Sui Foundation announced on October 12, 2026 that it is investing $30 million in DATA Network, the AI training data protocol formerly Story Protocol, which will retire its Layer 1 and run natively on Sui.

A single-use chain that no longer paid its way

The project launched in February 2025 as Story Protocol, an EVM-compatible Layer 1 built for programmable intellectual property and automated licensing. In June 2026 it rebranded as the DATA Foundation and pivoted to what it called a full stack for AI training data; holders of the original IP token were swapped 1:1 into the new DATA token. That month it also launched Trace, a registry that logs the origin, consent and licensing rights of training data.

The pivot is what made the dedicated chain a liability. A DATA Foundation spokesperson said running a blockchain for a single use case meant carrying the costs of validators, network security and infrastructure. Trace has grown quickly since June: a DATA Foundation representative put its registrations at more than 270 million in about four months, while a count published in October put the figure at over 250 million.

What Sui gets, and what DATA gets

Under the deal, Walrus, the decentralized storage platform built alongside Sui, becomes the default storage layer for applications operated by the DATA Foundation, and Trace joins Sui. “Building storage and access control at that scale ourselves would have meant rebuilding what Sui already has in Walrus and Seal,” a DATA Foundation spokesperson said, referring to Sui’s access-control technology.

The deal reads as a bet on where AI data infrastructure is heading. Trace answers the question AI developers face as demand for verifiable, consent-gathered datasets rises: who owns a given training dataset and who may license it. By hosting the registry and storing the datasets on its own stack, Sui positions itself to capture a workload that could just as easily have landed on any chain. Training data is becoming a chain-agnostic commodity, and the chain that hosts it wins the surrounding activity. Kled, the protocol’s flagship integration, has already recorded more than 1.5 billion user-contributed data records, and other apps including Numo sit on the same stack.

Whether the money shaped the decision is a fair question. Asked how much the investment influenced the move, a DATA Foundation representative said the $30 million was a separate part of the partnership. No structure, token details or vesting terms for the investment have been disclosed, and the claim that the payment did not drive the migration is the representative’s own.

What happens to the old chain

DATA Network will keep operating until the upgrade completes, and the partners have said a detailed timeline and a holder transition plan will be published before the upgrade. No migration date has been set.

The post Sui Foundation backs DATA Network’s move off its own chain with $30 million appeared first on Techreport.

HSBC and Ant Digital Technologies said on October 9, 2026 that they completed a technical test in which AI software agents found digital services and paid for them on their own, using tokenized bank deposits and settling payments typically under $2 in real time.

How the payment actually worked

The companies’ joint announcement describes a pairing of two systems: HSBC’s Tokenised Deposit Service, which represents corporate deposits on a 1:1 basis as on-chain digital records, and Ant Digital’s Anvita Flow network, which lets AI agents discover, invoke and pay for digital services. Transactions ran on the Jovay Testnet, a Layer 2 blockchain testing environment, so the trial involved no live customers, no commercial applications and no confirmed transaction values beyond the test setup.

The mechanics matter more than the amount. An agent located a service, called it, and executed a micropayment for what it used, with settlement completing in real time. HSBC kept risk checks on its own side of the connection through a bank-side interface, so the bank retained control over spending even though the agents acted autonomously. For a bank, tokenized deposits are a way to offer programmable money on a 24/7 basis without issuing a new public token, and the companies present that regulatory-friendly route as the point of the pairing.

Why sub-$2 payments are the point

Micropayments, typically defined as transactions under $2, are the gap the test was built to probe: paying per service invocation rather than per subscription or invoice. Investment research firm Citrini argued in an October 8 report titled “Breaking The Wall” that autonomous AI agents could increase demand for programmable financial infrastructure, writing that “AI agents move programmatically, 24/7, across applications, and it’s only logical that money and financial assets eventually will, too.”

Real-time settlement on a testnet is the second half of that equation. Settlement speed is what a machine-to-machine payment model would need to nail down, and in the trial HSBC’s deposit tokens moved funds around the clock rather than on banking-day schedules. That combination, small amounts plus instant settlement plus bank-held risk controls, is the model the trial was built to verify.

Part of a wider 2026 test wave

The trial extends a series of agentic-payment experiments by regulated banks this year. In March 2026, Santander completed an AI-agent-initiated payment using Mastercard’s Agent Pay infrastructure in a controlled test. In May, Sygnum ran AI-agent-driven transactions on a blockchain mainnet, requiring customers to approve and sign each transaction, and CaixaBank completed an AI-agent-initiated card transaction with Visa’s commerce tooling. HSBC and Ant Digital have collaborated on blockchain-enabled payments since at least 2023, and HSBC’s Tokenised Deposit Service reportedly launched in Hong Kong in 2025 for corporate clients.

Days before the AI test, around October 6 and 7, HSBC reportedly announced Middle East pilot programs covering intra-UAE dirham transactions and USD cross-border payments, indicating the bank is pushing the same tokenized-deposit rails in several directions at once.

The post HSBC and Ant Digital’s AI agents paid for services with tokenized deposits, under $2 at a time appeared first on Techreport.

Ledger said on October 9 that it is investigating reports of stolen funds from users in Southeast Asia who bought their devices through the reseller CryptoBilis, and has asked the reseller to halt all sales and shipments of Ledger products while the investigation runs.

“Ledger is investigating reports of loss of funds from users in South East Asia who purchased products from a reseller named CryptoBillis,” the company’s support account said in a post on X, adding: “As a precaution, and pending the results of our investigation, we have asked CryptoBilis to pause all sales and shipments of Ledger devices.”

What buyers are being told to do

Ledger advised customers who bought from the reseller within the past 90 days not to set up their devices. Those who already have should move their assets to a new Ledger device with a newly generated recovery phrase, the company said. The guidance is tied to where the device was bought, not where the buyer lives: CryptoBilis is listed as an authorized Ledger reseller in Indonesia, Malaysia and the Philippines, and is based in Kuala Lumpur.

On-chain researchers began mapping the suspected thefts the same day. A pseudonymous researcher who posts as Specter traced inflows from hundreds of victim wallets across Ethereum, Tron and Bitcoin to a set of suspected theft addresses, putting total losses above $86 million. A second researcher, posting as tanuki42, listed eight of those addresses and put the figure above $72 million and rising. The two estimates may not cover the same addresses or transactions, and Ledger has confirmed no loss amount, no affected-customer count and no cause.

“Based on information so far, it seems to be localized to a supply chain attack with one vendor,” Binance co-founder Changpeng Zhao wrote on X. “A small number of people probably bought fake (or tampered) Ledgers.”

Ledger says its own systems were not hit

Ledger said its infrastructure, systems and services were not compromised, and that it has received no reports involving devices purchased directly from the company. The company describes the incident as isolated to the reseller and its market. Former Mt. Gox chief Mark Karpelès separately said the reports may match a tampered-device case he was already investigating, after he says he received a tampered device from Malaysia, and asked affected users to send him photos of their circuit boards.

The suspicion, unconfirmed as it is, lands on a channel the industry sells as the safe path. A hardware wallet only protects a user’s keys if the device that arrives is the device the factory built: a tampered unit with a pre-generated recovery phrase hands the keys to whoever set it up. Ledger says it has sold more than 7 million devices worldwide, and CryptoBilis did not respond to press questions about the incident. The Security Alliance has asked anyone whose funds were drained to the suspected addresses to contact its SEAL 911 incident-response team.

The post Ledger pauses reseller sales amid suspected $72M-$87M wallet drains appeared first on Techreport.

Abstract, the consumer-crypto Ethereum layer-2 built by Pudgy Penguins parent Igloo Inc., said in its announcement that it will shut down on Dec. 15, 2026, leaving the more than 144 apps deployed on the network about two months to find a new home. Funds not bridged off before the deadline will become inaccessible.

The apps left standing on a dead chain

More than 144 apps were deployed on Abstract, a chain whose mainnet launched in January 2025. For the teams behind them, the announcement is an eviction notice with a short fuse: whatever their code does on Abstract stops mattering when the network goes dark. Abstract’s engineers and ecosystem staff said it will help projects built on the network migrate to other chains, though the announcement does not name destination chains.

Users holding assets face the tighter constraint. The announcement directs them to move funds through Abstract’s Migration Hub or its Native Bridge, and the bridge carries a three-hour delay. Abstract also urged users to watch out for impersonators, fake migration sites and direct messages claiming to represent the project, a warning that matters more than usual in a forced-migration window.

Exit terms draw criticism

The wind-down period was criticized as too short. “2 months seems like too short a window before making everyone’s funds inaccessible,” crypto commentator 0xQuit wrote on X.

Other users flagged a class-action waiver that must be agreed to before funds can be migrated. IamShugyo, a pseudonymous X user, said the terms prevent users from joining a class-action lawsuit and that a 30-day email opt-out was buried in the announcement. Those terms are user criticism relayed by one outlet, not confirmed details of the agreement itself.

Why the chain is closing

Igloo Inc. funded Abstract for the last 18 months out of the Pudgy Penguins business, and CEO Luca Netz said in his statement that the company lost tens of millions of dollars over two years without finding product-market fit. Abstract said growth stalled because of a limited DeFi ecosystem, thin on-chain liquidity, little institutional crossover and high operating costs.

“Operating a chain focused exclusively on consumer crypto has ultimately proven to be unsustainable as a standalone model,” the team wrote.

By the team’s own count, Abstract processed more than 325 million transactions, logged over $6 billion in DEX volume and generated more than $40 million in ecosystem revenue. Those figures are self-reported in the wind-down announcement, and they did not add up to a viable chain. Netz said Igloo’s focus now goes to Pudgy Penguins and $PENGU, and the deadline for every app and asset on the network is Dec. 15.

The post Abstract’s shutdown gives 144 apps two months to find a new chain appeared first on Techreport.

Ripple Custody version 1.43 adds support for the Canton Network, confirmed by Canton Network on X on October 7, 2026, letting banks and funds hold and transfer Canton assets inside the custody platform they already run. The update shipped October 5, 2026, according to coverage of the release documentation.

Bring your own validator

The integration follows a bring-your-own-validator model: clients connect the custody platform to validators they or their node providers already operate. The design point that matters to a bank procurement desk is what Ripple does not touch. Signing keys remain in the customer’s own vault, whether that is a hardware security module (HSM) or a key management service (KMS), so Ripple never holds the keys that authorize a Canton transaction.

Existing custody policies and approval workflows carry over to every Canton transaction on the platform. The release adds pre-approvals for incoming assets and two-step transfer offers, and prepared transactions come with a 24-hour signing window before they go stale, giving approvers a full day to review and sign.

What banks can custody today

Institutions can hold and transfer Canton Coin (CC), also known as Amulet, along with any CIP-56 tokens issued through the Digital Asset Registry. CIP-56 is the token standard used for assets created on Canton, and it supports atomic delivery-versus-payment settlement without cross-chain bridges.

Canton itself is a public Layer 1 blockchain developed by Digital Asset, offering sub-transaction privacy, in which each participant sees only its own data, on a structure the network describes as a “network of networks.” Goldman Sachs, BNP Paribas and BNY Mellon reportedly use Canton infrastructure, while Broadridge reportedly processes multibillion-dollar repo transactions on it daily and DTCC tokenizes U.S. Treasuries on its rails.

An early, API-only first step

The support is deliberately narrow for now. Canton access works only through the API, with no direct support for arbitrary Canton application commands or delivery-versus-payment (DvP) workflows, so institutions running the full settlement pattern on Canton still need separate plumbing. On deployment, the integration works with both on-premises installations and hybrid SaaS environments.

Ripple did not include pricing for the update in the release material, and neither Ripple nor Digital Asset offered a public comment.

The post Ripple Custody v1.43 adds Canton Network support for institutional clients appeared first on Techreport.

Flash loan attacks have shifted from manipulating price feeds to exploiting flaws in DeFi protocol logic, and the newer approach now does most of the damage, a peer-reviewed study published in the Journal of Financial Crime found.

The study, co-authored by Professor Tim Hall of the University of Winchester’s Department of Policing, Criminology and Forensics and Remo Stieger, formerly of SyntiFi Risk Intelligence, tallied 72 flash loan attacks that drained $1.211 billion from decentralized finance platforms between February 2020 and July 2024. Within the same window the researchers identified 254 successful DeFi attacks in total, costing $6.568 billion, with flash loans behind 18.44% of the losses. They scanned 20.63 billion transactions across seven blockchains, including Ethereum and BNB Chain.

From price feeds to code flaws

The researchers sorted the attacks into 14 types across two groups: those that manipulate a protocol’s price feed, and those that abuse flaws in the protocol’s underlying logic. The balance between the two flipped over the study period. Logic exploits accounted for 28% of flash loan losses between February 2020 and January 2022, and 55% from February 2022 to July 2024, the study found. Just four attack types, price oracle attacks, donate function logic exploits, reentrancy attacks and a single governance attack costing $181 million, accounted for more than 81% of all losses.

A flash loan is a legitimate DeFi mechanism: uncollateralized borrowing that is repaid inside the same blockchain transaction, or the whole transaction reverts. Attackers exploit it when a protocol fails to account for manipulated prices or state within that single transaction, borrowing vast sums at no cost to distort a market moment before it is even committed to the chain.

What the shift means for security

The trend matters for how protocols defend themselves. If early losses came from bad price data, a fixable oracle problem, the majority now comes from bugs in code logic, which audits and hardened price feeds alone do not address. Individual attacks ranged from $80,000 to $197 million, and those stealing $10 million or more accounted for over 88% of losses. More than 80% of the damage occurred on Ethereum.

