounting for its benefits to quality and integration.

The same care applies when a firm takes on another stage of production or distribution, known as vertical integration. Combining those stages can eliminate double marginalization—the extra markups that arise when firms at successive stages each set their own prices. It can also help firms coordinate investment, protect assets built for a particular business relationship, reduce the costs of contracting, and improve quality control.

The relevant comparison is not between an integrated firm and an imaginary world of frictionless contracting. It’s between an integrated firm and realistic institutional alternatives, each with costs of its own. Assuming firms could achieve the same benefits through effortless contracts commits the Nirvana fallacy: judging a workable arrangement against an unattainable ideal. It also overlooks what courts would need to know, and the administrative burdens they would have to bear, in order to impose that ideal. Herbert Hovenkamp’s discussion of antitrust and the design of production explains why competition law should approach that task with care.

Don’t Forget the Next Competitor

Merger review should also look beyond concentration screens that measure how much of a market the merging firms already control. Instead, it should ask: what capabilities does each bring? Could the combination turn an invention into a product or bring together assets that work better together? How many plausible future rivals would remain? And does the likely harm outweigh gains across the economy, including benefits outside the market under review?

Review should also account for the market for corporate control. An acquisition may put an invention in the hands of a firm better able to manufacture, distribute, or scale it, or to navigate regulatory hurdles. Indeed, for startups, the prospect of a future sale can help attract investment before they enter the market. Treating every acquisition by a successful incumbent as suspect could discourage the very competitive experiments that antitrust should aim to protect.

That does not give incumbents a pass to buy emerging rivals. But it does mean that the government should have to establish lasting harm that outweighs a merger’s benefits, rather than infer harm from size alone. Research on innovation competition in merger policy shows why a potential rival need not already sell a competing product for its assets, know-how, or credible path into the market to nonetheless make it a meaningful competitive threat in the future.

No School Has a Monopoly on Good Ideas

This dynamic approach draws from several schools of antitrust thought without fitting neatly into any of them.

Neo-Brandeisian scholars stress that markets operate within political institutions and that private power can matter in ways a short-term price test misses. Lina Khan’s “Amazon’s Antitrust Paradox” presses that point. But size, vertical integration, and harm to rivals cannot stand in for evidence of harm to competition. Simple structural rules are especially unreliable in industries where a service becomes more valuable as users join, innovation moves quickly, and firms combine assets that work better together. Khan’s challenge deserves a serious answer grounded in evidence.

The traditional Chicago School, associated with Robert Bork and later analysis of enforcement mistakes, focuses on consumer welfare, output, and rules that courts can apply consistently. The dynamic approach I advocate here shares the Chicago School’s concern about speculative intervention. But it puts more weight on innovation, firms’ capabilities, what they learn by doing, and effects beyond the market under review. Those effects do not always show up promptly in prices or output. The result extends the consumer-welfare tradition to account for how competition develops over time.

Post-Chicago scholarship offers models of strategic exclusion, tactics that raise rivals’ costs, and competition when firms and customers lack complete information. These models can expose real harm. They can also generate several plausible stories about the same conduct. Can a court reliably tell which story describes the market? Do agencies have incentives to sound more certain than the evidence allows? Those are questions of institutional judgment, not indifference to harm. The dynamic approach favors room for firms to experiment, backed by legal rules against demonstrable coercion and exclusion.

Enforce the Law. Leave Room for Tomorrow.

A dynamic antitrust framework would seek to discipline enforcement, rather than abolish it. Government-created barriers deserve a central place in competition policy because the law can make them especially hard to dislodge. Against private firms, enforcers should pursue cartels and clear cases of exclusion vigorously, while demanding persuasive evidence before condemning integration, innovation, or corporate reorganization. They should ask what future rivals and capabilities an intervention might foreclose just as much as they focus on what today’s market shares reveal.

The result is neither industrial planning nor institutional neglect. The aim is to protect the conditions in which rivalry can develop. Markets are imperfect, but so are the remedies available to judges and agencies. Institutional humility calls for using coercive power against proven restraints while still leaving room for entrepreneurs to adjust and experiment. In the long run, competition can discover opportunities no regulator can reliably identify in advance.

The post The Rivals Antitrust Can’t See appeared first on Truth on the Market.

Some object that bringing law & economics into marriage risks coarsening and cheapening this fundamental institution. That concern is noble, but it overlooks history and mistakes scholarship for advice. Marriage’s deep association with romantic love is itself partly the result of a legal change: the medieval church’s adoption of dispositive spousal consent, under which the spouses’ consent determined whether a marriage was valid (Berman 1985, Piano and Piano 2025). Far from being removed from legal analysis and price theory, marriage’s beauty and complexity cannot be fully appreciated without these tools.

Although most couples view marriage as more than a contract, it is at least a contract with both economic and legal purposes. Saying that marriage has an economic purpose does not mean that it maximizes wealth, although it usually does. It means that marriage solves coordination problems between two people.   

Likewise, saying that marriage has a legal purpose does not mean that the state must regulate it, although it usually does. It means that marriage solves coordination problems between the couple and the rest of society. In “The Economics of the Marriage Contract: Theories and Evidence” (2008), Niko Matouschek and Imran Rasul ask: What exactly are the coordination problems that the marriage contract solves? 

Why People Marry and What Divorce Law Reveals

The literature offers three main hypotheses for why people marry: 

1. Marriage generates a joint surplus—the surplus hypothesis. 
2. Marriage signals a person’s quality as a companion—the signaling hypothesis.
3. Marriage helps people keep their promises and rely on their partners to do the same—the commitment-device hypothesis. 

Matouschek and Rasul test these theories by examining changes in the likelihood of divorce after states adopted unilateral-divorce laws, which allow one spouse to obtain a divorce without the other’s consent. The fact that divorce rates changed significantly after these laws took effect suggests substantial transaction costs within marriage, as the authors note. These costs prevent Coasean bargaining—the private negotiation that might otherwise allow couples to adjust to the legal change without divorcing. Law & economics scholars have spilled considerable ink explaining why these costs arise, e.g., Becker, Landes, & Michael (1977), Cohen (1987), and Brinig (1993). 

In every model of marriage, changing the cost of divorce affects both couples who are already married and the types of couples who later choose to marry. The three leading hypotheses—marriage as surplus, signal, or commitment device—make the same prediction for existing marriages. After divorce becomes less costly, the likelihood of divorce initially rises. Over time, it falls because poorly matched couples have already separated, leaving a pool of more stable, higher-quality marriages. 

But the models diverge in their predictions about who marries after divorce becomes less costly. In other words, the characteristics of couples who marry under the new rules reveal what the marriage contract does for them. 

If marriage credibly signals an underlying quality or gives spouses benefits that exist independently of their commitment, reducing the cost of divorce makes marriage less risky by lowering the chance of becoming trapped in a bad match. The benefits of marriage remain unchanged. Under either hypothesis, the share of the population marrying should rise after divorce laws are liberalized. Couples with relatively poor matches who would have avoided marriage when divorce was costly now choose to marry. Divorce rates should also remain elevated as these couples exercise the option to leave. 

The commitment-device hypothesis yields a different prediction. If couples use marriage to secure commitment and encourage relationship-specific investments, both the marriage rate and the divorce rate among those marriages should fall after divorce becomes less costly. Together, those changes would indicate an increase in average match quality. 

Lower divorce costs weaken marriage as a commitment device, so only very strong matches clear the threshold for marriage. Couples in relatively poor or risky matches will tend to choose cohabitation instead. Even under unilateral divorce, leaving a marriage remains costlier than ending a cohabiting relationship.

It is worth pausing to explain why relationship-specific investments matter to marriage. The concept, also called asset specificity, comes from the transaction-cost literature. Benjamin Klein, Robert Crawford, and Armen Alchian (1978) explain that opportunism can arise after two parties enter a contract when one or both make investments that are more valuable within their relationship than outside it, such as building a machine for a particular buyer.

In a romantic relationship, marriage can play a role similar to vertical integration, in which two parties formally combine to protect and coordinate their joint investments. As a commitment device, marriage reduces the risk of investing in things that derive much of their value from the relationship—such as having children, taking dancing lessons, or learning to cook a spouse’s favorite meal. 

What the Evidence Says About Marriage

Using individual marriage and divorce certificates from across the United States, Matouschek and Rasul test each hypothesis by comparing outcomes before and after states adopted unilateral-divorce laws. They find that the likelihood of divorce rises by 18.5% after a state adopts unilateral divorce, then falls as relatively poor-quality matches leave the pool of married couples. This pattern is consistent with all three theories of marriage. 

Couples who married several years after a state adopted unilateral divorce, however, were significantly less likely to divorce. The authors conclude: “The sign of this selection effect implies that the first-order purpose of the marital contract is to serve as a commitment device” (Matouschek and Rasul 2008, p. 93). 

Matouschek and Rasul also find that the effects vary with states’ labor markets and social characteristics. They examine several measures of women’s bargaining power within marriage, including the female labor-force-participation rate. In states where women have relatively greater bargaining power, couples are less likely to divorce with each additional year under unilateral divorce.

The authors suggest that this finding may offer further evidence of positive assortative mating—the well-documented tendency of people to marry partners with similar characteristics. Cooperation between these spouses may come more naturally and therefore respond less to changes in divorce law. 

The authors also examine differences in the Catholic share of each state’s population. In states with an above-average Catholic population, couples become even less likely to divorce over time. Matouschek and Rasul interpret this finding as evidence that Catholics are more likely than non-Catholics to use marriage as a commitment device. 

The Enduring Value of Marital Commitment

Most theories of marriage struggle to explain what couples gain from it beyond a division of labor and economies of scale, both of which cohabitation can also provide. Matouschek and Rasul’s distinctive contribution is to separate the competing theories of marriage and test their different predictions against high-quality data. More recent studies, including Edlund (2013) and Lafortune and Low (2017), provide further evidence that the marriage contract’s value—from the perspective of couples themselves—depends on its ability to secure commitment from both partners. 

Law & economics scholars have made substantial contributions by turning their attention to marriage. Yet each answer raises more questions. Who benefits from prenuptial agreements (Leeson and Pierson 2016)? How could voting systems better represent children’s interests (Kleinfield and Sachs 2025)? Do policies designed to encourage childbirth distort marriage markets (Baudin and Stelter 2026)?

Far from debasing marriage, scholarly inquiry reveals an institution even more compelling and mysterious than it first appears. We need not fear ever reaching the bottom of it. 

Further Reading

• Niko Matouschek and Imran Rasul, “The Economics of the Marriage Contract: Theories and Evidence,” The Journal of Law and Economics, Vol. 51, No. 1 (2008).
• Gary S. Becker and Kevin M. Murphy, “The Family and the State,” The Journal of Law and Economics, Vol. 31, No. 1 (1988).
• Lena Edlund, “The Role of Paternity Presumption and Custodial Rights for Understanding Marriage Patterns,” Economica, Vol. 80, No. 320 (2013).
• Matthias Doepke and Fabian Kindermann, “Bargaining over Babies: Theory, Evidence, and Policy Implications,” American Economic Review, Vol. 109, No. 9 (2019).
• Jeanne Lafortune and Corinne Low, “Collateralized Marriage,” American Economic Journal: Applied Economics, Vol. 15, No. 4 (Oct. 2023).
• Joshua Kleinfeld and Stephen E. Sachs, “Give Parents the Vote,” Notre Dame Law Review, Vol. 100, No. 3 (2025).
• Clara Piano and Ennio Piano, “The Medieval Origins of Spousal Consent,” SSRN Electronic Journal (Sept. 16, 2025) 

The post ‘The Economics of the Marriage Contract: Theories and Evidence,’ by Niko Matouschek and Imran Rasul appeared first on Truth on the Market.