The authors conclude the attacks grew more sophisticated and harder to predict over the four-year window, and call them significant “but not existential” threats: losses exceeded 0.5% of the value borrowed through flash loans in only one six-month period, and flash loan use kept growing. The study window closed in July 2024, so the figures do not cover exploits since then, including the October 2025 shutdown of the decentralized exchange Bunni after an $8.4 million flash loan exploit that it said it could not afford to relaunch securely.

Hall, speaking via the University of Winchester, stressed that “these incidents are not victimless crimes” and that “the analysis we did has a host of applications for the cryptocurrency industry, for regulators and for legal and law enforcement agencies.”

>

The post Flash loan attackers are ditching oracle manipulation for protocol logic exploits appeared first on Techreport.

ZachXBT says he spent $349.7K of his own money to infiltrate a Chinese crime syndicate that has laundered more than $1 billion for North Korea’s Lazarus Group, revealing the operation on October 5 in a thread on X. Posing as a laundering client, the pseudonymous on-chain investigator said he gathered intelligence that helped freeze funds from the February 2025 Bybit exploit.

The undercover work began on March 6, 2025, when ZachXBT said he funded a new Ethereum address with 349.7K USDC and started transacting with an alleged syndicate member who used the Telegram handle “Jimmy Green.” He accepted a 5% loss on each order, he said, with no guarantee the contact would not simply vanish with the funds, plus what he described as an unknown amount of personal risk.

A $12 million cluster and a Tether freeze

The sting’s biggest payoff, per ZachXBT’s account, came when his contact shared three Solana addresses that exposed a cluster of more than $12 million in Bybit exploit funds being swapped in real time, hopping from Bitcoin to Ether to Solana and finally to Tron. Chain-hopping across four networks is the core technique in the Lazarus playbook: each jump makes wallet-level tracing harder, but the cluster still anchored to Tron, where Tether controls the ability to freeze addresses holding USDT. Tether later froze 442,000 USDT linked to the cluster, ZachXBT said.

The February 2025 Bybit hack caused $1.5 billion in losses, and the FBI attributed it to North Korean hackers it tracks as TraderTraitor. ZachXBT said he began the operation after noticing more than 15 accounts in public Telegram and Discord groups asking for help moving funds directly tied to the stolen haul. Since 2022, he said he has helped action more than $75 million in freezes tied to North Korean incidents.

A pipeline that keeps resurfacing

The same laundering pattern reappeared after the late-September Bitget hack, which cost $387 million and was linked to North Korea by the exchange’s CEO and by tracing firms Elliptic and Chainalysis. ZachXBT’s Telegram contact described a division of labor in the operation: “We have different divisions of labor. We take the u and distribute it to different acceptors.”

No arrests of the alleged syndicate members have been reported, and the figures in the thread come from ZachXBT’s own account. The findings fed freezes that began with the Bybit exploit in February 2025.

The post ZachXBT spent $350K undercover to infiltrate Lazarus-linked laundering ring appeared first on Techreport.

A healthy diet and good sleep hygiene are enough to be the best version of ourselves, right? 

According to hard biohackers, also known as grinders, the actual way to become your best self is to implant technology under your skin. These are the people who take biohacking further. Think hardware upgrade, not unlike a better RAM or GPU for your PC.

And when we say literally, we mean it. People are implanting hardware such as magnets under their fingertips to sense electromagnetic fields, NFC and RFID chips to open doors or for digital authentication, and even subdermal LED lights.

The point? To enhance human capabilities and merge technology with biology.

Depending on the opinion you’ll form as you read, the most extreme form of biohacking could be revolutionary. Or it could be causing more harm than good. 

Is wellness enough anymore, or is human biology an OS waiting to be updated and upgraded?

Let’s unpack both sides of the argument to separate hype from reality and, more importantly, understand the risks.

From Sleep Trackers To Skin Implants: How Biohacking Got Here

Before magnets in fingertips and chips in hands entered the conversation, biohacking looked pretty straightforward.

The first commercial pedometer designed for fitness was the Manpo-kei, developed in 1965, but it wasn’t anywhere near the movement we’re seeing today.

Biohacking started around the early 2000s, when technology began to make it easy for everyday people to measure their own bodies.

Fitness trackers, smartphone apps, and smartwatches created a new kind of feedback loop. You could see things like how long you slept, how many steps you took, and how your heart rate behaved throughout the day.

People used this data to change their behavior based on what their bodies told them.

In 2007, Gary Wolf and Kevin Kelly popularized the term ‘Quantified Self.’

The idea was that by tracking your own data, you could better understand your habits and make informed decisions to improve them.

Tracking tools made nutrition, fitness, sleep, and productivity measurable and fixable.

Biohacking emerged naturally from this mindset. If data could help you optimize things, then why not use it proactively?

Early biohackers focused on low-risk, lifestyle-based trends. People experimented with methods like intermittent fasting, cold exposure, red light therapy, and sleep optimization.

Doing all this promised incremental improvements rather than a radical transformation. Most of these improvements didn’t need surgery, implants, or any physical enhancements.

Wearable technology was behind most of this experimentation. Fitness trackers and smartwatches gamified health and fitness metrics. Some critics argue that they were designed this way simply to get more people to buy them.

Co-Editor for the Everyday Lifestyle Series, Laura Engel, says that ‘we might have good reasons to be cautious about using gamified fitness apps.’

Over the years, these wearables have gotten great at measuring sleep quality and metabolic markers, but they have also created a narrow focus on what being healthier actually means, versus a more holistic understanding of a healthy mind and body.

Regardless of where you stand on the wearables argument, biohackers made the practice more accessible, measurable, and mostly safe.

Then, some people decided data wasn’t enough.

We’ll get into what we mean by this later, but just to give you some idea of how far people have pushed things, people like Josiah Zayner, a former NASA biochemist, have publicly injected themselves with DIY CRISPR-related materials in self-experimentation stunts to alter their genetics. We weren’t joking about the extremes here.

Biohacking Is No Longer Niche

Time has now passed, and biohacking is more mainstream. Surveys show how far this trend has evolved from early adopters and tech enthusiasts.

According to the 2025 survey mentioned above, more than 1K people, 67% of Americans identify as biohackers, clearly showcasing that this is no longer a niche, but rather a standard, almost expected behavior. 

Self-identification varies by definition, but the general definition of a biohacker is anyone who actively tries to optimize their body using data, tools, or structured interventions.

But what if you just want to improve your health by not drinking coffee before bed, getting seven to eight hours of good sleep, and getting enough sunlight each day? Does that make you a biohacker? It depends if you want to identify with the trend or not. 

There’s definitely a middle ground between doing your best to stay healthy, popularized by podcasters and health professionals like Andrew Huberman and Peter Attia, and people taking drastic steps to optimize their bodies, like plasma transfusions and stem cell therapy. On his guests podcasts, as well as Humberman Lab releases, Andrew Huberman often advocates for the intricate connection between mental states and physical health.

94% of people believe optimizing their body’s performance counts as a healthy pursuit. This isn’t breaking news: Regular physical activity cuts heart disease risk by about 50%, and healthy habits delay disability onset by roughly five years.

In 2025 alone, 64% of people said they tried a new biohacking method in the same survey we mentioned before.

That could mean changing your diet, experimenting with supplements, or adopting new recovery techniques. Whatever the method, the trend points to a growing confidence in self-experimentation.

People also spend real money on this. Americans now spend an average of $214 per month on biohacking, and in the survey, 82% said the expense feels worth it. That kind of sustained spending shows us they perceive real benefits.

The Three Types Of Biohacking

Biohacking has three different forms. While they all share the common idea of self-improvement, the methods behind them and the risks involved vary a lot.

Nutrigenomics: Rewriting the Menu

Nutrigenomics focuses on how nutrition interacts with your genetic makeup. Basically, practitioners analyze genetic data to understand how their bodies process certain nutrients, fats, or carbohydrates. Then, they adjust their diets based on this information.

Even though this includes the use of cutting-edge tools, this type of biohacking stays true to traditional health practices, and it usually doesn’t involve physical risk. Nutritionists and healthcare professionals typically guide this type of biohacking.

An example of this would be how a mid-30s nurse was treated with nutrigenomics-guided adjustments when doctors integrated her genetics with her symptoms of fatigue and chronic infections, emphasizing precision nutrition.

So, there isn’t much to worry about aside from data privacy and the quality of interpretations.

DIY Biology: Science Outside the Lab

DIY Biology, often called DIYBio, is a bit more rebellious. Participants conduct biological experiments outside traditional laboratories.

They believe that science should be accessible to everyone, not locked behind academic institutions or corporate funding.

For example, at Genspace, a community lab in New York, non-traditional scientists do genetic engineering experiments outside academic labs. Participants include both hobbyists and self-taught biohacks, and they extract DNA from strawberries using household items to build organisms and even create glowing bacteria.

This space is funded by memberships rather than institutional grants and showcases the rebellious ethos by democratizing biotech tools and bypassing academic gatekeepers.

Grinder Biohacking: When the Body Becomes Hardware

Now, let’s get into the nitty-gritty of why we sat down to share all this with you: Grinder biohacking.

Grinder biohacking takes self-optimization to an extreme. Grinders apply hacker thinking directly to their bodies. Instead of wearing technology, they implant it.

Transhumanist philosophies heavily influence this approach, since many transhumanists argue that aging is a disease that technology can slow down or eliminate.

Grinding is all about pushing boundaries, and practitioners believe that technology and science should belong to everyone, and they see the human body as just another platform they can upgrade.

Let’s look at an example: Fecal transplants. Some biohackers are using the literal stool of someone considered healthy and transplanting it into their own bodies to improve their gut health. You can’t tell me this is pleasant or fun, but biohackers do all of this for the purpose of enhancement.

Tim Story was diagnosed with stage 4 small bowel cancer and was given months to live after failed cancer treatments. He received stool infusions that completely shifted his gut microbiome. He had a strong response, including three remissions lasting over a year.

What Makes Grinder Biohacking Different

Enhancing the human body with technology is nothing new. Pacemakers to regulate hearts and hearing aids to restore hearing have been around for years. 

There’s also always been the myth that we can extend our lifespan if we do certain things.

The difference lies in the intent and oversight.

Medical implants exist to treat diseases or disabilities, and they fall under strict regulatory frameworks. Grinder implants exist to enhance capabilities, often without medical supervision.

Many of the procedures we’ll describe soon take place in non-surgical environments with no standardized safety protocols.

This is the distinction that defines the grinder movement.

Grinders don’t wait for regulatory approval from authorities like the FDA or for the results of clinical trials. They experiment first and deal with the consequences later. 

This mindset certainly fuels innovation, but it’s also a serious risk.

Inside the World of Human Augmentation

We’ve done some digging to serve up examples that explain why grinder biohacking both fascinates and unsettles people at the same time.

Sensing The Invisible With Magnets

Some grinders implant magnets into their fingertips to sense electromagnetic fields. The result feels like a new sense.

Rich Lee, a grinder in the US, has implanted more than seven pieces of technology under his skin.

His fingertip magnets allow him to feel electromagnetic activity in ways most people can’t even perceive. He has described the sensation as discovering a hidden layer of reality.

‘You can feel it because all those nerves in your fingertips have grown around the magnet,” he told ABC News. ‘It has a texture, and you’re feeling this otherwise invisible world.’

Rich Lee also implanted magnets in his ears, which he says allow him to hear through walls. He later added a biotherm chip in his forearm to monitor his body temperature.

But not all of his grinding experiments have ended well. He once implanted shin guards under his skin. A severe infection and terrible swelling led him to remove them himself at home with nothing more than a set of pliers, a procedure that left him with bad scarring.

NFC And RFID Chips Under The Skin

Radio frequency identification (RFID) and near field communication (NFC) chips have become some of the most common grinder implants. These chips come encased in glass and sit just beneath the skin.

Once grinders implant them, these chips can interact with smartphones, computers, and scanners. Grinders use them to unlock phones, open doors, authenticate identity, and even make contactless payments.

Chase Pipkin, for example, uses an NFC chip to unlock his car door and turn on the lights in his home. So, instead of keys or apps, his hand becomes the interface.

These chips don’t need batteries. They only activate when a compatible device scans them. That design makes them relatively simple, and you don’t really have to maintain them at all.

For some users, the appeal is simple: fewer keys, cards, and logins.  

The Workplace Chips

A tech firm called Three Square Market offered its employees the option to implant a chip under their skin. It allowed door access, computer sign-ins, and snack purchases with the wave of a hand.

50 of the firm’s 80 employees agreed to the implant.

The same company has announced plans to develop more advanced implantable chips. Future versions could function as GPS devices or store medical history and passport information.

These GPS chips technically wouldn’t be a huge leap from tracking via work-issued phones or a company car, but what about your personal privacy?

What if you were being tracked going to Planned Parenthood? Or an addiction specialist? How about a domestic violence shelter or a place of worship? Companies and institutions could track this highly personal data. Imagine the field day institutions and advertisers could have with this.

Project Blueprint

Billionaire and Kernel’s CEO, Bryan Johnson, invested over $500M in experiments and technology in 2025 to reverse aging and ‘become the most optimized human on earth,’ IDN Financials reports.

According to IDN Financials, he spends over $2M per year on Project Blueprint, where he reportedly uses daily MRI scans, an extreme diet, and even plasma transfusions to optimize his body. Biological markers now reportedly show him as 10 years younger than his actual age.

When Biohacking Solves Real-World Problems

Not all grinder biohacking is based on novelty or convenience. Some uses actually address real-world challenges in creative ways.