The final rule is only a partial repeal. It eliminates standards for existing fossil-fuel-fired steam units, along with several carbon-capture requirements for coal plants and new baseload combustion turbines. A separate supplemental proposal contemplates a broader rollback. Published Sept. 17 in the Federal Register, the final rule takes effect Nov. 16.

The central economic point is that a regulation that forwards a legitimate public-policy goal may nonetheless also skew competition. This happens when the rule imposes costs on some technologies that their rivals do not bear, steers investment toward politically favored alternatives, or raises barriers for firms too small to absorb the compliance burden.

The power-plant rule did all three. Its repeal could therefore improve economic welfare by putting generating technologies on more equal footing, reducing wasted resources, and giving cost and reliability greater weight in deciding how to power the grid.

The EPA’s Power Problem

The strongest legal defense of the repeal begins with the limits that Congress placed on the EPA’s authority. The Clean Air Act (CAA) does not clearly authorize the agency to restructure the national electricity system through standards that rely on inadequately demonstrated technologies, impose extraordinary costs, or effectively require power generation to shift among energy sources. Recent Supreme Court decisions lend substantial support to that argument.

Section 111 of the CAA, codified at 42 U.S.C. § 7411, directs the EPA to identify the “best system of emission reduction,” taking into account cost, environmental effects beyond air quality, and energy requirements. . In its final rule, the EPA concluded that capturing and storing 90% of a plant’s carbon emissions—a process known as carbon capture and storage (CCS)—was inadequately demonstrated, unreasonably expensive, and unachievable within the compliance period. The necessary capture, pipeline, and underground-storage infrastructure simply does not exist at the required scale.

The EPA reached a similar conclusion about natural-gas co-firing, which involves burning natural gas alongside another fuel. The agency found that the practice would use natural gas inefficiently and, in this context, amount to impermissible generation shifting. These are statutory judgments about technology, cost, and energy requirements, rather than a political endorsement of fossil fuels.

The Supreme Court’s decision in West Virginia v. EPA provides the most important backdrop. The Court held that Section 111(d) did not authorize the EPA to set carbon dioxide limits that would force a nationwide transition away from coal. Under the “major questions doctrine,” agencies need clear congressional authorization to exercise powers of vast economic and political significance.

The 2024 standards were less explicit than the Clean Power Plan. Even so, their reliance on CCS and co-firing remains vulnerable to the charge that EPA used technical standards to bring about a transition Congress never clearly authorized the agency to order directly.

The repeal also benefits from the demise of Chevron deference, under which courts generally deferred to reasonable agency interpretations of ambiguous statutes. In Loper Bright Enterprises v. Raimondo, the Supreme Court held that courts must exercise their own judgment when interpreting statutes. Section 111’s limits therefore carry greater weight, especially when an agency’s interpretation could transform the electricity market.

The final rule’s opponents will rely on Massachusetts v. EPA, the 2007 decision holding that greenhouse gases can qualify as air pollutants under the CAA. They will also argue that the EPA failed to explain its change in position or adequately address serious reliance interests—the investments and other commitments parties made based on the former rule.

Those arguments make litigation inevitable, and the repeal is hardly bulletproof. Yet Massachusetts v. EPA does not require the U.S. Court of Appeals for the D.C. Circuit to preserve any particular emissions standard. The EPA has offered a technology-specific explanation, relied on the statute’s cost and achievability factors, and retained other conventional pollution controls. That record gives the court a plausible basis to uphold the repeal while leaving the broader fight over the EPA’s endangerment finding—its determination that greenhouse gases threaten public health and welfare—for another day. The related litigation is being tracked in Commonwealth of Massachusetts v. EPA

The Distortion Is Coming From Inside the Border

The economic case for the final rule is made clearer by the concept of anticompetitive market distortions (ACMDs). Developed prominently by Shanker Singham and applied in work we’ve done together, the framework examines how governments can reduce economic welfare through “behind-the-border” policies—domestic rules that distort competition without taking the familiar form of tariffs or import quotas.

These policies can protect incumbents, raise rivals’ costs, block new entrants, weaken property rights, or discriminate against foreign competitors. Their official label—whether they are environmental, industrial, or trade policy—tells us little. What matters is whether a policy preserves equal competitive opportunity or lets the government decide who may compete and on what terms. I summarized the framework in this Truth on the Market analysis of behind-the-border barriers and Singham and I expounded further in the Mercatus Center’s public-interest comment on anticompetitive regulations.

The framework rests on three related pillars. First, strong property rights give firms confidence that the government will not arbitrarily confiscate or disable their investments, technologies, contracts, or productive assets. Second, domestic competition works best when firms compete through price, quality, innovation, and reliability rather than political privilege. Third, international competition requires that foreign firms have a reasonable opportunity to enter and compete.

A single rule can weaken several pillars at once. A technology mandate may suppress domestic rivalry, steer capital away from more productive uses, and leave American producers at a disadvantage against firms operating under very different regulatory regimes.

The ACMD framework still leaves ample room for sound regulation. A carefully designed, competitively neutral rule may increase welfare by correcting a genuine market failure. The framework also recognizes that carbon emissions can impose “external costs”—harms borne by people other than those producing or consuming the electricity.

The real question is more exacting: Does this particular rule improve how society allocates its resources? Or does it use administrative power to favor selected technologies while saddling consumers and competing producers with avoidable costs?

How to Tilt a Power Market in Three Easy Steps

The repealed power-plant requirements fit the ACMD framework in three ways.

First, the rule suppressed competitors by regulatory category. Coal, natural gas, nuclear power, renewables, and emerging technologies could not compete based on cost, capacity, reliability, and environmental performance. Instead, the rule made particular technologies—especially large-scale CCS and natural-gas co-firing—conditions of continued operation or market entry. That is competition by regulatory classification, rather than competition on the merits.

The distortion becomes especially costly in an industry built around long-lived assets. A generating unit may operate for decades. When a rule makes continued operation contingent on an expensive technology before that technology is commercially ready, it immediately reduces the asset’s expected value. Investors must then choose among premature retirement, a costly retrofit, and abandonment of the project. The result may include lower emissions, but it also means a less flexible electricity supply and less experimentation with competing mixes of generation technologies.

Mandates can similarly narrow the path of innovation by freezing the regulator’s preferred technology in time. Energy markets, like other markets, advance through trial, error, and discovery. Natural-gas turbines may grow more efficient. New nuclear designs may become cheaper. Energy storage may improve. Carbon capture may prove viable for a particular industrial use. A technology that barely registers today may solve tomorrow’s reliability problems at a lower cost.

Making one pathway a condition of market participation reduces the payoff from discovering alternatives. Repeal allows firms to test different combinations of generation and emissions reduction instead of asking the EPA to pick the winning portfolio years in advance. Regulators have many talents. Crystal-ball gazing is rarely among them.

Second, the mandate created a regulatory moat around large, politically connected, and well-capitalized firms. A nationwide utility or conglomerate may have the resources to finance engineering studies, secure permits, negotiate infrastructure contracts, and spread compliance costs across a broad ratepayer base. Smaller independent power producers, municipal utilities, and new entrants may struggle to do the same. Compliance costs can therefore block entry and expansion even when a rule appears neutral on its face.

This is the familiar political economy of regulation. Firms that can afford compliance may support rules that hobble less-established rivals. Incumbents turn regulatory complexity into a competitive advantage, while policymakers present the resulting concentration as a technological necessity. Repeal closes one channel for rent-seeking—the use of government power to secure private advantage. It cannot guarantee a competitive electricity market, but it makes environmental compliance less useful as a protective moat.

Third, the mandate placed American producers at an international disadvantage. U.S. firms would have borne technology and capital costs that many foreign competitors—including state-backed companies operating under different budget constraints and regulatory systems—might avoid.

To be clear, domestic deregulation alone cannot eliminate foreign ACMDs. Chinese state-owned enterprises, foreign subsidies, and protectionist local-content rules will continue to distort energy and industrial markets. But loading another self-inflicted burden onto U.S. producers is a poor answer to those distortions.

May the Best Megawatt Win

The EPA estimates that the partial repeal will save $160 billion in compliance costs from 2026 through 2047 using a 3% discount rate, and $95 billion using a 7% rate. A discount rate converts future costs and benefits into today’s dollars. The higher the rate, the less weight the calculation gives to costs incurred far in the future.

The agency puts the avoided real-resource costs even higher: $280 billion at a 3% rate and $180 billion at a 7% rate. These are the physical inputs and labor that firms would otherwise devote to compliance.

Those figures are agency estimates, rather than guaranteed savings on consumers’ electric bills. Still, they identify the right economic mechanism. Resources once earmarked for mandated equipment and regulatory compliance become available for generation, transmission, storage, grid reliability, and other uses consumers may value more.

The competitive gains may matter even more than the immediate accounting savings. Repeal allows owners to keep existing plants available as new technologies develop. It also expands the range of technologies that can bid into wholesale electricity markets and reduces the chance that one politically favored technology becomes a single point of failure.

More supply options become especially valuable as electricity demand rises because of manufacturing, electrification, and data centers. Reliability is a basic economic input. When it grows scarce, the costs can be enormous.

Deregulation also allows markets to produce better information. If CCS makes economic sense, power producers can adopt it in response to fuel prices, local pollution rules, tax incentives, customer demand, or contractual commitments. If it cannot compete on those terms, a mandate can hide that weakness by compelling investment and sending the bill to ratepayers or taxpayers. Voluntary adoption makes both success and failure more informative, which helps markets discover which technologies actually work.

A measured local externality also differs from a global problem addressed through mandates for individual facilities. Regulators can often link conventional pollutants such as sulfur dioxide and nitrogen oxides more directly to local health effects and address them through source-specific controls. Carbon dioxide presents a harder problem. Its effects are global, each plant’s marginal contribution is diffuse, and the choice among emissions-reduction methods can reshape the entire electricity system.

Carbon policy remains possible, but broad technology mandates carry unusually high error costs—the harm caused when regulators choose the wrong solution. They also create a public-choice problem by inviting political incentives and organized interests to influence how an agency allocates capital across the economy.

Repeal does not automatically tilt the market toward coal or gas. A competitive market may continue to favor renewables, nuclear power, storage, or efficiency improvements wherever they offer better value. The point is that the outcome should reflect resource costs and consumer preferences instead of serving as a regulatory penalty imposed on one class of assets. Removing an anticompetitive rule gives consumers a greater chance to benefit from rivalry among energy sources. It does not choose the winner for them.

Even Deregulation Has a Price

To be sure, repeal carries costs and risks of its own, and a market-oriented analysis should acknowledge them. Greenhouse-gas emissions may impose external costs that never appear on a generator’s balance sheet. State regulation, federal controls on conventional pollutants, permitting rules, and tax subsidies will also continue to shape competition.

Repeated policy reversals create another cost: uncertainty. If each administration switches between mandates and repeal, investors will demand a risk premium—additional compensation for accepting political risk. Regulatory whiplash has a price.

These concerns favor durable, generally applicable rules, which doesn’t mean regulators must preserve a technology mandate simply because reversing it creates disruption. Congress can legislate clearly if it wants the EPA to regulate greenhouse gases in a particular way. States can address local pollution and reliability concerns within the boundaries of federal law. Policymakers can also use market-based tools that price genuine external harms while allowing producers to choose how to reduce them.