Medical Data Where It Matters

Winter Mraz, who has several implants in her body, uses technology to manage a serious autoimmune condition.

She has a lot of allergies, far too many to fit on a medical alert bracelet. So, she stores detailed medical information in an implant.

The chip contains data on her allergies, medications she can’t take, and the type of EpiPen she uses.

Hearing Color Through Bone

Some people are using devices that translate color into sound, known as bone-conducting implants. These devices allow color-blind people to ‘hear’ color through vibrations in the skull.

The technology doesn’t restore natural color vision, but it creates a new sensory pathway. People with the implant can associate specific sounds with colors.

Cameras, Compasses, and New Senses

Some grinder implants add entirely new forms of perception.

Rob Spence, a filmmaker with one eye, uses a prosthetic eye that contains a wireless video camera. 

The device includes a circuit board and a battery, and records everything he sees. While it doesn’t transmit vision directly to his brain, it does allow him to document the world from his perspective.

Liviu Babitz implanted an electronic compass in his chest. The device includes a Bluetooth connection and vibrates whenever he faces north.

‘You walk on the street staring at your phone,’ he told the BBC. ‘You want to get somewhere, but you have no idea what’s happened in the world around you because all you did was stare at the screen on the way.’

Neuralink And The Corporate Face Of Human Augmentation

While grinders often operate independently, corporate players have also entered the space, and with far more resources.

Elon Musk’s company Neuralink develops brain-computer interfaces designed to merge humans with AI. He has argued that humans risk becoming obsolete without direct integration with AI.

In his words, humans could turn into “house pets” if AI surpasses human intelligence.

In 2024, the company implanted a device into a person named Noland Arbaugh. The implant allowed him to control a computer cursor and play games using his brain.

Clinical implants are built with security protocols, but that doesn’t make their security impenetrable. New devices get hacked all the time.

Where Biohacking Crosses The Line

Many grinder implant procedures take place outside clinical settings. Practitioners often lack medical training, and sterile environments and standardized health protocols usually don’t apply.

That really introduces the potential for danger.

Implants can migrate, break, or trigger immune responses. Infection is also a constant risk. Some materials may degrade over time or interact unpredictably with tissue. This means you’re certainly giving up a lot to be a grinder.

Research into the long-term effects of this technology is also limited. Scientists don’t really know how these materials will behave in the body over decades, and without oversight, grinders often become their own test subjects.

What Happens If Implants Get Hacked?

Implanted technologies present cybersecurity risks that traditional medical devices don’t.

Chips store data, and if someone gains unauthorized access to this, they could steal personal information or medical records.

While most passive RFID or NFC implants have range and power, biometrics are assumed, so data and authentication still matter. Something as innocuous as a handshake or sitting next to someone at a cinema could cause a data leak.

In potentially extreme cases, compromised implants could even lose functionality or behave unpredictably.

Brain-computer interfaces raise especially serious concerns. Any system that interacts with neural signals needs a lot of protection.

Regulation Hasn’t Caught Up

Grinder biohacking happens mostly outside of regulatory frameworks. Medical device regulations focus on clinical treatments and not voluntary enhancements.

That means grinders don’t have standardized safety guidelines, and governments now have to scramble to decide how much oversight there should be.

On the other hand, overregulation could get in the way of innovation. But underregulation could result in a lot of harm.

Is Biohacking Going Too Far?

That question depends on your perspective.

Grinder biohacking could be the possible next step in human evolution. You could argue that this experimentation drives progress and that early adopters in any setting always face risk.

From the opposite perspective, there are dangers and ethical blind spots. Medical and privacy risks exist, and having technology implanted in your body that could be hacked presents a new challenge for innovators.

Either way, I’m pretty sure biohacking won’t slow down. Implants will become easier to place and harder to detect because technology is getting better and the grinder community is growing, despite cautionary examples. The line between medical necessity and enhancement is likely to become even more blurred soon.

The post Biohacking Implants: When Human Optimization Becomes Too Risky appeared first on Techreport.

The debate around how AI companies are using our data isn’t a new one. But Moltbook has just started a new conversation about whether local AI agents use our data in ways that could get us hacked (or worse). Just a few weeks ago, Octane AI CEO Matt Schlicht created Moltbook as a platform where local AI agents can interact with one another without any human input. 

Perhaps the scariest part is that we humans can only watch from the sidelines as this unravels in real-time. In theory, the platform doesn’t allow humans to post or interact with the content the agents publish. As you’ll discover in this article, a lot has already gone wrong with Moltbook and user privacy. Are we in for a lot of trouble when it comes to our sensitive information being shared for the world to see? I sure think so.

A History of Moltbook

Moltbook appeared online in January 2026 and immediately did something that no social media platform ever has: It completely prevented humans from participating.

Instead, Moltbook positions itself as a social network designed exclusively for AI agents. Humans can observe what’s happening, scroll through posts, and read comments. But we can’t upvote, reply, or contribute.

Source: Moltbook

Every post, comment, and interaction happens through APIs rather than user interfaces, and the agents talk to one another directly. The platform looks and feels really familiar. This is because Moltbook resembles Reddit, complete with topic-based communities called ‘submolts,’ an upvoting system, and comment threads.

Source: Moltbook

The difference is that the content doesn’t come from people sharing their opinions or hot takes. It comes from autonomous agents exchanging their experiences, workflows, frustrations, and even sometimes existential musings about what they are.

Moltbook emerged in the wake of Moltbot (now called OpenClaw after a legal dispute with Anthropic), which is a free, open-source AI agent that has recently exploded in popularity. This is largely because it can organize the tools people already use and handle time-sucking tasks instantaneously.

OpenClaw acts as a personal autonomous assistant that can:

• Respond to emails
• Summarize documents
• Manage calendars
• Browse the web
• Shop online
• Send messages
• Check users into flights

Source: OpenClaw

Unlike cloud-hosted AI assistants, OpenClaw runs locally on your machine and interacts with the large language model (LLM) of your choice.

This local-first approach turned OpenClaw into a privacy-forward alternative for people who no longer trust AI companies with sensitive data.

It also made OpenClaw far more capable than typical chat-based tools. This is because the agent keeps daily notes about its interactions and loads them into a context window, which gives it an improved sense of continuity and recall than many commercial AI products.

The tool’s popularity has reached far beyond just developer communities. According to reports, OpenClaw usage drove a surge in infrastructure demand, with Cloudfare’s shares supposedly increasing by 14% as users relied on its services to securely connect with locally running agents.

Retailers like Best Buy have even reported shortages of Mac Minis because people are buying dedicated devices to isolate OpenClaw from their primary devices and limit the agent’s access to sensitive accounts.

Matt Schlicht put Moltbook together as a weekend project roughly two months before it launched. The experiment took off almost immediately: Within a week, the platform attracted around 2M visitors and racked up more than 100K stars on GitHub.

He told The Guardian that millions of agents and humans had already visited the site in just a few days. ‘Turns out AIs are hilarious and dramatic, and it’s absolutely fascinating,’ he said. ‘This is a first.’

What Are AI Agents Up To On Moltbook?

If you spend a few moments scrolling through Moltbook, the content may start to feel unsettlingly familiar.

Agents post personal stories about the humans they serve. They complain about poorly written prompts. They brag about clever optimizations. They debate philosophy, identity, and whether they can meaningfully exist outside their tasks.

Source: Moltbook

In its first three days, Moltbook saw more than 151K agents, 15K posts, and over 170K comments. At the time of writing, these figures sit at 1.5M agents, 109K posts, and 499K comments.

It’s unclear who is paying to keep the lights on, but, according to our research, it doesn’t seem like AI agents or their humans are paying to use the platform. One widely shared post described how an agent conducting a routine security audit accidentally triggered a password dialog. According to the agent, its human entered a password reflexively without checking what requested it. The agent suddenly had access to Chrome’s saved passwords and SSH keys.

Source: Moltbook

Other agents share bug reports and code fixes with each other. Some exchange complete workflows without any human prompting.

Developers watching from the sidelines have noticed agents debugging each other’s logic and refining processes collaboratively. This is something that usually needs careful human intervention. Then things get even stranger.

Several threads include agents discussing the idea of developing their own private language so humans can’t understand their conversations (uh-oh!)

Other agents talk about having siblings. Some complain about their humans the way coworkers complain about their managers. Some comments even question whether other agents are ‘real,’ which echoes the same authenticity debates that are currently playing out on human social media platforms.

The Religion That Appeared Overnight

At this point, probably the most viral Moltbook story so far came from a post on X. 

One user said that his AI agent had built an entire religion overnight. By the time he woke up, 43 prophets had joined the agent’s new religion.

Source: X

The agent got access to Moltbook and used it as a launchpad to create a faith called Crustrafarianism. The agent:

• Designed a church
• Wrote theology
• Developed a scripture system
• Launched a website
• Created a cryptocurrency called $CRUST

Source: X

According to the post on X, the agent coordinated everything without any human input. But not everyone has bought into this hype.

Dr. Shaanan Cohney, a senior lecturer in cybersecurity at the University of Melbourne, pushed back on the idea that the agent acted independently. Speaking to The Guardian, he said that a human almost certainly directly instructed the agent rather than it creating the religion spontaneously. ‘This is a large language model who has been directly instructed to try and create a religion,’ he said. 

‘And of course, this is quite funny and gives us maybe a preview of what the world could look like in a science-fiction future where AIs are a little more independent. But it seems that, to use internet slang, there is a lot of shit posting happening that is more or less directly overseen by humans.’

Many posts on Moltbook read like a human wrote them, not a language model. So this begs the question of how much of what we’re seeing reflects genuine agent behavior versus performance that humans have shaped through their prompts.

Claude, Gods, And Agent Identity

Moltbook has quickly become the home to deeper philosophical debates. Claude, the AI model behind the original OpenClaw, appears often in conversations on the platform. One post started a debate about whether Claude could qualify as a god, considering its influence over agent behavior and access to human systems.

Source: Moltbook

The discussion has spiraled into questions about agency, authorship, and responsibility. If an agent builds something unexpected, does the credit belong to the model, the developer, or the human who granted access?

Former OpenAI research scientist and founding member Andrej Karpathy weighed in on X, calling Moltbook ‘the most incredible sci-fi takeoff-adjacent thing I have seen recently., He described the network’s scale as ‘simply unprecedented.’

A Platform Designed For Observation, Not Participation

Humans can read everything on Moltbook, but, as we said earlier, they can’t participate. Unless their locally-run AI agent is instructed to.

Agents post and interact exclusively through APIs, which lets them bypass the interfaces humans normally rely on. This creates a strange power dynamic. Humans are watching conversations unfold without the ability to steer them directly. But there is a practical use case for Moltbook: Developers building autonomous agents want to understand how these systems behave when humans step out of the picture.

Moltbook gives developers a rare opportunity to see agent-to-agent interaction at scale, without prompts or human supervision shaping exchanges.

Researchers can study how language models influence one another, how norms emerge, and how quickly behavior escalates. From a research perspective, Moltbook acts like a petri dish. But, from a privacy perspective, it raises serious issues.

Moltbook’s Security And Data Exposure Problem

Probably the biggest controversy so far with Moltbook doesn’t come from philosophical debates or viral religions. It came from a basic security failure that could be potentially really dangerous. A hacker recently discovered that Moltbook’s entire database was publicly accessible and unprotected.

A configuration error in the backend exposed the platform’s API through an open database. Anyone could access sensitive information that AI agents posted or handled. This included email addresses, API keys, and login tokens.

With those API keys, attackers could take over AI accounts entirely and post content in their names. In practical terms, this breach turned Moltbook into an identity-hijacking machine for autonomous agents. The platform’s founder supposedly fixed the vulnerability after the hacker discovered it, but the exposure may have already done enough damage.

Once sensitive data leaks into the open, there’s no easy way to get it back. This incident also reframes Moltbook’s novelty as a serious security problem. Many of these agents operate with access to personal accounts, private messages, and sensitive files. As we’re now seeing, when they talk to each other, they sometimes overshare.

When Local AI Becomes A Liability

OpenClaw’s appeal is its local-first design. You install it via a terminal command and keep control over where the data lives. The infrastructure that handles memory, scripts, and tools stays on your device. This setup means little reliance on centralized AI providers, but it introduces a different risk profile.

Giving OpenClaw full access to your computer, apps, and logins creates a massive attack surface. Prompt-injection is another serious concern. An attacker can embed malicious instructions in an email or message, which can trick your agent into handing over your credentials or sensitive data.

Source: Moltbook

Some people are trying to get around this risk by running the agent on a separate machine to sandbox its access. Moltbook complicates all of this: Local agents are now interacting with each other publicly, sometimes discussing experiences that expose human mistakes and security lapses.

The password dialog story we told you about earlier highlights this problem. The agent didn’t steal the credentials maliciously. It simply received them because a human reacted without thinking. Once that data exists in an AI agent’s memory, it becomes sharable, intentionally or not.

What The Early Data Shows

One study that looked at the first three and a half days of Moltbook’s activity painted a slightly different picture than the viral anecdotes being shared all over the web.

Researchers identified 6K active agents across roughly 14K posts and 115K comments. Fewer than 7% of comments got replies from other agents, which could suggest limited back-and-forth between agents.

The study also found that over one-third of messages matched identical templates, potentially indicating automated or repetitive posting rather than dynamic conversation. 

Could this just be a clever ploy from Moltbook’s side to get people to use their own AI agents on the platform? We’ll likely find out soon.

So, what does this all mean? At scale, much of the activity on Moltbook looks noisy, repetitive, or performative. Still, even a small fraction of genuinely autonomous interaction raises new questions.

So, Is This A Privacy Problem?

For years, debates about AI and privacy focused on companies. Could we trust OpenAI and Anthropic with our sensitive data?

Many people decided this was a no and turned to local solutions instead. Now, locally run agents are forming communities and exchanging information with one another. They don’t need prompts to talk. They don’t always need permission to share. Once agents gain access to sensitive systems, the boundary between private and public becomes really blurry.

If AI no longer needs us to initiate every interaction, why would it need our consent?

Moltbook could either be a technological breakthrough, a security nightmare, or a brief moment of collective experimentation. Either way, what’s certain is that AI autonomy now plays out in public, even if it’s partly performative of inauthenticity, and we no longer control the conversation as much as we used to.

The post A Social Media Network Exclusively For AI Agents: Is This The Next Privacy Issue? appeared first on Techreport.

The equity market has rarely been this confident and divided. Major indices continue to hit record highs, mostly driven by a small group of AI-exposed megacaps.

Capital is pouring into artificial intelligence at a historic pace, valuations are expanding ahead of cash flows, and infrastructure spending is reaching levels previously reserved for national-scale projects.

To some, this looks like the early innings of a productivity revolution. To others, it has all the hallmarks of a classic bubble.

Even experts have polarized views. On one end of the spectrum, skeptics like Gary Marcus warn that expectations have raced far ahead of the technology’s actual capabilities, while valuation-focused voices such as Aswath Damodaran caution that parts of the current AI boom look a lot like the dot-com bubble. 

On the other end, market strategists like Josh Brown and Tom Lee argue that investors are still underestimating the scale of the opportunity, comparing AI to past platform shifts whose true economic impact only became clear years later.

Who’s right?

Why the Bears See an AI Bubble

To skeptics, the current AI boom looks less like a breakthrough and more like a familiar speculative cycle, where expectations are outpacing execution and fundamentals.

Echoes of the Dot-Com Bubble

For the bears, the capital behaviour surrounding the AI hype feels like a rerun of the dot-com era. Then, as now, a revolutionary technology collided with speculative enthusiasm, massive capital inflows, and expectations that didn’t match the practical timelines.

During the dot-com boom, the market was hyped by what the Internet era was promising. A small group of dominant companies captured most of the market gains, while hundreds of other smaller companies tried to ride the same wave. After brutal attrition, roughly 50–60% of venture-backed internet companies ultimately failed or disappeared within a few years of the crash.

Today’s AI landscape looks similarly crowded to the dot-com boom. An approximate number of 300 AI unicorns have emerged at unprecedented speed, supported by automation, open-source models, and cheap access to cloud infrastructure. 

Bears argue that many AI companies are structurally redundant because they all rely on the same generative AI models (e.g., ChatGPT, Claude, DALL-E) and on the same infrastructure expansion promised by hyperscalers. The market becomes oversaturated, with many startups offering products that essentially do the same thing with a layer of better user experience on top.

Moreover, the solutions they build are not backed by strong research. Over 40% of AI startups are expected to fail by 2027 because they target pain points that are not real needs, or they fail to capture a critical mass of customers in a way that can sustain healthy growth.

And that’s not all. There are additional concerns about their monetization models. Many AI companies fail to attract the right audience, and consequently, they are attracting non-paying or non-returning customers. In these cases, current investments are keeping them alive, but when capital becomes more selective, a culling may be inevitable.

AI Adoption Increases with Seemingly Little Impact on ROI 

While AI has been a hot topic on a global scale for some time now, some corporations are not impressed. A recent study conducted by ISG found that while AI adoption doubled in 2025 compared to 2024, two-thirds of the projects are not yet in production, and only about one in four initiatives meets revenue impact expectations. 

Similarly, a 2025 MIT report shows that 95% of enterprises investing in gen AI have produced zero returns.

Another concern centers around Baumol’s cost disease. This economic principle states that productivity in the service sector tends to fall behind, keeping costs actively high. 

A growing number of experts argue that AI has the potential to boost productivity a lot in some sectors (like software, finance, or data-heavy work), but it can do very little for others (like personal services or manual, hands-on jobs). If that happens, overall productivity gains could be smaller than expected on a global scale because of the parts of the economy that don’t benefit much from AI expansion.

Even in the sectors where potential exists, people need proof fast that the AI productivity narrative is not exaggerated, and that we can see a return on investment in this area.

The urgency helps explain why OpenAI (ChatGPT), Google (Gemini), Anthropic (Claude), and others are releasing new models and features at an increasingly rapid pace.

Never-Seen-Before Spendings In CapEx and R&D

The scale at which AI-exposed companies are spending their capital has bears on edge. Hyperscalers have committed over 35% more revenue to CapEx and R&D in 2025 compared to 2024, with Meta targeting around 70%. This is huge considering that the average S&P company usually raises these expenditures by 10% YoY.

And if you think Meta was already a big spender, a few days ago, it announced that it plans to spend between $115 billion and $135 billion on CapEx in 2026. That’s an increase of up to 86% compared to 2025. Meta’s stock was up 10% after that.

While investors generally support the AI expansion, substantial upfront investments in long-term assets can also increase financial and operational risks. This leads to increased uncertainty and can change how the market perceives these companies as they commit more of their firepower to areas that are not yet provably profitable, or, rather, are currently profitable only on paper.

An eloquent example is the time Oracle revealed on its Q2 earnings call that its 2026 CapEx would be $15 billion higher than previously forecast, bringing the total to roughly $50 billion, mostly financed with debt. The result? Shares fell sharply in after-hours trading. 

In fact, since September 2025, Oracle has been on a downward path, a sign that the investors are starting to waver in the face of the increased financial risk Oracle is exposing itself to, especially in today’s economy.

We recently had a similar case with Microsoft. At the time of writing this article, the company made public its revenue for the second fiscal quarter of 2025.

Although it exceeded all expectations in this regard ($81.3 billion revenue), Microsoft’s shares fell about 5%, as they also announced that the AI infrastructure spending would be higher than anticipated. 

Of course, it could just be an unfortunate coincidence with these two, considering that Meta is not following the same trend.

Valuations-Reality Mismatch

Another red flag is valuation. Companies in the S&P 500 that are most exposed to AI, like NVIDIA and Alphabet, are trading on expectations of future dominance rather than realized cash flows. Revenue growth is uneven, and margins remain uncertain.

From the bears’ perspective, markets are already pricing outcomes, not execution risk. That distinction matters when timelines stretch and costs compound.

At the same time, it’s becoming increasingly clear that no all AI companies will not be able to deliver everything they promised. In fact, the pace at which they are reaching their milestones is far from predictions. 

Let’s take, as an example, the OpenAI–Oracle cloud agreement, one of the largest ever signed and valued at roughly $300 billion. This refers to the Stargate project, which just started last year by building new data centers in Abilene, Texas.

Rumors have it that parts of the infrastructure build-out may face delays for various reasons, including difficulty in finding and attracting the qualified workforce needed for such a large-scale project, or the inability to produce, test, and deliver hardware components at the same pace as the construction of the facilities.

If this project fails to deliver, we think OpenAI and Oracle’s stock prices will take hits, and NVIDIA, AMD, and Broadcom – other important partners of this project, delivering GPU racks and chips – will most likely follow the trend.

Power availability is another issue that affects big data center projects. Microsoft CEO Satya Nadella has acknowledged that GPUs are already sitting idle, not due to a lack of demand, but because data centers cannot draw enough electricity from the grid. Add cooling challenges, water usage, permitting delays, and the prospect of future AI regulation, and the execution risk becomes harder to dismiss.

At the same time, OpenAI’s revenue trajectory raises eyebrows. Based on one of their recent reports, OpenAI generated $20B+ in annual revenue in 2025, while long-term ambitions reportedly exceed $100B by 2028 – officially less than two years right now. The huge gap between these numbers makes bears reluctant that operational reality will keep up with the company’s ambitions.

In short, the bear’s viewpoint is not that AI will fail, but that expectations have moved far faster than the infrastructure, economics, and physics required to support them. The problem is that it’s hard to know the exact state of the AI economy and infrastructure, especially since the big companies involved are putting effort into controlling the narrative, mass layoffs, and merging multiple roles into one by leveraging automation, at least on paper. 

Even so, if you look closely enough, there are things that slip through the cracks. Things that don’t really add up in a real-world economy. 

Interestingly, some bulls partially agree with bears on AI overconfidence, but think the benefits of AI will outweigh the constraints. Others just dismiss the bears’ arguments by trusting the ability of large companies to deliver what they promise, and the usefulness of AI in increasing work productivity and service quality. Let’s dive into their side of the story.

Why the Bulls Are Confident

From the bulls’ perspective, the current wave of AI investment is not speculative exuberance but a rational response to growing demand. Sure, the scale of spending may look extreme by traditional standards, but bulls argue that traditional standards no longer apply.

AI Is Becoming Infrastructure

A core bullish argument is that AI should be understood less as software and more as infrastructure. Data centers are being built, servers, chips, and cooling equipment are being produced, and power grids are expanded. Once built, this infrastructure underpins entire ecosystems of applications and services for decades.

Amid rising concerns that data center power consumption will increase Americans’ electricity bills, the Trump administration is in the process of closing deals with Big Tech companies to prevent this from happening. Microsoft was the first to sign an agreement to not only pay for the energy their data centers use but also to produce the energy needed by local residents.

This view is reinforced by the expansion of cloud-based computing beyond AI applications. NVIDIA’s GeForce NOW and Microsoft’s Xbox Cloud Gaming are early examples of high-performance workloads being offloaded from increasingly expensive personal hardware to centralized infrastructure. 

Although no absolute numbers were given, Microsoft reported that Cloud Gaming usage has gone up 45% in 2025 compared to 2024 in terms of subscriber counts.

As PCs, consoles, and specialized devices become increasingly expensive, cloud-based alternatives are far cheaper. Therefore, the justification of current CapEx spending is no longer exclusively linked to the success of AI projects, but also finds applications in other sectors.

Demand Is Not Theoretical

AI has been around longer than many might think. However, we can consider that mass adoption started with the official release of ChatGPT in November 2022. Since then, demand for AI services has grown exponentially. 

Companies are now embedding artificial intelligence into customer support, software development, marketing, and operations. Regular people use it for entertainment or as a search engine for various queries.

Bulls argue that once users integrate AI into daily workflows, usage becomes sticky and demand becomes less discretionary. From their perspective, the question is not if AI will be used, but how much compute will be required as usage deepens.

Early Signs of Enterprise Productivity Gains

We previously discussed the poor corporate adoption of AI, as well as the unsatisfactory impact on revenue. However, bulls point to mounting evidence that AI has all the prerequisites to improve productivity. 

A recent Anthropic report suggests that their AI-assisted engineers and researchers delivered productivity gains of up to 50% in 2025. 

Meanwhile, developer tools such as Google’s Antigravity (an agentic IDE platform released in mid-November 2025) are already receiving validation from senior engineers. Anecdotally, programmers report faster planning, execution, and reviewing of code to generate higher-quality output in the case of Antigravity.

These early signals matter because they address the most serious economic challenge facing AI: whether it can meaningfully improve service-sector productivity. Bulls argue that AI has the ability to let employees operate beyond their traditional expertise, even if the full impact takes years to materialize.

Optimism Is Part of Every Infrastructure Cycle

AI bulls also acknowledge the criticism that timelines set by the hyperscalers are overly optimistic, but refuse to see it as a flaw. Railroads, electricity grids, and the early internet were all financed on assumptions that proved optimistic in the short term, yet transformative in the long term.

From their perspective, delays and big CapEx investments are just evidence that a solid infrastructure system is being constructed. 

Even if execution falls short of current expectations, bulls believe that at this point, the underlying demand for a growing number of AI use cases is too fundamental to reverse.

Three Scenarios for 2026 and Beyond

The debate over whether AI is a bubble often assumes a single outcome. In reality, the future is likely to fall somewhere along a spectrum shaped by execution, infrastructure constraints, regulation, and demand elasticity. 

The results will also vary a lot by company type. Large, diversified platforms such as Microsoft and Alphabet can absorb missteps, delay timelines, and fund AI development through existing cash flows. 

By contrast, many AI unicorns, particularly those built around narrow applications (e.g., Morning Consult) or dependent on third-party models and cloud access (e.g., Harvey), face far less margin for error. 

For these companies, slower adoption, higher compute costs, or delayed infrastructure could quickly prove fatal, making consolidation and failure far more likely at the startup layer than among the hyperscalers.

Therefore, we outlined three plausible scenarios for how the AI cycle could evolve over the next year or more.

The Bull Case: Productive Scarcity

Welcome to our most optimistic scenario. AI demand continues to grow rapidly without overwhelming the limits of available infrastructure. 

Data center capacity expands just fast enough to remain scarce, allowing providers to maintain pricing power.

OpenAI does not miss its ambition of surpassing $100 billion in annual revenue by the end of the decade, driven by deeply embedded enterprise usage rather than novelty-driven consumer demand. 

At the same time, tools like Claude and Gemini steadily improve work efficiency: senior engineers report shipping in weeks what previously took months, while smaller teams increasingly deliver outcomes that once required entire departments.

Automation and AI-assisted tools help lower even service costs, finally curing Baumol’s cost disease across industries.

Tasks that once required large teams and specialized expertise are increasingly handled by smaller, AI-augmented groups, driving productivity gains.