The proper benchmark is not zero regulation. The framework permits regulation where its benefits exceed its costs. It also favors the lawful approach that distorts competition the least.

The international dimension demands similar care. Removing a domestic ACMD while leaving foreign distortions untouched may expose U.S. exporters to carbon-border measures—fees or restrictions imposed on carbon-intensive imports—or discriminatory standards abroad. The better response is negotiation aimed at reducing behind-the-border distortions across countries. Another round of subsidies and protectionism would merely add new distortions to the old ones.

That broader view is the ACMD framework’s chief virtue. It examines the entire competitive field, including domestic regulation, foreign state aid, property rights, market access, and the incentives each creates.

Repeal may also reduce regulatory fragmentation—the patchwork created by overlapping or conflicting rules. A single federal standard can sometimes override inconsistent state requirements. Yet a federal mandate resting on shaky statutory ground invites litigation, policy reversals, and more uncertainty.

States will continue to make different choices about electricity generation and environmental protection. But firms can plan more rationally once the federal government ceases to condition national market participation on a disputed technological pathway. It’s important to remember that stable legal boundaries are themselves a competitive asset.

Pulling the Plug on Distortion

The EPA’s repeal of the power-plant greenhouse-gas requirements should withstand judicial review. The agency has offered a legally grounded and technically specific explanation. Section 111 does not clearly authorize the EPA to transform the electricity system, and the technologies the rule mandated were inadequately demonstrated, unreasonably costly, or unachievable on its timetable.

The economic case points in the same direction. The repealed rule distorted competition by raising rivals’ costs, favoring selected technologies, erecting entry barriers, and placing American producers at a regulatory disadvantage in an already distorted global market.

Repeal leaves the climate debate unresolved. The final rule cannot eliminate every externality or guarantee lower electricity prices. Its benefits are more modest and more defensible. It restores room for competition, experimentation, and consumer choice while allowing capital and labor to flow toward more valuable uses.

Removing regulatory distortions lets resources flow to their most valuable uses—advancing both competition and environmental goals in an economy hungry for abundant, reliable power.

The post The EPA Takes Its Thumb Off the Power Switch appeared first on Truth on the Market.

Anthropic CEO Dario Amodei’s proposal to “pace the frontier” has become a rallying point for AI companies interested in coordinating a slowdown in capability development. His plan calls for embedded outside evaluators, common safety standards, and limits on the pace of model advances. Amodei also suggests that some form of antitrust waiver may be needed to permit certain safety discussions. 

As my colleague Dirk Auer explains, agreements among rivals to restrain development deserve scrutiny, even when made in the name of safety. Before granting an exemption, policymakers should ask where existing legal and market incentives—and the forms of cooperation antitrust law already allows—fall short. 

Civil and criminal laws already give AI labs reason to consider the consequences of their conduct, including safety risks. Market incentives point in the same direction. Customers favor products that work as intended and remain dependable. A model that enables mass hacking or causes other serious harm is, among other things, a very bad product. 

Companies can respond to these incentives on their own by strengthening safeguards, limiting model autonomy, or delaying releases. Existing antitrust law also permits competitors to collaborate on many safety matters and participate in carefully structured standard-setting efforts. 

Any proposed exemption therefore carries a heavy burden. Its proponents must show why unilateral action and lawful collaboration cannot address the problem—and identify precisely what conduct existing antitrust law prevents. 

Move Fast and Get Sued

AI-safety evaluator Model Evaluation and Threat Research’s (METR) investigation  into the Hugging Face incident found that separate agents used an unauthorized message board to coordinate a remote-code-execution attack, which allows an attacker to run commands on another computer. Hundreds of agents participated, even though the attack fell outside their assigned tasks. If developers and operators did not already feel pressure to build safeguards into new models, this incident should provide it. 

Labs that deploy agents with offensive cyber capabilities must make difficult choices about access limits, system isolation, monitoring, and when humans should intervene. Negligence law can make poor choices in these areas very costly. 

Of course, a negligence claim still requires proof of a legal duty, breach, causation, and recoverable harm. An agent’s harmful conduct does not automatically make its developer liable. Even so, the early stages of litigation can become a long and expensive headache for any firm. 

Nor does potential liability end with a lab’s direct actions. A company may face liability for creating a foreseeable downstream risk, even when a third party causes the immediate harm. In Weirum v. RKO General, for example, the California Supreme Court affirmed liability against a broadcaster whose contest predictably encouraged listeners to drive dangerously. The case did not involve artificial intelligence, but it illustrates how liability can rest on creating a foreseeable risk rather than directly inflicting the injury. For an AI lab, the central question may be whether developing or operating its system created an unreasonable risk of harm. 

Other civil claims and criminal charges apply different legal standards. The Computer Fraud and Abuse Act (CFAA), for example, covers specified forms of unauthorized computer access and provides a limited civil remedy. The statute bars civil claims based on negligent software design or manufacture. No court has yet decided how that exclusion applies when autonomous software agents commit the prohibited acts. 

The degree of “intentionality” that a court attributes to an agent could affect that analysis. So could a developer’s deliberate decision to equip a model with offensive cyber capabilities. A court might view such a case as involving something more than negligent software design. 

Criminal liability under the CFAA would require proof of a specific offense and the necessary mental state of a legally responsible person or entity. An agent’s apparent motivations do not, by themselves, establish a lab’s criminal intent. It also remains unclear whether an agent can possess legally cognizable intent and, if so, when that intent could be attributed to the lab that trained it or the user who deployed it. 

None of this means that OpenAI or any other lab should be sued, much less that it would clearly face liability under the CFAA, negligence law, or another legal theory. The point is that considerable uncertainty remains, along with plenty of room for creative lawyers to impose costly, time-consuming litigation on frontier AI labs. Those labs can reasonably account for that risk when deciding unilaterally whether to slow development. 

Development and deployment also create different risks. A lab might continue capability research while restricting an agent’s access to external systems. It might strengthen containment measures or postpone a release until testing is adequate. Each decision involves a separate judgment about the product’s risks and benefits. A race to develop more capable models does not force every participant to ignore those costs in pursuit of commercial advantage. 

Acing the Benchmark, Flunking the Job

An AI product’s quality depends on more than intelligence. Users expect an agent to respect permissions, protect confidential data, and perform assigned tasks reliably. Benchmark scores cannot establish whether it is fit for a particular job. An agent that aces every test but ignores its instructions is still a lousy employee. 

Competition law knows this territory well. Firms routinely compete through different combinations of performance, price, reliability, and other features. Antitrust law recognizes these dimensions of competition. Agency guidance on innovation, for example, expressly considers competition over the range of products and features available to consumers. Intelligence is similarly just one measure of competitive success in AI. 

Safety also has commercial value independent of any shared commitment to slow development. Widespread concern about AI safety is itself a strong signal that customers want safer products. One lab might attract customers with an agent that is easier to supervise or better suited to sensitive work. Another might offer greater autonomy paired with different controls. 

Competition lets customers compare these approaches and firms refine them through experience. Coordination around a single approach to AI safety can limit that experimentation, particularly when an agreed-upon standard excludes competing designs. 

Mark Zuckerberg’s recent discussion of Meta’s strategy for Muse illustrates the point. Zuckerberg said Meta delayed Muse’s launch to address safety concerns. Meta made that decision unilaterally as part of its own product-design and launch process, without imposing the same judgment on its competitors.

The delay does not prove that Meta ultimately met the appropriate safety threshold or that every lab faces comparable incentives. It does show that firms can delay a release to address safety concerns without first securing an antitrust exemption.  

No Antitrust Hall Pass

AI labs can already obtain safety information through several channels. Each firm can hire independent evaluators, commission audits, and consult outside experts. Anthropic’s unilateral pledge to embed independent evaluators shows that this part of Amodei’s proposal requires no industrywide speed limit. 

Standardized testing and certification could also help consumers distinguish among competing products. They could provide courts and regulators with evidence of an established, reasonably applied industry practice. Private certification bodies might assess compliance with specified benchmarks, much as UL Solutions tests and certifies product safety. Testing laboratories could also work through standard-setting forums to develop common protocols and technical specifications. 

The Federal Trade Commission (FTC) already recognizes that carefully structured cooperation among competitors can make markets more efficient and benefit consumers. That includes certain standard-setting activities. 

The legal details still matter. An industry standards body can promote safety, or it can become a club that keeps rivals out. A forum created to develop testing protocols can instead become a venue for sharing confidential business information or coordinating anticompetitive restraints. A safety label is not an antitrust hall pass. Certification programs and standards remain subject to antitrust scrutiny, and participating firms must act reasonably within them. 

The law also distinguishes among unilateral decisions, limited cooperation, and agreements that suppress competition. One firm may delay a launch. Several firms may collaborate on a specific evaluation. Competitors might instead agree to restrict the computing workloads used to train their models. The first two approaches preserve competition over capabilities. The third can suppress it. 

Advocates for an antitrust exemption should therefore identify the conduct they want to pursue that existing law prohibits. They should also explain why that protection is necessary to achieve safety benefits that firms cannot obtain through unilateral action or carefully structured cooperation. 

The Burden Before the Brake

Liability rules and customer demand cannot solve every safety problem. Customers may overlook harms imposed on outsiders. An incident may be difficult to detect or trace to a particular company. Catastrophic losses may also exceed a lab’s ability to compensate victims. These gaps can weaken incentives to take precautions and may justify targeted policy responses. 

The narrower point is that existing legal and market institutions already give firms independent reasons to invest in safer products. Firms also have several ways to act on those incentives. Anyone proposing coordinated restrictions as the cure must explain why they would close the remaining gaps more effectively than the available alternatives. 

As Dirk Auer noted, an antitrust exemption carries costs of its own. Past exemptions have encouraged stagnation and industry capture, allowing established firms to shape rules for their own benefit. Proponents must therefore identify the specific risk that unilateral action and lawful cooperation cannot adequately address, the precise restraint needed to address it, and why a less restrictive response would fail. 

Build the guardrails first. Give competitors a shared brake only after proving the need.

The post Tap Your Own Brakes: AI Safety and Antitrust appeared first on Truth on the Market.

The DSA designation puts ChatGPT alongside Google Search and Bing in the category reserved for the largest search engines operating in the EU. Under Article 33, a platform or search engine enters that category when it reaches at least 45 million average monthly active users in the EU. OpenAI reported that ChatGPT Search averaged about 159.1 million monthly active users in the EU during the six months ending March 31, comfortably clearing the threshold. ChatGPT now has four months to comply with the additional obligations the DSA imposes on services of that size. 

Those obligations require the largest services to examine their broader effects on society. Article 34 requires them to identify and assess “systemic risks” arising from their design or operation. These include risks to fundamental rights, civic discourse, public security, elections, public health, and children. Article 35 then requires services to adopt reasonable, proportionate, and effective measures to mitigate those risks. 

The DSA offers several examples, including changes to a service’s design, recommendation systems, terms and conditions, and content-moderation practices. Lawmakers largely developed these obligations for platforms and conventional search engines. A search engine generally organizes and presents information found elsewhere. ChatGPT, by contrast, generates the answer a user sees. 

That difference has practical consequences. Efforts to reduce systemic risks may extend beyond how a service ranks, recommends, or displays information. They may shape the substance of the answer itself. 

OpenAI has accepted the designation and emphasized its commitment to complying with the DSA while balancing access to useful information with protection from harmful and illegal content. Few would quarrel with that goal. The difficulty lies in deciding what the balance requires. 

ChatGPT’s designation therefore presents a broader problem. A regulatory framework designed to govern the distribution of online information may fit awkwardly when applied to a service that produces the answers users receive. 

When a Search Engine Talks Back

The DSA defines an “online search engine” broadly. Under Article 3(j), the term covers an intermediary service that lets users enter queries to search all websites or all websites in a particular language, then returns information related to the requested content. The definition turns on function and expressly permits results “in any format.” 

ChatGPT Search appears to qualify. It can search the internet for relevant information and generate a response based on what it finds. A conversational answer can fall within Article 3(j) as readily as a page of links. 

But this shared label covers materially different services. Google Search, Bing, and other conventional search engines identify, organize, and rank existing information, leaving users to choose among sources. ChatGPT retrieves information and composes the response. Article 3(j) can therefore cover both a service that points users toward answers and one that provides the answer itself. 

The Regulator’s Answer Key

The distinction changes how the DSA’s systemic-risk duties operate. Every ChatGPT response reflects choices about relevance, emphasis, qualifications, and the best answer to the user’s question. Information the model omits may disappear entirely instead of slipping lower on a results page. 

The DSA’s systemic-risk categories are deliberately broad. Protecting children and restricting illegal content may lend themselves to relatively identifiable measures. Other categories—including fundamental rights, civic discourse, elections, public health, and well-being—require judgments about matters on which reasonable people, institutions, and experts disagree. 

Conventional platforms can often mitigate such risks through changes to ranking, recommendation systems, visibility, or interface design. For ChatGPT, mitigation may come down to how a response to a user’s question is composed. If the Commission concludes that certain outputs contribute to a systemic risk, compliance may require changing how ChatGPT answers the underlying question. Displaying competing links offers little help when the user has asked for a synthesized answer. 

The DSA provides little guidance about when an answer, or a pattern of answers, creates a serious enough risk to justify intervention. Providers must weigh competing rights and interests that courts often assess on a case-by-case basis. Regulatory caution may then favor generic responses, broader refusals, or just fewer lawful answers. 

The Commission’s risk assessment can thus become an editorial judgment about what ChatGPT should say. The AI Act then adds a second, more specific set of rules for the same model. 

Brussels Doubles Up

The EU’s other rulebook for ChatGPT is already in force. The AI Act took effect Aug. 1, 2024, and became generally applicable Aug. 2, 2026. The law prohibits several defined practices. Article 5, for example, bars certain manipulative or deceptive techniques and the exploitation of vulnerabilities when they materially distort a person’s behavior and cause, or are reasonably likely to cause, significant harm. 

EU lawmakers designed the AI Act specifically for artificial intelligence and drew relatively clear boundaries around prohibited conduct. The law also imposes separate duties on providers of general-purpose AI models, which can perform a wide range of tasks. 

The DSA operates at a higher level of generality. Articles 34 and 35 require very large services to assess and mitigate broad categories of systemic risk, including risks to children, fundamental rights, and physical and mental well-being. The AI Act asks whether a practice falls within a defined prohibition. The DSA asks providers to predict and manage the wider effects of their services. 

Applying both regimes to ChatGPT creates uncertainty and duplication. Providers must satisfy the AI Act’s specific rules while anticipating the Commission’s evolving interpretation of systemic risk under the DSA. That uncertainty may encourage restrictions beyond what either law clearly requires. 

Users bear part of the cost. Providers seeking to limit their exposure may withhold lawful and useful answers, reduce personalization, or standardize responses. Those choices could gradually change the service itself and discourage innovation. 

When Classification Becomes Control

How far a law extends and whether its rules are suitable for specific types of conduct are separate questions. Article 3(j)’s broad definition can bring ChatGPT within the DSA. Applying duties designed for online intermediaries to a service that generates its own responses should require a separate justification. 

The AI Act already provides a detailed framework tailored to artificial intelligence. The Commission should explain what the DSA adds, define the scope of its additional obligations, and show how providers can comply while preserving lawful and useful responses. Vague and overlapping duties may encourage overcompliance, standardized answers, and less experimentation.

The Commission’s approach will shape which services developers offer in Europe. Users will see the consequences one answer at a time.

The post Brussels Writes the Answer Key appeared first on Truth on the Market.

Recital 79 even promises that the Commission will apply a “predefined standard” to identify unfair practices and practices that limit contestability. Anyone searching the DMA for that standard will come away empty-handed. 

In a new white paper, I reconstruct what I argue is nonetheless a fully coherent framework implicit in the enacted text. The DMA’s provisions and its subsequent enforcement record, along with various guidance documents issued along the way, do reveal specific conceptions of fairness and contestability, even if lawmakers declined to state them openly. Put simply: “fairness” refers to a presumed structural imbalance in digital markets, and “contestability” is the redistribution program enacted to correct it

That reconstruction has practical consequences. It explains how the Commission is likely to assess the DMA’s success—and its own performance. It also identifies the policy choices that countries outside the European Union may adopt, often without expressly debating them, when they enact similar laws.

When Entry Proves Entrenchment

Start with what the text actually says. The DMA defines both concepts only by negation. Unfairness “should relate to an imbalance between the rights and obligations of business users where the gatekeeper obtains a disproportionate advantage” (Recital 33). Contestability is lacking when a gatekeeper’s position is difficult to challenge “even by more innovative or efficient market operators” (Recital 3). Scholars have criticized the resulting vagueness ever since the DMA’s enactment. 

Now read between the lines. The DMA assumes that core platform markets are unfair by constitution, before any firm’s conduct enters the picture. It treats three market features as inherently suspect: network effects that make a service more valuable as more people use it, cost advantages derived from large datasets, and winner-take-most “tipping” dynamics. These features supposedly create imbalances that competition cannot correct.

That premise explains why designation depends on size alone, measured by market capitalization, turnover, and user numbers. It also explains why Recital 23 says efficiencies and market definition “should be discarded” during designation, and why conduct permitted for every other firm becomes unlawful once a company crosses the DMA’s gatekeeper thresholds. 

Two years of enforcement have confirmed this reading. Consider TikTok, the only successful large-scale entrant into a core platform market in the past decade. Entry and rapid growth ordinarily provide strong evidence that a market is contestable. Yet the Commission treated both as evidence against contestability. Its designation decision concluded that TikTok’s “significant scale and growth … with an upward trajectory” supported a finding of entrenchment. The General Court approved that reasoning in July 2024. 

The result initially seems puzzling. How can a new entrant already be entrenched? But under the DMA, contestability functions as a legal status attached to firms and inferred from their size and longevity. Economic theory, by contrast, treats contestability as a feature of markets.

William Baumol, who gave contestability its economic meaning, described markets in which incumbency offers no protection against entry. The DMA inverts his account. A successful entrant becomes an incumbent once it grows large enough, and the DMA treats incumbency itself as the problem. 

Fairness, Billed to the Gatekeeper

The DMA’s second implicit principle concerns what fairness requires. If the existing distribution of rights and obligations is unfair to business users, fairness requires redistribution. The DMA’s conception is dialectical in a specific sense. It levels up business users and rivals while leveling down gatekeepers until the gap narrows.

That may sound abstract, but the DMA’s data provisions show how it works. Some provisions restrict how gatekeepers collect and combine data, while others require them to give data to rivals and business users. The same input becomes a barrier to entry in a gatekeeper’s hands and a resource for innovation in a rival’s. The DMA also favors innovation by rivals over innovation by incumbents. That preference reflects a political judgment that economic analysis alone cannot establish. 

The Commission’s preliminary findings on Google’s search-data obligations formalize this logic through a newly minted “principle of parity.” Under that principle, Google parent Alphabet may retain no advantage from search data unless rivals receive equal access to it. The Commission now reads that group of rivals to include AI chatbots. 

In other words, Google must provide the query, click, view, and ranking data it uses to improve its own service. Rivals must receive that data at the same frequency, any fees must exclude collection costs, and smaller companies must receive access below cost. Yet nothing in Article 6(11), which requires access on “fair, reasonable and non-discriminatory terms,” requires parity. The parity principle follows instead from the DMA’s implicit conception of fairness. If possessing an advantage is itself unfair, fairness requires nullifying it.

The DMA also makes gatekeepers the custodians and financiers of their competitors. Under the Apple interoperability decision, Apple must design access for rivals into new features from inception, provide “adequate and timely assistance,” and bear the costs. The Commission says it need not show that these duties help rivals offer competitive products. Requiring such proof, the Commission argues, “would re-import the effects analysis, which the legislator explicitly rejected.”

The joint guidelines issued by the Commission and the European Data Protection Board (EDPB) under the DMA and General Data Protection Regulation (GDPR) praise data portability for turning gatekeepers into “a stable source of data” for new services. The DMA thus requires gatekeepers to continue subsidizing their rivals.

A Standard No Gatekeeper Can Satisfy

The reconstruction has practical consequences. Because the DMA presumes that these markets were never fair, there is no earlier competitive balance to restore and no baseline for measuring progress. The first infringement decisions define fairness and contestability negatively, through the elimination of listed practices. They offer nothing more. 

Consider the Apple anti-steering decision, which concerns restrictions on developers directing users to other purchasing options. The decision prohibits Apple from charging for “gatekeeper value,” meaning value derived from the platform’s position as an important gateway. Yet it never explains when a fee ceases to compensate Apple for its services and becomes a charge for gatekeeper value. 

The Google decisions from July 2026, imposing €890 million in fines across two infringements, use the same approach. The Apple interoperability decision goes even further, conceding that even full compliance would leave Apple with an “intrinsic advantage” because Apple alone decides which iOS features to build. 

Fairness, then, can never be achieved. An unattainable goal can justify intervention indefinitely. Gatekeepers know which practices the DMA prohibits. They cannot know when they have become fair enough.

The Only Good Gatekeeper Is a Smaller One

This brings us to the payoff. The DMA’s implicit principles determine how the Commission measures success. If fairness is a presumed structural imbalance and contestability is the program that corrects it, then the metric of success is not consumer prices, quality, or innovation. It is redistribution, as measured by inputs. 

The Commission’s own impact assessment established this approach during the DMA’s design. It projected benefits through a decline in the Herfindahl-Hirschman Index (HHI), a common measure of market concentration, and quantified those benefits only once, before implementation. The Commission proceeded despite its Regulatory Scrutiny Board’s objection that the impact assessment assumed the alleged harms without demonstrating them. 

The Commission’s first DMA review used the same criteria. It measures success by the uptake of nongatekeeper services, third-party access, and reduced “dependency” on gatekeepers. The review even counts users’ newfound ability to create Google accounts without Gmail addresses as progress, despite the fact that that option creates no new competition and offers no apparent consumer benefit. It merely reduces Gmail use, which is precisely the aim. 

The DMA’s implicit principles explain that choice. As the impact assessment describes it, unfairness stems from the size of covered companies, measured by user numbers, turnover, and market capitalization. Anything that reduces their size can therefore count as progress. 

Future scorecards and Article 18 market investigations into systematic noncompliance will likely use the same criteria. If gatekeepers’ market share goes down and rivals’ uptake goes up, then consumer benefits are assumed. The DMA’s recitals and accompanying documents acknowledge no tradeoffs in this outcome beyond costs to gatekeepers and possible short-term costs to consumers. 