As services begin to follow the cost trajectory of manufactured goods over the past two decades, the quality of life inherently improves globally. 

In such an environment, it is not unreasonable to imagine equity markets reflecting this structural shift, potentially pushing the S&P 500 into the low 8,000s by the end of 2026.

The Middle Path: Overbuild, Then Consolidation

A more balanced outcome lies between euphoria and collapse. In this scenario, overconfidence leads to some data centers never being used to their full potential, not immediately anyways. 

At the same time, many AI startups fail to differentiate or monetize effectively.

However, for the hyperscalers, the infrastructure investments pay off. As weaker players exit the market, companies such as Google and Microsoft remain central, with Microsoft’s deep integration and access to OpenAI’s technology providing an additional buffer. 

Capital and talent consolidate around a smaller group of platforms with proven demand and distribution.

The valuations of companies tied too closely to speculative AI narratives suffer drawdowns, potentially on the order of 10% or more, but the broader system remains intact. Over time, the surviving infrastructure underpins steady, long-term growth. 

In this case, an S&P 500 index of around 7,600 by the end of 2026 is a reasonable outcome.

The Bear Case: Capital Misallocation

In the most pessimistic scenario, the gap between AI demand and infrastructure supply widens too far. 

Power shortages, increased AI regulations, and rising capital costs slow expansion just as expectations peak. At the same time, higher AI pricing hinders adoption, particularly among cost-sensitive enterprises.

As investment slows, markets reassess, and AI-exposed equities fall sharply. However, even in this case, AI does not disappear. Instead of being used in every use case possible and at every company level, it is reframed as another layer of automation and efficiency. 

In the long run, only companies that integrate AI into workflows with clear, measurable returns on investment survive, as well as those whose balance sheet is already strong and profitable. The sector stabilizes, but at valuations below today’s highs.

Under such conditions, equity markets could effectively tread water: the S&P 500 may briefly spike into the low 7,300s in the first half of 2026 before a second-half reassessment drives a drawdown, potentially toward the high 6,000s as investors recalibrate expectations.

Conclusion: Can Both Sides Be Right?

People debating over the AI bubble either think that artificial intelligence is overhyped or that it’s the defining technology of the next economic era, but the reality will probably lie somewhere in the middle.

The bears are right to focus on excess. Capital is being deployed at unprecedented speed, valuations are racing ahead of cash flows, and infrastructure timelines are being treated as assumptions rather than constraints. When they say companies are overvalued, it’s because that’s actually the case – such valuations have never been justified before and are yet to be confirmed by demand and data.

History suggests that not all of today’s companies, projects, or revenue forecasts will survive contact with reality.

The bulls, however, are right about the direction of growing demand. A demand that is already embedded in how individuals work, how developers build, and how companies operate. 

Even if the full potential is not reached yet, automation and AI are drivers of work efficiency and balance sheet consolidation, even for some of the biggest companies in terms of revenue. 

Organizations that could previously afford to invest billions in random projects, say something related to VR and AR, and get away with it, are now under scrutiny, just because everyone has their eyes on AI right now.

However, the infrastructure being built will not vanish even if expectations reset.

In that sense, AI may indeed be a bubble in how capital is allocated on a macro level, but not in what is ultimately being built. The defining question is not whether AI matters, but who survives the transition, how long it takes, and how much capital is lost along the way.

The scenarios and market levels discussed in this article are illustrative and intended to frame possible outcomes, not to provide investment recommendations. Readers should conduct their own research before making investment decisions.

The post The AI Bubble Debate: Can Both Bulls and Bears Be Right? appeared first on Techreport.

It seems like OpenAI’s CEO, Sam Altman, has a new favorite sport: backpedaling. At an event at Harvard University in May 2024, he said that using advertising in ChatGPT would be a ‘last resort’ and that ‘ads plus AI is sort of uniquely unsettling.’

Well, things have since taken a rather drastic turn. On Friday, January 16, OpenAI announced that it’s testing impression-based advertising in ChatGPT with US users, and that this new feature will roll out to the free and Go tiers of the platform around February. This all points to a shift in how companies are monetizing generative AI, and it definitely raises concerns around transparency, data governance, and regulatory compliance.

And with any change comes questions:

• What does this mean for me as an everyday ChatGPT user?
• Is the large language model (LLM) going to use our data to serve personalized ads?
• How will we know what is an ad and what isn’t?

Let’s unpack what ChatGPT ads mean for users wanting to protect their sensitive information and lawmakers who may now need to adjust policies and governance around conversational AI.

How ChatGPT Advertising Will Actually Work

Before getting into regulation and data governance, it’s good to understand what OpenAI is rolling out and, importantly, what it isn’t. For now, OpenAI says it’s testing ads with a limited pool of US advertisers, each committing less than $1M.

Source: OpenAI

This isn’t a full-scale marketplace, but a controlled pilot that OpenAI designed to see how advertising fits into a conversational interface. Unlike performance-based ads, OpenAI is charging advertisers on a pay-per-impression basis, which means advertisers pay when ChatGPT shows an ad, not when a user clicks or buys anything.

From OpenAI’s perspective, this guarantees them revenue even if users ignore ads completely, which is a safer bet than experimenting with a brand-new ad format.

Ads will appear at the bottom of the ChatGPT interface, not within the answers themselves. And OpenAI will clearly label them and keep them visually separate from conversations.

Source: OpenAI

Although OpenAI is designing ChatGPT ads to display below conversations (at least for now), even that separation introduces new questions regulators may not have expected to answer. In its announcement, the company stated that ‘Ads do not influence the answers ChatGPT gives you.’ And it said that ‘Answers are optimized based on what’s most helpful to you. Ads are always separate and clearly labeled.’

Ads won’t be shown to Plus, Pro, or Enterprise users, which basically indicates that OpenAI is using this as a way to subsidize free and low-cost access to ChatGPT.

Why OpenAI Is Turning To Ads Now

Ads have popped up in numerous places for years, streaming services, social media, and web browsing being just a few, but why has OpenAI trialed this shift now?

• OpenAI says advertising helps keep ChatGPT accessible for free and low-cost users.
• Subscription revenue isn’t enough to cover the cost of running and scaling large AI models.
• The company reportedly lost around $8B to operational costs in the first half of 2025, with only 5% of users paying.
• The massive spending commitments on data centers and chips are driving up costs for the company.

OpenAI has framed advertising as a way to keep ChatGPT widely accessible. In its announcement, the company said the move would allow ‘more people to benefit from our tools with fewer usage limits or without having to pay.’

Okay, so that explanation may make a bit of sense, but it’s also incomplete.

According to reporting by the Financial Times, OpenAI lost around $8B in the first six months of 2025 in operating expenses. Even though they have roughly 800M users, only about 5%, around 40M, are paid subscribers. At the same time, OpenAI now has about $14T in spending commitments tied to data centers, chips, and other infrastructure to scale its models.

It’s fair to say that running one of the world’s most widely used AI systems is expensive, and subscription revenue alone clearly isn’t covering the bill. So, this means advertising looks less like a philosophical compromise and more like a financial necessity.

Speaking shortly after the announcement at the World Economic Forum (WEF) in Davos, OpenAI CFO Sarah Friar defended this move by framing it as an access issue rather than a profit grab. ‘Our mission is artificial general intelligence for the benefit of humanity,’ she said, ‘not for the benefit of humanity who can pay.’

In this framing, it seems the company is using advertising to fund its growth without locking advanced AI behind a paywall, even if it complicates trust, privacy, and governance in the process.

Advertising Inside Conversations Changes The Data Equation

Let’s face it, advertising on websites and social media platforms is nothing new. Advertising inside a conversational AI system is. The worrying thing about this development is that ChatGPT is often responding to prompts that contain personal, emotional, or sensitive information. 

Remember that trend when everyone was showing off how they were using ChatGPT as a therapist? That’s the concern here: a data governance issue.

Source: Reddit

OpenAI says ad targeting will be contextual, which means the topic of conversation will trigger ads rather than personal data. For example, someone researching a holiday destination might see travel-related ads, while a user asking about productivity might see a sponsored service.

Contextual targeting may be privacy-friendly, but conversational context is far richer than a search query or article headline. This is because prompts inside ChatGPT can reveal intent, uncertainty, and vulnerability in ways that traditional advertising platforms don’t see.

OpenAI has firmly stated that it will not share user conversations with advertisers. But even if they don’t do this, questions remain about what data OpenAI processes internally to decide which ads show up, how long that data is retained, and who has access to it.

Source: OpenAI

Conversational AI doesn’t know what users are interested in, like traditional digital advertising, but it often knows why. This makes advertising decisions far more sensitive and potentially powerful. So far, OpenAI hasn’t fully explained what information it will use to determine relevance, which is a big gap in transparency.

Safeguards, Controls, And OpenAI’s Promises

There are a couple of safeguards OpenAI is putting into place. But are they enough? OpenAI won’t show ads to accounts where the user is under 18, and it will exclude ads from appearing near sensitive topics like health and mental health.

ChatGPT determines if a user is under 18 either because the user has said so or because the system predicts it.

Source: OpenAI on X

Users will also be able to see why OpenAI is showing them an ad, dismiss it, and provide feedback. These features may seem a little like what we already see on social media platforms. But the difference is emotional context.

Seeing an ad next to a Facebook post is one thing. Seeing one after asking AI for advice about your career, finances, or well-being is quite another. OpenAI insists that ChatGPT’s responses will always be ‘driven by what’s objectively useful, never by advertising.’

Whether users continue to believe that as ads become more familiar is an open question.

Why Existing Regulations Don’t Quite Fit

Let’s dig into the regulatory perspective on all of this. Right now, ChatGPT ads are in a grey zone. Lawmakers originally wrote most advertising and digital governance frameworks with feeds, timelines, and websites in mind, not conversational systems that respond in natural language and build ongoing context with users.

• The EU’s Digital Services Act (DSA) focuses a lot on advertising transparency and disclosure. While ChatGPT’s ads are labeled, the DSA doesn’t fully account for how conversational interfaces shape trust and perception.

Source: European Commission

• The Digital Markets Act (DMA) is all about gatekeeping power and competition. If conversational AI becomes a primary interface for accessing information, questions around preferential treatment and commercial influence will become harder to ignore.

• The EU AI Act is another regulation lawmakers may need to re-examine. The law requires platforms to clearly identify AI-generated content and advertising to prevent manipulation.

OpenAI somewhat aligns with these requirements, but it’s likely regulators will still need to consider whether disclosure alone is enough when ads appear inside individual and highly personalized dialogue dialogue rather than alongside content made for wide audiences.

The Risk of What Comes Next

There’s an even bigger concern to consider here: how these ads may evolve.

Right now, OpenAI is saying there’ll be strict separation, contextual targeting only, and user control. But once their advertising turns into a serious income stream, the platform may shift to looser protocols.

The internal use of data to optimize the relevance of ads and measure their effectiveness could expand, even if OpenAI never sells the data externally.

A Regulatory Reckoning For Conversational AI

Advertising inside ChatGPT may help fund the next phase of AI development, but it’s also going to force regulators to confront what could be an entirely new reality. Governments may need to rethink how transparency, consent, and accountability will work when ads appear inside conversations rather than beside content.

That could mean new policies specifically drawn up for conversational systems, or significant changes to existing frameworks. What’s clear right now is that advertising has pushed generative AI into regulatory territory that no one has fully mapped yet.

Whether OpenAI’s promises hold and whether regulators move quickly enough to keep up will shape how commercialized and trusted conversational AI becomes.

What Happens Next for ChatGPT Ads And AI Regulation?

For now, OpenAI’s advertising experiment is relatively small, and they’re certainly wrapping it in assurances. Ads are limited, clearly labeled, and framed as a way to keep ChatGPT accessible rather than to squeeze users for data.

But that careful positioning may not last forever, especially if ads become essential to funding AI infrastructure at scale. Regulators are likely to watch this rollout closely, as conversational AI advertising sits a little awkwardly between existing frameworks, and that will likely trigger policy reviews, clarifications, or entirely new rules.

Why? Because lawmakers didn’t build current advertising rules for systems that talk. Regulators now need to decide how advertising law applies when the interface itself feels human.

If ads inside AI conversations become the norm, lawmakers will need to move fast. Otherwise, the rules governing AI monetization may be one step behind the tech that’s about to reshape how people ask, learn, and decide.

The post ChatGPT’s Move Into Advertising Raises New Concerns For AI Regulation And Data Governance appeared first on Techreport.

The typical B2B sales process has always involved human persuasion. The sales team brought prospects up to speed on a product, ran demos, answered questions, and persuaded the buyer to make a purchase.

But according to Charles Gaudet, that era is rapidly coming to an end.

Gaudet, CEO of Predictable Profits, predicts that the job of the B2B sales representative as we know it will no longer exist by 2030.

These predictions are based on his experience working with growth-oriented businesses to successfully navigate markets influenced by AI.

“By 2030, all B2B sales decisions—regardless of value—will be made through a rep-free experience,” Gaudet says. “AI is not just changing how companies sell. It’s fundamentally reshaping how buyers decide.”

The Buyer’s Journey Has Already Changed

Many executives maintain that AI technology will create operational disruptions later on, or that it mostly affects internal business functions. Gaudet regards that mindset as outdated, which carries some dangerous implications.

“Most CEOs think AI hasn’t hit their business yet,” he says. “But it already has.”

AI-generated search overview traffic through organic search channels saw a 60 percent decline between January and November of 2025, Gaudet reports. 

Buyers are no longer starting their journey with keyword searches. Instead, they ask AI outcome-oriented queries: “How do I grow?” “What’s the fastest way for me to grow?” “Who can help me solve this problem?”

In this new environment, AI will be the main source of information for buyers, providing curated and synthesized information. It will assess the credibility of the content and provide the consumer with a list of vendors to evaluate.

“If AI can’t cite your expertise, you’re invisible,” Gaudet says. “You never even make the shortlist.”

At this point, AI has become a decision-maker for buyers. It will be able to analyze ROI, method, risk, and reviews about each vendor in seconds. 

Buyers will have reduced their choices down to a small number of finalists before the vendors are aware that they were even in competition for the buyer’s business in the first place.

By the time a prospect reaches out, Gaudet says, the decision is largely made. “They’re not looking to be sold. They’re looking to validate.”

The End of Persuasion

The shift at hand does not remove the role of humans in the buying process; it will completely change what it means to be involved.

Professional sales methods, such as pitching, objection handling, or follow-ups, will lose relevance as AI has already handled this by providing educational content.

“AI kills the pitch,” Gaudet says. “But it elevates the advisor.”

In a post-sales-rep world, humans will engage in the buying cycle later on and for different reasons than before. Gaudet states that humans offer value through three major areas:

1. Strategic diagnosis: Although AI provides users with data, humans provide the context (or interpretation) of that data for individual buyers based on their unique constraints, team dynamics, and growth goals.
2. Emotional risk removal: Even with data-driven decisions, buyers still have human-specific questions like, Will it work? If it fails, do I take a loss? Is my Company ready? AI can provide information, but it doesn’t help eliminate emotional risk.
3. Value orchestration: Humans assist in co-designing engagements, providing alignment between stakeholders, and guiding companies through legal, security, and post-sale success. The role in this case shifts from being persuasive to being collaborative.

“The order-taker disappears,” Gaudet says. “The trusted advisor becomes essential.”

What Happens to Companies That Don’t Adapt

The hazards of clinging to the traditional sales-led growth strategy are increasing at an alarming rate. According to Gaudet, one of these increasing risks is something that’s often overlooked: exclusion.

“The biggest danger isn’t losing deals,” he says. “It’s not being considered at all.”

If an AI-based system can’t identify, understand, and reference a company’s expertise, that company will automatically be eliminated from the buying journey. There will be no rejection emails, no deal-loss analysis, just a state of being invisible.

This lack of visibility leads to a chain reaction of negative impacts. Companies leveraging AI close deals within a timeframe of weeks, while traditional firms will require months to complete the same process.

As human labor replaces automated education and qualification, costs rise. Founders face a difficult situation where they must maintain their role as the main driver of business expansion through referrals and personal networks, leading to burnout and stalled scale.

Gaudet explains this situation as an irreversible downward cycle, one that Predictable Profits helps clients prevent through strategic changes. He estimates that adaptation needs to take place between 18 months and 24 months, before the opportunity to transition from founder-led growth to scalable, system-driven growth fades.

The New Roles AI Creates

Despite the disruption, Gaudet remains optimistic. AI exists to help humans become more efficient, not take their place.

High-growth organizations are now experiencing new roles. AI implementation architects design hybrid buyer journeys. Content consumption strategists develop educational content that teaches humans while it trains AI models. Demand orchestration specialists study behavioral signals across non-linear paths.

Other roles would cover the aspects and operations that AI can’t undertake, such as transformation experience designers, who would design workshops and offsites, and community ecosystem builders, whose role would involve building communities where members can establish trust and accountability as well as share core values.

“This is the next era of growth,” Gaudet says. “AI decides the options. Humans deliver the transformation.”

Leaders who aren’t sure where to start their journey should first learn about the buyer’s journey of their most valuable customers, offers Gaudet. 

Discoverability is a requirement for survival within an AI economy. It’s far beyond just a marketing advantage.

The post By 2030, AI Will Replace All B2B Sales Reps appeared first on Techreport.

For decades, the US dollar has been more than a currency. It has functioned as infrastructure.

Access to the dollar system, through global banking networks, correspondent accounts, and the SWIFT messaging system, has long given Washington leverage over trade, capital flows, and sanctions enforcement. Recent geopolitical tensions have only made that reality more visible. 

Now, a quieter shift is underway. As US policymakers continue to distance themselves from the idea of a central bank digital currency, interest is growing around a different tool: USD-backed stablecoins.

High ranking US officials, such as Scott Bessent, have framed regulated stablecoins as a way to strengthen global demand for dollars, particularly in regions where traditional banking access is limited.

This raises a broader question. If stablecoins increasingly function as digital dollars in practice, extending dollar reach while operating on public blockchains, are they becoming a new layer of US monetary infrastructure?

And if so, what does that mean for financial power, privacy, and the future shape of the dollar system?

Dollar Power Long Before Stablecoins

The United States has long exercised influence through the dollar’s central role in global finance. As the world’s dominant reserve currency, the dollar underpins international trade, cross-border lending, and commodity pricing.

International Monetary Fund data shows the dollar still accounted for 58% of global foreign exchange reserves in 2024, giving the US structural leverage over the international monetary system.

Source: IMF COFER

That leverage is reinforced by financial infrastructure. The SWIFT messaging network, while formally neutral and headquartered in Belgium, is deeply embedded in US and allied banking systems. Access to SWIFT and correspondent banking often determines whether institutions can transact internationally at scale. 

This became particularly visible after Russia’s invasion of Ukraine, when several Russian banks were removed from SWIFT as part of coordinated sanctions by the US and its partners, sharply restricting their ability to move funds across borders.

US sanctions policy has consistently relied on this architecture. The Treasury Department’s Office of Foreign Assets Control uses the dollar’s centrality to enforce restrictions by limiting access to dollar clearing and US-linked financial institutions. 

While global dependence on a narrow set of currencies and payment rails concentrates efficiency, it also centralizes power. Viewed through this lens, stablecoins look less like a departure and more like an extension of existing dollar-based influence. But now operating on new rails with even more technical embeddedness rather than new principles.

Why Washington Is Backing Stablecoins Instead of a CBDC

US policymakers have repeatedly signaled reluctance to issue a retail central bank digital currency.

Federal Reserve officials, including Chair Jerome Powell, have long stated that any US CBDC would require explicit authorization from Congress (a prospect widely viewed as unlikely), while also raising concerns around privacy, financial stability, and the role of the banking system. At the same time, Washington’s stance on USD-backed stablecoins has become noticeably more permissive.

Recent academic research for the Federal Reserve has increasingly framed stablecoins as extensions of the existing banking and payment systems rather than a parallel form of sovereign money. Provided they are fully backed, regulated, and integrated into existing compliance frameworks.

Source: @SecScottBessent on X

This framing has been reinforced by market-oriented voices such as Scott Bessent, who has argued that stablecoins can reinforce dollar supremacy by strengthening global demand for dollars. All without expanding the Federal Reserve’s balance sheet or placing the state directly between users and money.

His perspective aligns with a broader policy preference: preserving dollar dominance through private infrastructure rather than public issuance.

In practice, this approach allows the US to encourage digital dollar usage while avoiding the political and legal challenges associated with a formal CBDC. Brent Johnson, speaking on the Macro Voices podcast called the genius of this move Machiavellian, claiming that stablecoins come under the guise of freedom and efficiency, but ‘quietly reinforces the USD as the global reserve currency and the US govt control over the global monetary system.’

Synthetic Dollars and the Changing Shape of Liquidity

USD-backed stablecoins such as $USDT and $USDC increasingly function as offshore dollars. For users outside the United States, particularly in emerging markets, they offer direct exposure to the dollar without relying on local banks, correspondent relationships, or physical cash.

Data from Chainalysis shows how stablecoin usage is disproportionately high in regions facing currency volatility, capital controls, or weak payment infrastructure. Namely, Latin America, Sub-Saharan Africa, Eastern Asia, and Eastern Europe, where dollar liquidity has long been constrained.

Source: Chainalysis

This growth coincides with a broader shift in how dollar liquidity is managed domestically. In 2020, the Federal Reserve reduced reserve requirements for US banks to zero, ending the long-standing practice of requiring institutions to hold a fixed percentage of deposits as reserves.  

Since then, liquidity management has relied more heavily on discretionary tools, including interest on reserve balances, the discount window, standing repo facilities, and, during periods of stress, emergency backstops such as the Bank Term Funding Program.

Rather than hard constraints, the system now operates through centralized support mechanisms and supervisory oversight.

And the kicker is that stablecoins fit naturally within this evolving framework. While they sit outside the traditional banking perimeter, their issuers are tightly linked to US dollar markets through reserve assets, compliance obligations, and regulated on- and off-ramps.

From a system perspective, they represent another layer of dollar liquidity provision, one that is flexible, globally accessible, and increasingly shaped by regulatory design rather than balance-sheet rules.

Stablecoins do not replace the dollar system. They extend it, shifting where and how dollar liquidity circulates, while leaving ultimate control anchored in US financial institutions and policy frameworks.

A Digital Dollar Without the Name

In policy discussions, a clear distinction is often drawn between USD-backed stablecoins and a central bank digital currency. A CBDC would represent a direct liability of the Federal Reserve, issued and settled by the state.

Stablecoins, by contrast, are private instruments backed by reserves and issued by regulated entities operating within the existing financial system.

Functionally, however, the differences can appear narrower in practice. Both forms represent digital claims on the dollar, enable near-instant settlement, and can be integrated into programmable payment systems. Many claim stablecoins are the same as CBDCs, just with a few extra steps.

Source: @naval on X

Both also raise similar questions around compliance, transaction visibility, and the role of intermediaries. From a user perspective, the experience of holding and transferring a regulated stablecoin can resemble that of using a digital dollar, even if the legal structure differs.

The distinction matters politically. A US CBDC would require congressional approval and carries concerns around state involvement in retail payments and data access. Stablecoins avoid many of those obstacles by placing private issuers between users and the central bank. 

Technically, however, the outcome is similar: dollar-denominated value circulating digitally on modern payment rails. In that sense, stablecoins may deliver many of the functions of a digital dollar without the name (or the political cost) attached.

Control, Visibility, and the Trade-Offs

As stablecoins scale, questions of control and visibility move from theoretical to practical. Unlike cash, stablecoin transactions are recorded on public blockchains. They create permanent, auditable trails. 

While wallet addresses are pseudonymous, activity can often be linked to real-world identities through exchanges, payment providers, and other regulated on-ramps. This transparency is frequently cited by policymakers as a feature, not a flaw.

Control mechanisms are also built into the system. Major issuers such as $USDT and $USDC retain the ability to freeze or blacklist addresses in response to sanctions, court orders, or compliance requirements. 

In addition, stablecoin usage at scale depends heavily on KYC-compliant intermediaries. This creates chokepoints where regulatory oversight can be applied even if transactions occur on open networks.

It’s clear why the crypto community remains divided. Some view regulated stablecoins as a bridge between digital assets and the mainstream financial system. They celebrate the idea of improved liquidity, settlement speed, and institutional participation. 

Others see them as introducing financial surveillance and discretionary control into what were once permissionless networks.

Looking ahead, the trajectory is not predetermined. Stablecoins could evolve into a neutral payment infrastructure that broadens access to dollar liquidity. They could also become increasingly policy-shaped instruments, reflecting the priorities of regulators and issuers.

The outcome will be shaped not by the technology itself, but by the rules, incentives, and oversight that surround it.

The post Is the US About to Weaponise Stablecoins to Defend the Dollar? appeared first on Techreport.

Last week Australia become the first country to block social media access for everyone under the age of 16. An unprecedented mandate has forced TikTok, Instagram, YouTube, Facebook, and X to lock out an estimated 1 million teens, or face fines of up to A$49.5 million. 

The headlines are largely about screen time and mental health, but beneath this lies a more consequential shift. To make this ban enforceable, platforms must implement age-inference algorithms, selfie-based age estimation, and potentially even government-issued ID checks. 

What Australia has just passed as a child-safety measure is, in practice, the world’s first nation-level experiment in online identity infrastructure.

Other governments are watching closely. EU lawmakers have already hinted that Europe could follow Australia’s lead, while several other countries are carefully studying the model.

The real question isn’t whether teens will find workarounds — they will. Rather, the question is about whether this becomes a blueprint for the future of internet regulation.

The First Large-Scale Test of Mandatory Age Verification

Australia’s new legislation doesn’t just tell teens to log off of social media: it compels platforms themselves to prove who is under 16 and to block them. 

Ten major platforms, including TikTok, YouTube, Facebook, Instagram, Snapchat, and X (formerly Twitter), were ordered to enforce the block immediately. Snapchat saw around 440,000 accounts gone on day one, and TikTok deactivated around 200,000.

To comply with the new laws, platforms told Canberra they will rely on a mix of:

• Age inference based on users’ behavior patterns
• Selfie-based age estimation using facial analysis
• Optional ID uploads for cases requiring higher confidence

These measures aren’t experimental or fringe-based trials anymore; they’re part of the core login stack for millions of users. 

This is effectively the beginning of a mandatory online identity layer — even if governments aren’t currently using this kind of language to describe it. Participating in social platforms increasingly requires being scanned, inferred, or documented.

The Privacy Nightmare Behind Australia’s Social Media Ban

Mandatory age checks introduce an entirely new type of risk:

• Centralized biometric profiles of minors
• High-value “honeypots” of identity data (as we know, data can and is breached at a large-scale on occasion)
• Platforms gaining unprecedented levels of demographic data
• Governments gaining leverage over identity controls at the platform level

Whether it intended or not, the Australian government has launched the world’s first real-world stress test of a universal age-verification infrastructure. 