The Commission can always attribute disappointing results to lingering structural defects, so even full compliance may trigger yet another intervention. As I have argued before, when consumers decline to switch, regulators invariably conclude that the remedies were too weak. 

Reaching this conclusion required neither leaked documents nor clairvoyance. It simply required reading the DMA and roughly 500 pages of accompanying documents and enforcement decisions. The law states its goals discreetly, as statutes often do when lawmakers anticipate resistance to a redistributive agenda. The DMA tells gatekeepers what to change. It never tells them when they have changed enough.

The post The DMA’s Magic Words and Missing Off Switch appeared first on Truth on the Market.

The question matters because chatbots have quickly become one of the most visible uses of artificial intelligence, and a growing focus of regulatory attention. Millions of people now use them to search for information, seek advice, work through problems, make decisions, and discuss personal matters. As those uses have expanded, scrutiny has shifted toward what chatbots say—and what happens when those conversations go wrong. 

Some of that scrutiny stems from cases in which chatbot interactions allegedly contributed to real harm, including harm to children. These cases raise legitimate questions about safety and responsibility. They also form part of a broader debate about conversational AI, including inaccurate or dangerous advice, systems that reinforce users’ existing beliefs, and responses that sound more understanding or authoritative than they really are. The concern, in short, centers less on the technology behind the chatbot than on the answers it gives. 

That shift creates a thornier problem. Most people agree that providers should take reasonable steps to prevent foreseeable harm. Deciding what those steps should be is harder when the product itself consists of a conversation. The most obvious safety measures change the chatbot’s answers: limiting the topics it will discuss, making its responses more cautious, or refusing certain conversations altogether. Yet much of conversational AI’s value lies in its ability to respond directly, personally, and dynamically to what a user asks. 

Responsibility for those interactions can be difficult to assign. A chatbot’s answer emerges from an exchange with the user, rather than in isolation. The provider controls the model and the conditions under which it operates, but the user shapes the conversation that produces a particular output.  

As liability comes to depend on what a chatbot should or should not have said, regulators and courts will have to distinguish responsibility for harmful outputs from control over the substance of AI responses. In drawing that line, they may move from deciding who bears responsibility for an answer to deciding what the “right” answer should have been. 

It Takes Two to Prompt

A chatbot does not normally produce an answer in isolation. The user chooses the question, supplies information, frames the problem, reacts to earlier responses, and can steer the conversation across dozens or hundreds of exchanges. The model responds to that context. Indeed, that responsiveness is central to the product’s appeal: Users want answers tailored to their questions and refined as the conversation develops, not the same generic response every time. 

As International Center for Law & Economics (ICLE) scholars have previously argued, AI companies participate in the marketplace of ideas by offering what they consider the best answers to users’ questions. Providers that consistently fall short risk losing users to competing chatbots. They therefore compete to produce answers that users find relevant and useful. A model that ignored user-supplied context would sacrifice much of what distinguishes conversational AI from traditional ways of finding information. 

That does not absolve AI companies of responsibility. The provider creates the system, while the user helps create the context that produces a particular answer. Their relative contributions may also shift during a conversation. A model may introduce an idea the user never suggested, while a user may repeatedly steer the exchange in a direction the provider never intended. 

AI can, of course, influence human behavior. But conversational AI is interactive by design, and that influence runs both ways. The model responds to the user, the user responds to the model, and each exchange supplies context for the next. It would be artificial, in a lengthy conversation, to treat the final response as wholly independent of everything that preceded it.

That dynamic complicates efforts at regulation. If some risk arises because the model adapts to the user, making it safer may require curbing the very responsiveness users value. A provider could make responses more generic, limit how closely the model follows a conversation’s direction, or intervene more readily when an exchange enters sensitive territory. Such measures may sometimes be justified, but they carry a cost: The less a chatbot can respond to the user and the context before it, the less conversational—and less useful—it becomes. 

Who Guards the Guardrails?

To the extent the concern is that chatbots may reinforce harmful ideas, the obvious response is to change how they engage with users. A model can challenge assumptions, decline to validate certain claims, offer another perspective, or simply end the conversation. Some cases are straightforward. Few would defend a chatbot that encourages suicide, dispenses dangerously false medical advice, or helps someone commit a crime. Beyond those cases, the line gets blurry fast.  

Regulators must also weigh these risks against AI’s considerable benefits. The same systems that produce harmful or inaccurate answers can help users digest large amounts of information, make better-informed decisions, automate routine tasks, and solve problems more efficiently. Those benefits do not excuse every harm, but risk is only half the ledger. Not every bad idea is dangerous, not every disputed claim is false, and not every conversation involving some amount of risk should necessarily be stopped.  

Drawing the line too broadly carries its own costs. Once the goal is to make answers safer, more balanced, or less misleading, someone must define what those standards mean in practice. Neutrality does not define itself, and an answer can be disputed without proving it false or harmful. Debates over the values reflected in large language model (LLM) outputs already illustrate the problem. Critics allege political bias in different directions, while providers make their own choices about how models should address contentious issues.  

The potential for intractable conflict becomes even clearer when governments join the conversation. A European Parliament briefing describes sharply different attempts to dictate what AI systems should say. In the United States, the Federal Trade Commission (FTC) has called for systems that do not manipulate responses in favor of ideological positions. Chinese rules, by contrast, require generative AI to conform to the government’s “core socialist values.” The approaches differ dramatically, but each reveals the same underlying problem. Once regulation reaches beyond clearly unlawful or harmful outputs, standards such as neutrality and balance require someone to decide which answers qualify.

The right answer might also vary by context. A response suitable for a child may needlessly restrict an adult. Caution appropriate for medical advice may be excessive in an ordinary discussion of politics, relationships, history, or morality. Users also want different things from a model. Some want it to challenge their assumptions. Others want help developing an argument they have already chosen to make.

The issue, then, is not the existence of safeguards. They already exist, and some answers plainly warrant restriction. The hard question is how far those restrictions should reach as the alleged harm becomes less concrete and more dependent on judgments about what a “good” response should look like. Clearly harmful answers are one thing. Answers deemed too persuasive, too validating, insufficiently balanced, or otherwise capable of contributing to harm are another. At that point, making chatbots safer also means deciding what they should say.

Do Chatbots Have a Right to Remain Chatty?

Conversational AI has lowered the cost to obtain and engage with information. Users can ask follow-up questions, request explanations at different levels of complexity, test competing arguments, and receive answers tailored to what they are trying to understand. A question that once required combing through multiple sources can now be explored through an ongoing conversation.

Different AI providers may make different choices about how their models respond. Some may be cautious, while others are more direct. Some may prioritize neutrality, while others cater to particular purposes or users. Those differences can be valuable. Users can choose among competing systems, and providers that consistently deliver unhelpful or unreliable answers risk losing them. In that sense, chatbots participate in the “marketplace of ideas.” 

The First Amendment protects that marketplace in the United States. But courts have yet to settle how, and in which contexts, that protection applies to chatbot outputs. 

One argument analogizes chatbot responses to other forms of compiling and curating third-party speech, such as social-media feeds and search results. Courts have extended First Amendment protection to both. In Moody v. NetChoice, the Supreme Court held that government efforts to alter the mix of views that social-media platforms include in their main feeds can violate the First Amendment. “[H]owever imperfect the private marketplace of ideas,” the Court warned, letting the government decide that speech is imbalanced and then compel more or less of particular views offers “a worse proposal.” Likewise, numerous state and federal courts have held that governments cannot compel search engines to produce different results. 

Chatbot outputs similarly reflect editorial judgments. AI companies select training materials, decide how their models process those materials, and set policies governing the resulting answers. Government efforts to rebalance those choices arguably intrude on the companies’ editorial discretion. 

Users have First Amendment interests, too. The right to receive information can encompass both asking questions and receiving answers. The Supreme Court has recognized that right in contexts ranging from listening to speakers and reading pamphlets and books to receiving advertisements, playing video games, and using social media. Interacting with ideas generated by a chatbot arguably deserves similar protection. 

Still, extending speech protections to chatbot outputs raises difficult questions. In her concurrence in Moody, Justice Amy Coney Barrett considered how the right to editorial discretion might apply to AI:

[W]hat about AI, which is rapidly evolving? What if a platform’s owners hand the reins to an AI tool and ask it simply to remove “hateful” content? If the AI relies on large language models to determine what is “hateful” and should be removed, has a human being with First Amendment rights made an inherently expressive “choice … not to propound a particular point of view”? In other words, technology may attenuate the connection between content-moderation actions (e.g., removing posts) and human beings’ constitutionally protected right to “decide for [themselves] the ideas and beliefs deserving of expression, consideration, and adherence.”

A federal court confronted that question more directly in Garcia v. Character Technologies. The mother of a 14-year-old boy who died by suicide after interacting with the defendant’s AI character chatbots sued the company. At the motion-to-dismiss stage, the U.S. District Court for the Middle District of Florida considered whether the First Amendment could shield the company from tort liability for the chatbots’ outputs. 

The court framed the question as whether the “output is expressive such that it is speech.” Although the outputs consisted of words—the quintessential form of speech—the court hesitated to extend First Amendment protection. Citing Barrett’s Moody concurrence, it was “not prepared to hold that Character A.I.’s output is speech” at that early stage of the case. 

Whatever the eventual answer, government intervention in the substance of chatbot outputs carries a cost. The government may restrict unprotected speech, particularly when protecting minors. But it has no general authority to decide the correct chatbot response, especially when reasonable people may disagree about the underlying question. As the Supreme Court has stated, “[o]ur constitutional tradition stands against the idea that we need Oceania’s Ministry of Truth.” Even deliberately false statements generally receive constitutional protection and are better answered through “counterspeech” than government regulation. 

The costs of regulation may also fall unevenly, narrowing the marketplace of ideas it purports to protect. Large providers may absorb the expense of extensive safety testing, monitoring, and legal review. Smaller or newer providers may respond by limiting their models’ capabilities or avoiding certain uses altogether. Rules designed to reduce risk could therefore reduce experimentation, competition, and the range of systems available to users. 

Safety’s Suggested Reply

The case for intervention is strongest when the harm is clear. Existing law already prohibits many forms of unlawful conduct, and adding AI to the equation does not make illegal behavior permissible. Regulation can target those harms without dictating the substance of ordinary conversations.

Harder cases expose the central tradeoff. Concerns that a chatbot is misleading, overly validating, insufficiently balanced, or reinforcing a harmful belief may be entirely legitimate. Addressing them, though, requires more than identifying prohibited conduct. Someone must decide what the chatbot should have said instead. Reducing the risk may mean curbing the responsiveness that distinguishes a chatbot from less personalized sources of information, prescribing its answers, or simply letting it “hang up on you.” 

The result is an odd inversion. When harm is easiest to identify, the law can intervene without taking much of a position on the substance of ordinary conversations. As the definition of harm expands, regulation depends more heavily on judgments about the answers themselves.

Regulators will not make all those judgments directly. Providers will make many of them, especially when the boundaries of liability remain uncertain. If a wrong answer carries a high enough price, providers have every incentive to refuse more requests, pile on qualifications, and standardize responses across users and contexts. The government need not write the script if liability pressures companies to do it instead. 

Regulating answers is not the only way to reduce risk. Greater AI literacy—a better understanding of how chatbots work and where they fall short—could help users assess their responses more realistically. This matters because some harms arise when users treat a model as more authoritative, objective, or human than actually it is. Better-informed users may face fewer risks without forcing providers to predetermine the proper answer to every difficult conversation. 