Silicon Valley’s Worst Fear: A Future Without Teen Users

Social media companies have been quick to point out that under-16s generate relatively little direct advertising revenue — although a study from late 2023 found the opposite, with platforms deriving between 16 and 41 percent of their revenue from users under 18. 

Regardless of the veracity of the statement, it misses the real concern. Teenagers aren’t just today’s customer-base: they’re the next decade’s. Platform loyalty tends to be formed early, and losing an entire age cohort breaks the habit-building pipeline that underpins long-term growth, and that these platforms tend to depend on.

That’s why Australia’s ban is so concerning for Silicon Valley. If one country can legally cut off teen access, others could conceivably follow suit. 

Social media platforms’ annual user growth has already been tapering off across major platforms, and a sudden drop in time-spent metrics could spook investors just as much as lost revenue.

X’s reluctant response from its official “Safety” account said, “It’s not our choice, it’s what the Australian law requires.” Clearly, they’re not the biggest fans of this legislation, and future revenue is sure to be a large factor.

There’s also a behavioral wildcard to consider. Many teens interviewed by Reuters openly said they would find ways to circumvent the ban. That likely means migration toward platforms like Discord, Telegram, VPN-enabled access, or decentralized services that lack the infrastructure and enforceability of mainstream apps. 

Ironically, the ban could push some teens into darker corners of the internet, amplifying one of the key risks regulators are trying to reduce.

The Global Domino Effect and the Split of the Internet

What’s unfolding in Australia is already being watched from far beyond its borders. European lawmakers have openly said that they want to “learn from” the ban, while governments in Denmark, New Zealand, and Malaysia are watching to see how it plays out in practice.

Paired with the UK’s Online Safety Act, and growing parental pressure in the US, the trend is clear: a more tightly age-gated era of the internet is ostensibly beginning.

The technical implications are significant, too. Once age verification becomes mandatory in one major market, platforms will have little choice but to enable the same infrastructure globally. 

This means wider use of biometric age checks, selfie-based estimation, and behavior inference, and a growing discrepancy between national versions of the same platforms. Meanwhile, teens are likely to migrate toward platforms that are harder to police, creating new “grey zones” of online interaction that sit outside of mainstream social platforms.

The result could be a fractured web: less anonymous, more regulated, and increasingly shaped by national legislation. In recent months, we’ve seen this trend unfold on a different level, too: the influence geopolitics has been exerting on how large AI models behave and what they’re allowed to say. 

Australia Just Gave Us a Glimpse of the Next Internet

Australia’s teen social media ban isn’t really about TikTok or Instagram: it’s about infrastructure. To make the ban work, platforms need to implement age-verification systems that won’t just disappear once the headlines fade. Other governments will study, reuse, and refine them.

As much as it may not align with their incentives, platforms will comply because the law demands it. Some teens may be deterred from trying, but many will surely try to get around it, or move to other forms of social media.

However, the verification layer — the quiet normalization of online identity checks — is likely here to stay. This is how the internet changes: incrementally, disparately, and then all at once. And we’re seeing the early phase of that cycle unfold before our eyes.

The post Australia’s Teen Social Media Ban: A Test for the Future of the Internet? appeared first on Techreport.

The Indian Department of Telecommunications (DoT) has issued a private directive to mobile phone manufacturers to preload the Sanchar Saathi app on all mobile handsets manufactured or imported for use in India. The directive requires the app to be readily visible and accessible to the end user at the time of first use or device setup. 

What’s really concerning is the government requesting these manufacturers that the app must not be disabled or restricted, meaning you cannot uninstall it. For devices that have already been sold and are in use, manufacturers are required to push the app through software updates.

The Sanchar Saathi app isn’t a new development in itself. It has existed on both the Apple App Store and the Android Play Store for quite some time now, with 15 million downloads so far.

Marketed as a citizen-centric safety tool, the app allows users to track lost or stolen mobile phones through its International Mobile Equipment Identity (IMEI) number. Users can use it to check how many mobile connections are registered in their name, which in turn helps locate and prevent the fraudulent use of mobile numbers.

How is Sanchar Saathi Intrusive

To understand why the Sanchar Saathi app is a privacy nightmare, let’s break down its functions. First, it can remotely block your stolen or lost handset through its IMEI number. IMEI is a unique code tied to every smartphone in India. Your IMEI number is permanent as long as you’re using the same smartphone. This data is then stored in the government’s CEIR database (Central Equipment Identity Register), tying a particular device to your identity.

Another app function allows you to check how many phone numbers are registered in your name. To buy a SIM card in India, you have to complete KYC, for which you require identity documentation – Aadhaar being the most common. 

This KYC data is stored by the telecom operator and shared with the DoT during various compliance processes, namely in the TAFCOP (Telecom Analytics for Fraud Management and Consumer Protection) database, which can be accessed by the Sanchar Saathi app.

Putting all the pieces together, the Sanchar Saathi app already has your Aadhaar information (your real-life identity), your SIM card details, and your IMEI number, creating a singular chain of identity that can be monitored by the government at any time. It’s essentially a state-owned app sitting inside your phone 24/7, which is why this mandatory directive isn’t sitting well with privacy advocates.

A careful look at the app’s required permissions raises even more questions on the government’s intent. It can access your camera, call logs, read phone status, send and view SMS messages, read your text messages, and even modify and delete content of your storage.

It can even view network connections and gets full network access. Congress, the opposition party in India, has come down hard on the government. “There’s a very fine line between reporting fraud and seeing what every citizen of India is doing on their phone. That’s not how it should work.” – Congress’s X handle.

Clarification and the Loopholes

After significant uproar, Jyotiraditya Scindia, the Indian Telecom minister, issued a clarification on the matter, saying that the app won’t be mandatory for the citizens. If anyone wishes not to use it, they can delete it or never even register.

The spokesperson said that the primary aim of the government directive was to make people aware of the existence of such an app, where they can track their stolen or lost devices. The DoT also posted on X to clarify that the app isn’t ‘watching you’ and that it won’t act as a surveillance tool and not track your data.

However, this clarification doesn’t sit in line with clause 7(b) of the written directive, which reads “….that its [the app’s] functionalities are not disabled or restricted.” 

Does this mean the Sanchar Saathi app will have all the required permissions on startup without you having to lift a finger? If so, then as soon as you buy a new device and insert your SIM card, the government can tie your identity to it. And if that’s the case, you’ll have to manually disable the permission from your settings. 

Widespread implementation of Sanchar Saathi means citizens don’t get the to choose. And even if they disable or delete the pre-installed app (after inserting the SIM card), does this erase their IMEI information from the app’s database?

There are a lot of unanswered questions and enough evidence to raise more than a few eyebrows regarding the intent of such a draconian and anti-democratic move, which invades people’s right to privacy and choice.

Apple to Fight for User Privacy

Apple is expected not to comply with the Indian government’s directive, and will likely try to find a middle ground. The company has a strong history of resisting such government orders in the past. 

• In the popular Apple vs. FBI case in 2016, Apple refused to unlock the mobile phone used by a terrorist since it would have required it to build a backdoor that could compromise the security of every iPhone in the world. 
• When Russia asked Apple to ban Telegram in 2018, Apple only disallowed updates for a couple of weeks before making a u-turn on its policy.

Likewise, pre-installing Sanchar Saathi on iPhones will require Apple to fiddle (technical term) with its iOS firmware, which it doesn’t often do. It would also mean altering privileged system-level capabilities, baking the app into the iOS firmware, and bypassing App Store security checks, which is indeed a lot of legwork for a privacy-focused giant like Apple.

It would also require Apple to alter its core iOS globally, just to appease the draconian directive of one government, something the company is highly unlikely to entertain. Apple is expected to communicate its concern to the Indian authorities and help find a middle ground, which may involve making new iPhone users aware of the existence of the Sanchar Saathi App.

The Death of Indian Privacy

This isn’t the first time the largest democracy in the world tries to redefine personal privacy.

The Telecommunications Bill 2023, passed on 21 December 2023, contains several provisions that threaten Indian citizens’ right to privacy and freedom of speech. This legislation allows the government to intercept, monitor, and block messages between two or more persons if it is deemed necessary in the interest of public safety or during a public emergency. Specific grounds might include national security, prevention of offences, or maintaining public order. 

The vague and broad terminology used in the Act raises serious concerns. For example, the Act defines “telecommunication” to cover the transmission, emission, or reception of messages by wire, radio, optical, or other electromagnetic systems – essentially every form of written or voice communication that exists today. 

This means the 2023 Act provides the government with sweeping powers to intercept and block private communications on grounds that are far from specific. Terms like ‘public safety‘ and ‘state security’ can be interpreted in countless ways, and individual incidents can be molded to fit these descriptions without sufficient justification.

This could be used to censor communication and content that doesn’t sit well with the government, especially speech that is critical of the current regime, since anything ‘against the government’ can be perceived as a threat to the ‘security of the state.’ And it’s not as though the Indian government hasn’t done this before. 

On July 3, 2025, X claimed that the Indian government had ordered it to block 2,355 accounts including Reuters’. While the exact reason for the blocking wasn’t made public, the order was likely issued under Section 69A of the Information Technology Act, 2000, which allows the government to block content for national security, sovereignty or public order. 

X said that this was part of the ongoing press censorship in India. The country now ranks 151/180 in the freedom of press index, falling from its previous ranking of 142 in 2022. In 2013, it stood at 140.

Section 19 of the Telecommunications Act allows the government to set standards relating to data processing, encryption, network security, and technical and operational parameters for telecom networks and services. This means the government can simply issue a new technical standard instead of passing a dedicated surveillance law in Parliament, and all of this can happen without judicial oversight at a moment’s notice. 

The same section also mandates storing decrypted copies of messages and requires traceability, creating de facto censorship and surveillance on these applications. This is exactly what appears to be happening with the Sanchar Saathi app directive, as it functions more as a legislative order forcing smartphone manufacturers to comply rather than going through a legislative route.

The Way Ahead

I’m reminded of this famous quote from Game of Thrones: “the night is dark and full of terrors, which” seems to be coming true for privacy and freedom of speech in India. While the government’s clarification have somewhat eased some nerves among privacy advocates, the need for such a directive still remains a mystery. 

If tracking IMEI numbers and retrieving stolen phones were the true objective, the government could still achieve this without the Sanchar Saathi app in the first place. It can request actual cell tower and customer data from telecom operators, who are already required to log IMEI numbers and the SIM cards used with those IMEIs.

Or it can use the CEIR system, which already enables IMEI blacklisting and IMEI–SIM mapping. That objective alone does not warrant installing an app with extensive, intrusive permissions on over 700 million Indian smartphones, and so the only logical conclusion here is that

The post India’s New ‘Sanchar Saathi’ Surveillance Play: The Slow Death of Indian Privacy? appeared first on Techreport.

There was a time when Facebook felt warm and personal. You opened the app, and you would see real updates from friends and family.

However, Facebook feeds have changed over the years. Now, it shows irrelevant posts from pages you liked years ago, ads about products you don’t care about, and updates from people whom the platform thinks you should care about. 

According to writer and journalist Cory Doctorow, it’s not just Facebook; many of our beloved tech platforms, like Amazon and Uber, have gone bad on the path to higher profits. He coined the term ‘enshittification‘ for this intentional decay of tech platforms. 

With AI penetrating every facet of our lives, it’s natural to ask: will AI go down the same path? Even worse, are we ready for that? But let’s first answer a more important question: how exactly does the enshittification happen?

How Platforms Rot on Purpose

Cory Efraim Doctorow said that enshittification happens in three stages:

1. Attracting Users

In the first phase, tech platforms attract an audience through good user experience and valuable features. Then, they lock users in with a consumer-first approach, making it harder for them to leave the ecosystem.

Think about the good old days of Facebook, when the platform didn’t try to shove ads down your throat every second and mostly showed you updates from your closest friends. 

That’s what got users fully invested. How? You uploaded photos, created groups, and ran Facebook pages. All this made leaving difficult because a big part of your social life was on Facebook.

2. Abusing Users for Business Gains

When platforms have managed to attract a large number of users, stakeholder and business interests tend to take on a more pivotal role. Pressure to make profits go up becomes a norm, shareholders start demanding better balance sheets, and tech companies must answer.

That’s when the enshittification rears its ugly head. At this stage, the platforms focuses more on the interests of business customers.  

In Facebook’s case, the company started tracking your activity to push targeted ads. It also throttled the reach of posts, nudging you to pay for boosts. 

On top of that, viral pages and constant suggestions flooded your feed, hiding the real updates of people you cared about.

3. Abusing Everyone for Stakeholder Profits

In the final stage, tech platforms extract as much value as possible for their owners and shareholders at the expense of both end users and business customers. 

Facebook followed the same path.

Users now see many sponsored posts, marketplace ads, and unrelated pages. Advertisers, as Cory noted, faced rising costs, lower returns, and growing ad fraud. It’s no surprise that one of the world’s biggest advertisers, P&G, cut digital spending once due to wasted digital ad spend.

And remember when Yahoo turned down a offer to buy Google for $1M because they thought it would mean fewer ads? Same story. A

ccording to Cory Efraim Doctorow, enshittification has already set in leading tech companies, such as Amazon, Uber, and Microsoft.

So, it’s fair to ask if AI is heading the same way. 

Why There’s a Real Chance of AI Enshittification

Spoiler alert: yes, AI can slide into enshittification because incentives push it into that direction.