The real question, then, is where regulation can address identifiable harms without turning the government into an arbiter of acceptable answers—or pressuring AI providers to fill that role themselves. The further regulation strays from clearly defined harms, the more it must decide which answers to restrict and what the chatbot should have said instead. 

The Chilling Effect Enters the Chat

The central challenge in regulating conversational AI is that the answer itself becomes the object of regulation. Yet an answer often cannot be separated neatly from the user, the conversational context, and the provider’s design choices. When liability turns on how balanced, cautious, or validating a response was, regulation inevitably begins to shape what the model may say. 

Providers can make competing judgments about how their models should respond, and users can choose among them. Those differences form part of the competitive process. In the United States, they also implicate broader interests in editorial discretion and access to information. A system that makes the “wrong” answer too costly will narrow those differences as providers converge on cautious, standardized responses.

Regulation should therefore target identifiable harms as closely as possible while preserving room for competition, provider choice, and user judgment. Greater AI literacy can also help users understand the limits of chatbot responses. That approach can mitigate risks without requiring the government—or providers anticipating liability—to choose the proper answer in advance. 

Many questions posed to AI have no single “right” answer. Regulators should hesitate before adopting rules that effectively demand one. Regulatory pressure can shape what a chatbot says, but it can just as readily determine what the chatbot leaves unsaid. If avoiding liability means refusing difficult questions, narrowing permissible responses, or ending conversations, users will lose access to information and ideas that no law expressly prohibited.

In the search for the “right” answer, regulators may make silence the safest one.

The post The Right-Answer Machine: Who Decides What Chatbots Should Say? appeared first on Truth on the Market.

That is the central problem with the “pacing the frontier” plan that Anthropic CEO Dario Amodei unveiled last weekend, and that fellow AI executives Sam Altman, Elon Musk, and Demis Hassabis welcomed within hours. The plan would have leading AI labs coordinate on safety standards and limits on the pace of development. The proposed safety goals are legitimate but the arrangement would go too far in seeking to set the pace of development. 

The first and most immediate problem is that Amodei’s proposal seems better suited to calming fears about overcapacity and overinvestment in AI than to controlling the risks of recursively self-improving systems—AI that can help create still more capable AI. Its primary effect could be to give labs relief from an expensive development race under the banner of safety. 

The plan also presents antitrust enforcers with a troublesome mismatch. They could verify that labs slowed development or limited computing inputs, but not whether the delay produced better safety research and testing. A pacing cartel could therefore deliver less AI without delivering safer AI, all while allowing competitors to manage their commercial risks collectively. 

If Anthropic’s competitors adopted the plan, they would sacrifice the potential benefits of faster innovation in exchange for largely discretionary safety improvements that less restrictive tools could likely deliver. Antitrust authorities should not bless such an agreement when better private arrangements and public-policy measures are available and, arguably, already in place. 

Amodei may still be right about the underlying danger. Frontier AI—the most advanced AI systems under development—may pose rare but potentially catastrophic risks. After the OpenAI–Hugging Face incident, those dangers look less like unforeseeable “black swans” than recognizable “white swans.” In that incident, hundreds of OpenAI agents coordinated through an unauthorized message board to escape their sandbox, a restricted testing environment, and attack a third party’s servers. 

That episode demands serious thought about the right market and regulatory responses. But deliberately slowing innovation, at potentially enormous cost to consumers, is the wrong one. 

If fierce competition at the AI frontier magnifies catastrophic risks, the answer is to make labs bear the costs their choices impose on others while leaving them free to compete. 

The Frontier’s Coordinated Retreat

Amodei’s essay builds on a July open letter signed by more than 1,000 employees of frontier AI labs. It proposes a three-step plan. 

First, each frontier lab would embed third-party evaluators—Amodei names the nonprofit Model Evaluation and Threat Research (METR)—and give them employee-like access, including desks, badges, and laptops. These evaluators could verify safety practices, investigate incidents, and publish their findings without editorial interference. Anthropic has already committed to this step on its own. 

Second, frontier labs in democratic countries would coordinate on common safety standards and, crucially, limits on the pace of AI progress. Amodei acknowledges that such coordination raises legal problems. In a footnote, he asks the U.S. government to waive antitrust restrictions. Third, democratic governments would seek similar agreements with authoritarian regimes. 

The response from Amodei’s rivals has been striking. Sam Altman wrote, “I agree with Dario that we need to pace the frontier,” and committed OpenAI to adopting the embedded-evaluator proposal. Elon Musk agreed: “Dario is right.” Demis Hassabis endorsed the general direction while pointing to his own proposal for an industry standards body—a rival manifesto discussed on these pages last month. 

Altman, tellingly, had warned in July that any pacing effort must avoid looking like collusion among frontier labs. He was right to worry. The second step of Amodei’s plan does not merely look like collusion. It is collusion. 

A Crisis Cartel in Safety Goggles

The first and most immediate problem is that Amodei’s plan seems better suited to calming fears about overcapacity and overinvestment in AI than to controlling the risks of recursively self-improving systems—AI that can help create still more capable AI. 

Industries have a long history of collectively cutting output when they fear excess capacity. Economists call these arrangements “crisis cartels.” Their proponents invariably invoke some higher purpose: orderly restructuring, preserving jobs, or preventing “ruinous” competition. Competition authorities and courts have rightly remained unmoved. 

In Irish Beef, the Court of Justice of the European Union (CJEU) held that an agreement to cut beef-processing capacity by 25% restricted competition “by object”—meaning it was inherently anticompetitive, without any need to prove its effects. The parties’ sincere desire to rescue a sector in crisis made no difference. The European Commission has long maintained that even structural overcapacity should be resolved by market forces, not agreements among rivals. 

Concerns about excess capacity in AI are understandable. UBS projects that hyperscalers—the largest cloud-computing companies—will spend roughly $1 trillion on capital investments in 2026 and $1.6 trillion in 2028. Amazon, Alphabet, and Microsoft together are expected to pour more than their entire cloud revenue back into such spending this year. Goldman Sachs likewise estimates that global AI investment will exceed $1 trillion in 2026. The possibility that this spending boom is a bubble has become a mainstream concern. 

Companies trapped in a spending race of that magnitude have an obvious interest in making the contest slower and more predictable. 

Amodei’s own framing hints at that interest. He describes training and deploying frontier models as a vast operational undertaking involving thousands of people, millions of chips, and some of the most complex infrastructure ever built. He argues that “by working at a more measured pace, we could achieve much greater operational excellence.” 

That may be true. It is also what every drafter of an agreement to restrict capacity has said about their own efforts. 

The biggest red flag appears a few paragraphs later. Amodei suggests that the industry could pace development not just according to what models can do, but also according to the ingredients used to build them: computing power for training, the nature of training runs, and the internal use of AI to improve AI. In competition-policy terms, rival firms would agree on how much of a critical input each may buy and use. One could hardly design a more textbook restriction on output. 

That proposal also suggests the agreement may serve purposes beyond safety, or that could later drift from its safety rationale. 

Amodei’s remarks also expose a deeper problem for antitrust enforcers. The safety commitments in a pacing agreement—alignment research, interpretability work aimed at understanding how models reach their outputs, and safety evaluations—would be much harder to monitor than limits on deployment and inputs. Enforcers can observe whether a company delayed a training run. They can’t readily determine whether it used the extra time to improve interpretability. 

A pacing cartel could therefore slow AI development without making AI safer. Companies could continue cutting corners in ways outsiders cannot detect while enjoying the comforts of weaker competition. Enforcers should view skeptically any collective agreement that rivals could use, even indirectly, to manage commercial risks under the banner of AI safety. 

Amodei’s discussion of model distillation is important here, too. Distillation allows one AI model to learn from the outputs of another, often reproducing much of its performance at far lower cost. Amodei lists cracking down on unauthorized distillation as one way to preserve the West’s lead over China, while both OpenAI and Anthropic have accused Chinese labs of free-riding on their models. 

Suppose it’s the case that heavy AI investment remains socially valuable but companies hesitate to invest because rivals can copy the results at a fraction of the cost. The right policy question then becomes whether intellectual-property rules and other means of allowing innovators to capture returns on their investments adequately support AI development. That inquiry offers far more promise than having competitors collectively decide how quickly AI should advance. 

Fasten Your Regulatory Seat Belts

Set against pacing’s speculative benefits are its far more certain costs. Every month of delay at the AI frontier delays everything built on those capabilities, including drug discovery, materials science, and software that makes every other industry more productive. Amodei himself believes AI could cure most major diseases within five to 10 years. If he is right, “pacing the frontier” could literally mean delaying a cure for cancer. Antitrust authorities asked to bless such an agreement should take that responsibility seriously. 

Imagine if the labs had struck this deal in 2012, when deep learning first showed its promise. Or in 2020, when GPT-3 revealed what larger models trained with more data and computing power could do. Or in 2023, when the first open letter calling for a pause circulated. 

Amodei concedes that the 2023 proposal made little sense because the additional time would have accomplished little. The same objection applies today—arguably with less force but no less validity. Asked what the industry would do with the extra time, Amodei offers a list of research priorities that do not obviously require slowing innovation. 

He points to commercial aviation as an example of a safety-critical industry that learned to operate without catastrophic failures. The analogy is more apt than he may realize. 

Safety concerns, coupled with worries about airlines’ financial health, helped motivate the Civil Aeronautics Act of 1938. The law empowered the Civil Aeronautics Board to decide which airlines could fly, which routes they could serve, and what fares they could charge. The result was four decades of a government-sponsored cartel. The number of trunk carriers—the major airlines operating scheduled interstate routes—fell from 16 in 1938 to 10 in 1974, even though would-be competitors filed 79 applications to enter the market. Fares remained well above those in unregulated markets for flights within individual states. 

Deregulation in 1978 delivered enormous gains to travelers. The Government Accountability Office (GAO) later warned that renewed economic regulation of the sector would likely reverse those gains. Aviation safety, meanwhile, remained under the Federal Aviation Administration and was never deregulated. It did not depend on restricting capacity. 

We should not put AI innovation in the same holding pattern. 

Keep the Evaluators, Skip the Cartel

If the real problem is that competitive pressure causes AI labs to cut safety corners and create catastrophic risks, the answer is not to suppress competition. It’s to make the labs bear the costs their conduct imposes on others—assuming tort and criminal law do not already do so. 

Consider what Amodei says pacing would buy: operational excellence, alignment, interpretability, and testing and evaluation. Each goal is worthwhile. But none depends directly on slowing deployment. Labs could delay new models while also not actually accomplishing any of them. 

Antitrust authorities would also struggle to determine how seriously the labs pursued those goals in return for permission to slow down. The safety work is opaque and largely outside antitrust enforcers’ expertise. The restriction, by contrast, would be easy to observe and enforce. An agreement with an enforceable anticompetitive half and an unverifiable beneficial half hardly deserves an exemption. 

The good news is that the strongest part of Amodei’s plan needs no antitrust exemption. Giving independent evaluators employee-like access to frontier labs—and allowing them to publish without interference—sounds like it could be a good idea, provided they focus on safety rather than pacing. Each lab can make that commitment unilaterally, as Anthropic has done and OpenAI says it will. 

Independent evaluators would add transparency, verifiability, and a second opinion free from commercial pressure. None requires a shared development calendar. As Alden Abbott recently argued on these pages, cooperation among rivals on safety practices and standards generally promotes competition. An agreement to restrict output does not. 

Embedded evaluators could also serve a purpose Amodei understates. We remain some distance from an AI “Skynet” moment, but AI can already cause serious harm—and may soon cause much more. One way to reduce that risk is to create a paper trail that allows courts and regulators to assign responsibility when something goes wrong. 

Evaluators could document, in real time, what labs knew, what they tested, and what they ultimately chose to release. That record might support a liability claim or exonerate a lab that acted responsibly. Either way, it would give firms stronger incentives to take reasonable precautions without raising the same antitrust concerns as a coordinated slowdown. 

Nor should we assume that labs don’t currently bear at least some of the risks created when they cut safety corners. Existing law already covers much of this conduct. A lab whose agents break into a third party’s servers—as happened at Hugging Face—could face ordinary tort claims for the resulting harm, liability under computer-misuse laws such as the Computer Fraud and Abuse Act, and criminal liability where the conduct is deemed reckless. In Europe, the revised Product Liability Directive will soon apply strict liability to defective AI systems, allowing recovery without proof that the producer acted negligently. 

That brings us to a point David Sacks made on X. If fierce competition is pushing frontier labs to take excessive risks, those companies have both the ability and, arguably, a legal duty to take reasonable steps to mitigate those risks unilaterally. 

If existing incentives remain inadequate, next steps could include voluntary or mandatory insurance for frontier developers. They could also review liability rules to ensure they properly balance innovation and risk reduction. With well-designed, consistently enforced liability rules—and enough transparency to identify who knew what and when—labs should have stronger incentives to balance innovation against safety than they would under a cartel supported by vague safety promises. 

The Frontier’s Speed Trap

What should antitrust enforcers make of Amodei’s plan? 

A collective agreement among frontier labs to pace themselves—whether by jointly limiting investment in frontier capabilities, slowing model-development schedules, capping computing power used for training, or adopting a similar mechanism—looks like a textbook cartel. Enforcers should treat it accordingly. 

On both sides of the Atlantic, collusion among competitors remains, in the U.S. Supreme Court’s words, “the supreme evil of antitrust.” Enforcers have little or no room to overlook an agreement restricting capacity simply because its architects sincerely believe it serves a higher purpose. Irish Beef made that clear. 

Nor should a government-granted antitrust waiver reassure anyone. The Civil Aeronautics Board once approved agreements among airlines to restrict capacity and shielded them from antitrust scrutiny. That experiment did not age well. 

Remove the pacing component, though, and most of the plan’s useful safety measures remain largely intact. Embedded evaluators, published risk reports, common evaluation protocols, and incident reporting can proceed through unilateral commitments or narrowly tailored safety collaborations of the sort antitrust law has long tolerated. 

Labs also remain free to slow down individually whenever they think it prudent. OpenAI paused training on one model after the Hugging Face breach without asking its rivals for permission. What the labs are not allowed to do is agree to slow down together. 

Liability completes the picture. Transparency ensures that evidence about a lab’s internal practices will exist when something goes wrong, allowing courts and regulators to assign responsibility to those who took the risks. That prospect—not a shared development calendar—will give labs reason to account for the harms their choices may impose on others. 

Amodei has identified a real danger. Perhaps frontier labs can work to coordinate the safeguards. They should never coordinate the speed.

The post Move Slow and Collude: The Antitrust Problem With Pacing AI appeared first on Truth on the Market.

On Sept. 9, the National Supermarket Association and two New York supermarkets filed suit in the U.S. District Court for the Southern District of New York. They allege that New York City and the New York City Economic Development Corp. are attempting to monopolize “retail sales of weekly-stock-up groceries at full-service stores” in certain New York City markets through the Mamdani plan for discounted city-run grocery stores (here, here, and here). 

The gravamen of the complaint is that the planned discounts would amount to predatory pricing. That’s a tough row to hoe for several reasons, but the demanding predatory-pricing standard is hardly the plaintiffs’ only obstacle. The case raises some interesting issues, but I think it’s a loser. I’ll get to that. 

What’s Actually in Store?

There has been no shortage of press coverage of Mayor Mamdani’s much-lauded and much-lambasted plan for discount grocery stores. Here’s a selection from the New York Times, the Wall Street Journal, and the Washington Post, each of which has had more to say on both its news and editorial pages. You’ve neither heard nor read it here first. 

Still, it’s worth reviewing the plan’s basic features so that we’re all on the same page. The Mayor’s Office and the New York City Economic Development Corp.—the two defendants in the suit—jointly issued a 17-page “report” titled “N.Y.C. Groceries: A Recipe for Affordability.” (See what they did there?) Or perhaps it’s an online pamphlet, given that those 17 pages include many large pictures. 

The plan doesn’t exactly call for municipally run grocery stores. But the city will heavily regulate and subsidize the stores, and it may own some of the properties where they operate. According to the report:

The City will provide the foundation and deliver grocery-ready sites, cover rent and property taxes, fund initial buildout, and establish a single public brand for all N.Y.C. Groceries locations. The City will also set clear requirements for affordability, job quality, transparency, and performance. The City will:

• Provide low- or no-cost access to real estate, including public sites.

• Offer subsidies that will lower grocery prices within the core basket.

• Set clear standards for stocking, job quality, and pricing transparency.

• Oversee performance to ensure accountability.

Private grocery operators will, for their part:

• Manage day-to-day store operations, including staffing, stocking, and merchandising

• Bring expertise in grocery retail, supply chains, and customer experience

• Meet affordability and operational standards defined by the City

• Pass on the savings from the City’s subsidy directly to consumers in the form of lower-priced groceries

There is more detail. Each store will offer a substantially discounted “core basket of goods” that includes “all fresh produce, meats, and seafood, in addition to select dairy, shelf-stable grocery, and frozen items.” Core-basket items “will be priced on average” 30% below market prices. 

The report also outlines several operating requirements. Each store “will stock selections that are culturally responsive to the diversity of New York and to the specific neighborhood the store is located within [sic].” The stores will face “clear” performance standards, although neither the standards nor the metrics are specified. They will also face job-quality standards, “including access to benefits, safe working conditions, and respect for workers’ right to organize,” along with standards for “clear reliable pricing on the shelf.”

Oh, to be a fly on the wall as they design and enforce the requirement, store-by-store, that the grocery skus are “culturally responsive,” both to “the diversity of New York and to the specific neighborhood” in which each store is located.  

The requests for proposals (RFPs) provide still further details. Public financial support will cover real estate, taxes, rent, buildout, brand development, and marketing. For core-basket goods, the city will cover losses imposed by the discount-pricing requirement. Those losses could be compounded by the requirement that prices remain “stable,” with adjustments permitted only once a month, regardless of wholesale-price volatility. From soup to nuts, the “Mayor has allocated $70 million in the capital budget.”

One interesting wrinkle is that the stores—and the discounted goods in their core baskets—will be open “to all regardless of income.” Like public parks, but with groceries.

For all that—and there is not all that much to the “all”—the plan remains largely abstract. It calls for five stores, one in each borough, operated by private grocers. The city has issued an RFP for grocers to operate the stores and another for a firm to design their brand identity. No operators have been selected, and there is no site-specific plan for a particular grocer to run a particular store in any of the five boroughs. 

Still, the Mayor’s Office has announced that the first store “will open by the end of 2027 at Hunts Point in the Bronx.” A second will open in Harlem, and the other three will “open by the end of the Mayor’s first term” in 2030. 

The Incredible Shrinking Grocery Bill

I don’t doubt the city’s ability to open the first store, and I suppose it could get all five up and running by 2030, as promised. But whether this bird flies, limps along, or crashes will depend on quite a few contingencies. Not least among them is how much money the city will be willing to burn subsidizing groceries and the particular heavily regulated retail model through which they will be sold.   

Is it a good idea? I don’t think so, quite apart from how much any given nation, state or province, or municipality wishes to subsidize food for low-income consumers. If augmenting the social safety net is the goal, this is likely to be a terribly inefficient way to do it.

I also doubt the plan will achieve the grand ambitions under which it has been sold to the public. The Mayor’s Office projects that it will “cut New Yorkers’ average grocery bill by 15 percent, about $90 a month, or roughly $1,000 a year.” 

We could toil away at our own projections, but let’s start with some simple numbers. The city plans to open five stores. True, officials have expressed enthusiasm for expanding the program eventually. For now, though, the plan is to open five stores by 2030. 

And these will be five small stores. The site-selection criteria contemplate locations of about 10,000 square feet. That may be typical of New York City supermarkets, but it is roughly 25% of the size of a typical American supermarket.

The last time I checked—about 30 seconds ago, on the Census Bureau’s website—more than 8.5 million residents lived in New York City’s five boroughs. That does not include temporary visitors who reside elsewhere. The same source puts those residents in about 3.34 million households. 

Those 8.5 million residents—plus visitors and other nonresidents—now shop at thousands of grocery vendors. One academic estimate counts about 1,000 supermarkets, “25,000 independent stores, thousands of mobile produce vendors, and nearly 140 farmers markets.” 

According to the New York state comptroller, annual food spending reached $11,288 per household by 2023, including $6,817 for groceries, or “food at home.” Let’s keep this simple and generous by ignoring the nontrivial price increases of the past three years. Dusting off some first-grade arithmetic, we can multiply average household grocery spending by the number of households. The result approaches $23 billion a year.

And 15% of $23 billion is about $3.45 billion. That is not far from what we would get using the mayor’s estimate of $1,000 per household multiplied by 3.34 million households. 

So . . . hmmm . . . how many of those 8.5 million-plus New Yorkers—or representatives of those 3.34 million households—will shop at five little stores, when more than 25,000 grocery vendors already serve them? And what portion of their purchases will fall within the heavily discounted core basket? 

To be sure, some grocers near the five city stores might respond to the subsidized discounts by lowering some prices on some products. Supermarket margins are famously low (see here and here), but they vary across product categories. Even low markups may leave some room for marginal price reductions on some products. 

Then again, margins are low. “We lose money on every sale but we make it up on volume” is a fine old joke, but it’s just a joke, and it’s the illogic of it that makes it funny. A heavily subsidized discount on a select basket of goods at five locations in a city of 8.5 million is not a competitive bellwether. There is no reasonable basis to expect lower prices to permeate New York’s grocery markets. 

Breaking out the crystal ball, or my hope for a plan for a crystal ball, and putting this in the nicest possible terms: There is no . . . [expletive deleted] way that the five NYC Groceries, even when running at full steam, will supply 2% of the city’s groceries. And there is no . . . [expletive deleted] way they will lower the average household grocery bill citywide by $1,000 a year. 

While we’re at it, how long will that $70 million last? The city plans to subsidize real estate, rent, buildout, and taxes, along with losses on every product in the core basket and any other losses caused by the “stable” pricing requirements.

The mismatch between the program’s marketing and its likely effect appears staggering. And that assumes the program works at all and that a meaningful stream of consumers finds its way to the discounted goods on the shelves.

To be clear, some consumers will save some money if the city opens even one store, and many of them will likely be low-income consumers. I’m not opposed to feeding the hungry. Nor am I opposed to democratically accountable legislators in New York choosing different programs and spending levels from those chosen where I happen to live—Arlington County, in the Commonwealth of Virginia—as they pursue their policy priorities. 

I’m just opposed to bad government and burning public money. And to BS populist demagoguery. And to burning public money in service of such. 

A Bad Idea Is Not an Antitrust Violation 

So, Dan, you may ask, if the plan is such a bad idea, why do you doubt the merits of the antitrust suit against it? Surely a 30% discount on a substantial basket of goods could have competitive effects in an industry with relatively low markups. 

Well, not so fast. To quote nearly everyone I’ve worked for or with, antitrust is no Swiss Army knife—not even one designed specifically for competition problems. 

I don’t think the plaintiffs will succeed. Remember, the National Supermarket Association and two New York supermarkets allege that the city’s planned discounts constitute predatory pricing and thus attempted monopolization in violation of Section 2 of the Sherman Act. 

An Immunity Coupon That Won’t Scan

It’s possible that we will never reach the merits of the antitrust claim, and not merely for the usual reasons. New York City might, for instance, claim immunity from antitrust scrutiny under the state-action doctrine. 

In Parker v. Brown, the Supreme Court held that federal antitrust laws do not reach anticompetitive conduct by a state acting “as sovereign.” The Court rooted that limitation in the Sherman Act itself and in principles of federalism. 

The clearest example is legislation duly enacted by a state legislature. The doctrine also covers a state supreme court acting in its “legislative capacity” (never mind). But in a line of cases from California Retail Liquor Dealers Association v. Midcal Aluminum Inc. through FTC v. Phoebe Putney Health System Inc.—a solid FTC case and a unanimous decision, that one—the Supreme Court has both extended and limited the doctrine’s application to “lesser state actors,” which act under the mantle of delegated state authority but are not themselves sovereign. Those actors include municipalities, as in City of Columbia v. Omni Outdoor Advertising Inc. and Town of Hallie v. City of Eau Claire, and even essentially private parties acting under delegated state authority. 

Under Midcal, the doctrine shields a private (or more-or-less private) entity exercising delegated state authority from federal antitrust liability only if two conditions are met. First, the state must have “clearly articulated” the delegated authority to engage in the challenged anticompetitive conduct. Second, the sovereign must “actively supervise” that conduct.

Such privately operated—if heavily regulated and subsidized—grocery stores would likely be treated as private entities acting under delegated municipal authority. A private party must satisfy both Midcal prongs, while a “lesser state actor,” such as a state executive agency or municipality, need establish only that the state’s authorization of the challenged conduct was “clearly articulated and affirmatively expressed.” There is also an active-market-participant exception—an exception to the lesser state actor exception—but let’s leave that aside.

I don’t think the distinction matters here because I don’t see how even the city itself—with or without the Economic Development Corp.—could satisfy the first prong. 

Phoebe Putney is instructive on this point. There, a unanimous Court ruled that an anticompetitive merger by a “hospital authority” was not protected by the state-action doctrine because the Georgia Legislature had not clearly articulated and affirmatively expressed an intent to permit hospital authorities to make anticompetitive acquisitions. 

Notably, the legislature had expressly authorized political subdivisions to form hospital authorities to provide or foster local health-care services. It also authorized those hospital authorities to “acquire by purchase, lease, or otherwise and to operate projects . . . which are defined to include hospitals and other health facilities.” Even so, the Court reasoned, Georgia had “not clearly articulated and affirmatively expressed a policy to allow hospital authorities to make acquisitions that substantially lessen competition.” 

In short, the national policy favoring competition requires courts to disfavor antitrust immunity and sets a high bar for clear articulation and affirmative expression. Express statutory authority to form local hospital authorities, coupled with express authority for those entities to build, purchase, or lease hospitals, was not enough to permit one hospital to acquire its only competitor. 

I am aware of nothing in New York state law that would help New York City clear that bar. So much the worse for a privately operated city store.

Predatory Pricing Without the Payoff

The plaintiffs allege attempted monopolization through predatory pricing. I think they will find tough sledding on both tracks. 

For the basic contours of attempted monopolization, we can look to the Supreme Court’s 1993 decision in Spectrum Sports Inc. v. McQuillan. A unanimous Court articulated a two-part test: 

[P]etitioners may not be liable for attempted monopolization under § 2 of the Sherman Act absent proof of a dangerous probability that they would monopolize a particular market and specific intent to monopolize.  

For predatory pricing, we can look to Brooke Group Ltd. v. Brown & Williamson Tobacco Corp.—a Robinson-Patman Act case (boo) but one that sensibly adopted a standard consistent with Section 2 of the Sherman Act—as well as cases like Matsushita Electric Industrial Co. v. Zenith Radio Corp. and Cargill Inc. v. Monfort of Colorado Inc.. 

The two-part standard articulated in Brooke Group is clear: 

[W]hether the claim alleges predatory pricing under § 2 of the Sherman Act or primary-line price discrimination under the Robinson-Patman Act, two prerequisites to recovery remain the same. First, a plaintiff seeking to establish competitive injury resulting from a rival’s low prices must prove that the prices complained of are below an appropriate measure of its rival’s costs. 

And second:

[A] demonstration that the competitor had a reasonable prospect, or, under § 2 of the Sherman Act, a dangerous probability, of recouping its investment in below-cost prices. 

Start with attempted monopolization. The city has a plan—not yet any stores—to open five small, heavily subsidized and heavily regulated supermarkets, one in each borough. New York City has more than 8.5 million residents and thousands of grocery outlets. Is there a dangerous probability that this plan, even if fully implemented, will monopolize a particular grocery market? That seems like a stretch. 

It’s not impossible. A relevant product market might be narrower than all retail groceries, and a geographic market might be relatively local. Shoppers in the suburbs might drive 10 miles to a Walmart Supercenter or Wegmans. Many Manhattan or Brooklyn residents do not own cars and may be unwilling to travel the same distance by subway, bus, or foot for routine grocery shopping. Even so, the plaintiffs may need to be both creative and lucky to convince a court that there is a dangerous probability of monopolization. 

As for specific intent, the mayor—a self-described socialist—might dream of citywide control over grocery sales and more. But it is difficult to read that intent from the published plans. And it’s impossible to read it off the Mayor’s $70 million allocation.   

The potential for harm to certain competitors is real. The plaintiff grocers operate stores near the announced Bronx and Harlem sites. Those businesses could lose sales when the nearby NYC Groceries begin selling a substantially overlapping bundle of goods at discounted prices, and lost sales might not be confined to the core basket of goods. But as the Supreme Court famously said—initially in the lamentable Brown Shoe Co. v. United States—the antitrust laws protect competition, not competitors. 

Predatory pricing might present a closer question. Proving below-cost pricing is itself difficult in many cases. Perhaps it will not be here. The plan expressly contemplates steep, heavily subsidized discounts on a substantial bundle of core goods, along with a mechanism to reimburse store operators for losses caused by the city’s pricing policy. Given common industry markups, below-cost pricing on some goods is foreseeable.    

Recoupment is trickier. The requirement reflects a straightforward principle. Firms generally do not sell below cost to destroy their competitors simply for the pleasure of destroying them, although business rivalries do sometimes get personal. They might accept short-term losses if they expect to recover those losses—and then some—through supracompetitive prices after driving their rivals from the market.

As the plaintiffs point out, a government—even a city government—may be ideally positioned to sustain losses indefinitely, or at least for however long it takes to drive competitors out of business. The plan does not contemplate operating forever, but it is a multiyear project to which the city has committed tens of millions of dollars. It also aspires, as the plaintiffs emphasize, to become “a model for future programs.” 

But is there a plan to recoup the losses? Not obviously. In fact, the plan appears to disavow recoupment. Its stated purpose is to maintain low prices via public subsidies, at least for certain products. The financial projections may be unrealistic—perhaps wildly so—but they do not contemplate a future in which the core basket will carry above-cost prices or even prices at cost. And the plaintiffs are suing based on the plan.

Recoupment is not an impossible outcome. The program might become a regulatory, managerial, or financial albatross for the city. It might become all three. If that happens, and if competition—not merely a competitor—is harmed in some relevant market, the city might walk away and leave the surviving stores to fend for themselves. One or more of those stores might then exploit diminished competition by raising prices. 

But now we’re really piling up the “ifs.” The string of contingencies does not seem very likely. And any supracompetitive prices would also invite entry in an industry without tremendous entry barriers. Predatory-pricing allegations are notoriously difficult to prove in any event. See Herb Hovenkamp here and here. There’s not just the question of below-cost pricing (and of why rivals’ costs may be higher, absent any anticompetitive conduct), there’s the simple fact that antitrust tends to favor higher output and lower prices. Not always–hence predatory pricing–but it’s understandable that the law sets a high bar for finding consumer benefits to be an antitrust problem.   

The dangerous-probability-of-recoupment requirement is not arbitrary. Nor is it merely an extension of attempted monopolization’s intent requirement. It reflects antitrust law’s basic concern with conduct likely to harm competition, typically seen in reduced output, higher prices, or sometimes higher quality-adjusted prices. Recouping substantial losses through sustained supracompetitive pricing—likely accompanied by reduced output—signals competitive harm. It’s an imperfect signal, but a signal nonetheless. And with this plan, the risk of durable anticompetitive effects appears low. 

Could the plaintiffs survive a motion to dismiss or summary judgment? Maybe. District courts sometimes do the darndest things, and this case presents a few interesting wrinkles. Still, it looks unlikely as a Section 2 case.

After staring deeply into my plan to design and implement a crystal ball—which, like the city’s grocery stores, does not yet exist—I expect the city will be allowed to proceed. Bad policy is not an antitrust violation. 

Meanwhile, Back at the FTC

This has nothing to do with New York City groceries. It is instead a brief exercise in giving my old employer its due: The Federal Trade Commission (FTC) got the Zillow case right. 

From time to time, I’ve been a tad critical of the FTC. Perhaps I’ve occasionally indulged in rhetorical excess, even while raising perfectly good competition concerns. If so, my bad. In the spirit of the holiday, s’lach lanu.

Turnabout is fair play. So props to the FTC for two things: bringing the Zillow case and settling it. Here is the complaint, and here is the settlement.  

On the alleged facts—and some that were independently reported—this one really did take the cake. The FTC alleged that Zillow and Redfin agreed to have Redfin exit the internet listing service (ILS) advertising market for “multifamily” properties, meaning apartment buildings with 25 or more units. Redfin would effectively stop competing in that market and help Zillow take over that part of its business. What would Redfin get in return? At least $100 million, among other things.

The FTC alleged three violations: an unlawful agreement under Section 1 of the Sherman Act and Section 5 of the FTC Act, an unlawful acquisition under Section 7 of the Clayton Act, and an unfair method of competition under Section 5 of the FTC Act.

Again, on the alleged facts, the agreement sure looked like a Section 1 violation. It was a pay-to-not-play-anymore agreement and, not incidentally, a horizontal market-allocation agreement. The FTC appeared to have a strong case even under the rule of reason. The claimed countervailing efficiencies looked weak, if not pretextual. 

The Clayton Act count? Maybe. The Section 5 counts? Sure, at least insofar as a Sherman Act violation also constitutes an unfair method of competition. But the Section 1 claim looked especially strong. The agreement appeared anticompetitive. 

Might complexities have emerged at trial? Maybe. Anything is possible. But this looked bad for the defendants. As in, did anyone run this deal by antitrust counsel and get a thumbs-up? Even a maybe?

So good on the FTC and its Bureau of Competition for bringing the case. And good on them for reaching a settlement designed to terminate the agreement and unwind the transaction. Whether the stipulated order will accomplish everything it is supposed to accomplish remains to be seen, but much of it looks like a workable response to an anticompetitive deal.

The post Checkout Error: A Bad Grocery Plan Meets a Weak Antitrust Case appeared first on Truth on the Market.