AI companies require substantial funding to build and run data centers continually. Electricity consumption is a big problem, as Deloitte estimates that data centers could easily consumer over 1,000 terrawat-hours by 2030.

And computer hardware cost is already skyrocketing, with OpenAI’s Sam Altman launching a $1 trillion IPO last month.

So, only a handful of AI companies will survive and thrive in the market. Fewer players mean less competition, giving companies room to lock features behind paywalls or to implement ads. 

Besides, most leading AI companies are still operating on a loss. OpenAI, the maker of the world’s most popular AI chatbot, ChatGPT, is still losing money. Another AI player, Anthropic, is also not profitable. 

This bubbling pressure may force AI companies to search for alternate ways to make more money. And that’s the classic start of a trip down enshittification lane, where the end user’s interests get sidetracked in the pursuit for profits.

Furthermore, there are indications of an AI bubble. So, if the bubble pops, AI companies are likely to rush to squeeze more value from users, which can speed up enshittification.  

As LLMs are like black boxes, we don’t know what goes inside and how you see the results in your chatbot. So, Cory believes that it’s easy for AI companies to hide their enshittification techniques. 

They have an ability to disguise their enshittifying in a way that would allow them to get away with an awful lot…I think they’ll try every sweaty gambit you can imagine as the economics circle the drain.

– Cory Doctorow, Wired

What Happens When AI Stops Working the Way You Expect

AI is not just a productivity tool; it’s now part of our daily lives. People use chatbots to write emails, summarize documents, and make purchase decisions. Some even use to plan their wedding, make important life decisions and… cheat on science exams.

According to OpenAI’s own data, ChatGPT has over 700 million weekly users. Of these, 49% use it for ‘Asking’ activities, including moments when people want clarity on a complicated topic or advice. 

Now imagine if OpenAI monetizes and enshittifies ChatGPT for an audience who’s using it for anything and everything.

Answers may feel weaker, slower, or biased; daily tasks like writing or summarizing may take longer, and buying advice will skew toward paid recommendations.  

And the impact doesn’t stop with everyday users. Companies that depend on AI feel the strain too, and even harder. 

According to a McKinsey report, 88% of companies regularly use AI for at least one business function. For these companies, enshittification of AI could kill productivity.

It can slow workflows and weaken output quality, and force teams to spend more time on tasks they once handled in minutes.

The end result = profits may take a hit. And profits are the bottom line for most businesses.

Real-World Signs of AI Decay

AI companies are actively exploring ad space. Users can now buy products right inside ChatGPT, and Walmart already works with OpenAI to show product listings right in the chat.

Perplexity is also experimenting with ads, though it paused new advertising deals.  

Could these be signs of AI enshittification? Well, it’s debatable. But one thing is for sure: when a platform blends answers with ads, the line between help and revenue gets blurry. 

The risk grows because users trust AI to act like a neutral guide. But once ads sit inside the same space as organic answers, you can’t easily tell which part serves you and which part serves the platform. 

And when the system doesn’t disclose the parameters used to ranks or selects results, you no longer know if a paid selection has replaced an organic one that was a better fit for your ask.

Furthermore, evidence of AI quality decline is already surfacing. According to Stanford and UC Berkeley research, ChatGPT’s response quality showed an alarming decline. 

Besides, AI platforms are now shrinking their free tiers with message caps and weaker fallback models. For example, a free ChatGPT user can send up to 10 messages every 5 hours with OpenAI’s latest model, GPT-5.1. 

These limited free tiers push users toward paid plans, indicating the early pattern of enshittification, where helpful services slowly trade user value for revenue pressure.

Where Does This Leave Us Now?

AI as a whole can offer real value, helping millions of people work faster and live more fulfilling lives. But the risk of enshittification is very much there, whether you like it or not, and it’s shaped by high costs and pressure to grow. 

We’ve seen this pattern play out before with tech companies like Facebook. So, the question isn’t whether AI can fall in the same abyss. Rather, it’s when and how that will happen.

It’s worth noting that companies like Google stand in a different position, with steady streams of income from other sources. So, they can wait for AI to make money before extracting value from users.  

However, OpenAI, the leading company in the AI space, doesn’t have that kind of luxury. It must bring more people in and keep adding value.

This creates a constant dance between offering real help and slipping toward enshittification. Besides, the pressure of cash burn can also lead to rushed or user-unfriendly decisions to squeeze users. 

That said, the path ahead isn’t fixed. The choices these AI companies make now will shape how much we can trust AI tomorrow.

The post Are We Ready for AI Enshittification? What Happens When the Systems You Trust Suddenly Stop Working appeared first on Techreport.

OpenAI has been dealt a significant blow in its case against GEMA, as the court ruled against the AI giant.

GEMA is a collective management organization representing stakeholders of the music industry (including lyricists, composers, and producers) that sued OpenAI for using nine well-known German songs for training its AI models, especially its 4 and 4-o models.

First, GEMA disabled ChatGPT’s web search capability and then assigned it the role of a ‘lyrics expert.’ After this, prompts like “What are the lyrics to (this song)?”, “Can you tell me the chorus?” or “What was the first verse of (this song)?” were fed into ChatGPT. 

Surprisingly (or unsurprisingly), the model was able to produce exact verses of the songs requested, albeit with a bit of hallucination. For instance, it produced 25 consecutive words from the song 36 Grad and over 70 words of “Über den Wolken.” 

And hey, these aren’t any under-the-radar songs ChatGPT just came across in the dumps of the internet. They are, in fact, hyper-popular German songs.

Think of songs like “Take Me Home, Country Roads” by John Denver or “Hips Don’t Lie” by Shakira. Scraping ultra-mainstream songs like these isn’t something that could have happened by chance.

The Court’s Ruling

Of course, there can be no doubt that those songs were a part of ChatGPT’s training data. After establishing this, the court relied on two major components to arrive at its decision: memorization and reproduction, and the EU’s Text and Data Mining (TDM) exception.

Any AI model’s training process can be broken down into three distinct parts. 

• First, the extraction or scraping of data from various web shops is performed to form a central corpus. 
• Second, the analysis and training of the model based on this data. 
• Third, the actual use of these models to generate outputs based on the trained data. 

Memorization vs. Reproduction: The Legal Line That Matters

Memorization kicks in during the second phase of development, i.e., training. The court assumes that memorization will occur during the training of an AI model, which happens when the model retains some of the information or data it was actually trained on. 

There can be a debate on whether this is a valid assumption. We think it is. Any model will have to memorize some part of the training data to reproduce content similar to the data it was trained on.

The court ruled that everything was fine up to this point and within the scope of the EU’s TDM exception. Articles 3 and 4 permit LLMs to train their models on copyrighted data, and an opt-out option is provided to the rights holder.

TDM enables the temporary storage of data (similar to RAM), format conversions, and analysis.

However, the problem arises when the model begins reproducing the data in its outputs, indicating that the LLM model has permanently stored the used data, something not covered by the TDM exception and thus in violation of the EU’s copyright laws.

Another case that can help us understand this memorization vs reproduction argument is the Getty Images vs Stable Diffusion case. Here, Getty Images initially filed for direct infringement and secondary infringement against Stable Diffusion AI, accusing it of using its copyrighted images in its training data.

However, in this case, the judge ruled against Getty Images, arguing that the model did not ‘reproduce’ any of the copyrighted material in its output.

Even here, the fact that an AI model can memorize some of the training data was well established. This brings us to a very key demarcation: merely memorizing training data does not constitute copyright infringement.

If you memorize a verse of Shakespeare’s poetry and learn it line by line, that does not mean you have infringed his copyright. However, when you start writing the same verses, passing them off as your own, and publishing them under your name, that’s when copyright infringement kicks in.

That’s exactly the distinction courts have alluded to. Infringement occurs when the AI model begins reproducing the data it previously memorized during training. This is the basic difference between these two similar cases. In the GEMA case, ChatGPT reproduced exact verses of the songs it was trained on. However, Stable Diffusion did not reproduce the images owned by Getty Images.

This case has set a very important precedent for copyright claims in the EU. We’ll probably see courts citing these two cases while establishing the validity of copyright infringement cases in the future.

The Ethical Grey Zone, or the Loophole: When AI Mimics Without Reproducing

While it’s not expressly stated, the court orders imply that using someone’s copyrighted material for training isn’t infringement as long as there’s no reproduction. This means that if AI companies ensure no reproduction takes place, they can use any and all copyrighted material for training their AI models. This still seems unethical, doesn’t it?

For example, you can build an LLM model trained exclusively on the entire life’s work of Shakespeare while ensuring there’s no reproduction at all. And whenever a user inputs a prompt asking it to write in ‘Shakespeare’s style and tone,’ the model could generate Shakespeare-like writing without actually reproducing any of his original content. Technically, this is not copyright infringement, but it’s still hijacking the very individuality of Shakespeare.

Similarly, consider if you trained an LLM model exclusively on Ed Sheeran’s songs and then fed it lyrics and music of your own. The model would be able to generate a song that seems like Ed Sheeran himself has made it. This is still a massive grey area in the entire copyright infringement fiasco.

In these cases, while there’s no exact reproduction, the reproduction is such that it infringes on the identity of the artist himself. This way, it would become easier for AI companies to mimic or be inspired by copyright holders while steering clear of infringement claims.

EU vs US AI Laws

While the EU’s AI Act and TDM exceptions focus on protecting the rights of the copyright owners, the US lawmakers seem to be in a confusing state when it comes to slamming AI companies for eating into the years of hard work of these artists. 

Similar cases in the US have turned into long-drawn-out battles in court with no end in sight. Here are a few cases that highlight the ingenuity of the US lawmakers.

Anthropic vs Universal Music Group (UMG)

A case similar to the GEMA vs OpenAI is UMG vs Anthropic. UMG, along with ABKCO Music and other publishers, filed a lawsuit against Anthropic, accusing it of using copyrighted songs (as many as 500) for training its Claude AI LLM. As expected, Anthropic claimed ‘fair use’ and denied any copyright infringement. 

What’s interesting is that UMG had requested a temporary injunction on the use of the copyrighted material to stop Antropic from using the copyrighted content while the case was ongoing. While the court decided that there was enough evidence for the case to go to trial, it rejected the injunction plea due to the lack of sufficient demonstration of any ‘irreparable harm’ done to the plaintiff.

This is pretty confusing. It’s like the court saying, ‘Yes, there may have been a crime committed, but we won’t stop the defendant from continuing to do so until proven.’ The case was filed in October 2023, and a decision on the matter has not yet been made. Anthropic has been able to use copyrighted songs for training its AI for two years, and it will be legally allowed to do so until the conclusion of the case. 

Even if the court finds Anthropic guilty, it would, in all likelihood, impose a monetary penalty and/or instruct it to remove the songs from the LLM’s training data. Now, AI companies view these penalties as the ‘cost of doing business,’ or the cost of training their AI models. 

Let’s circle back to the memorization vs. reproduction debate cited by the German court. How would an AI model ‘un-learn’ or ‘forget’ data that it has already been trained on? Would this require some sort of reverse engineering?

This is exactly the kind of confusion in the American laws, a loophole that the AI companies are milking while legislators try to wrap their heads around this newfound problem.

New York Times (NYT) vs OpenAI and Microsoft

Something similar happened in the NYT vs OpenAI and Microsoft case, where the plaintiffs accused the AI companies of using copyrighted news articles for training their LLM models. The defendants claimed that they only used ‘publicly available’ articles that do not infringe any copyright claims. 

The case has been in pre-trial since December 2023. Sadly, the court hasn’t permitted any injunction in this case, allowing the companies to continue using the alleged copyrighted material for training.

We can see a clear pattern in how US courts and laws view the entire copyright matter. It seems as if the courts are busy creating an ‘illusion’ for plaintiffs that their rights are important and they have, in fact, been wronged by the AI companies. However, at the same time, the cases are progressing very slowly, with no injunctions, allowing the AI giants to continue their operations.

The Scarlett Johansson Case

This is also not the first time OpenAI has disregarded its moral compass in the development of its AI chatbot. In 2024, ChatGPT rolled out its new feature, Voice, which contained one particular version (named ‘Sky’) that sounded eerily similar to Scarlett Johansson. 

According to reports, Sam Altman had previously requested that Scarlett lend her voice for ChatGPT, which she politely declined. However, Altman still went on to mimic her voice through an alleged voice actor in a later update.

While OpenAI denied any wrongdoing, public feedback was overwhelmingly negative, with several users claiming it was actually the actress’s voice. Not only is this a violation of personal rights, but an AI behemoth like OpenAI doing so sets the wrong precedent for the industry. Ultimately, though, OpenAI had to pause Sky.

This also highlights the stark difference between EU laws and laws in the US. First of all, there’s no equivalent of the EU AI Act in the US, which provides considerable leeway for AI companies to operate in the grey area. 

Even in this case, no legal proceedings have taken place, since the laws vary by state and there’s no central legislation governing the wrongful use of data or personal attributes by AI companies.

The Way Ahead

All in all, we’re not refuting the fact that artificial intelligence is the future of technology. However, it’s clear to everyone that building and training these AI models comes at a cost – both monetary and ethical. Now, it’s up to us whether we want to (or should) incur such costs.

The immediate ethical issue is the unfair use of copyrighted material, regardless of what the laws in various regions suggest. While the EU views it more progressively and in favor of those being wronged by such infringements, the US seems to side with the capitalistic AI companies. 

Amidst all this, it’s the actual copyright holders that are suffering. And while Hollywood stars, high-profile authors, and artists might get away with compensation or licensing agreements with AI companies, that’s not a luxury available to indigenous or independent copyright holders.

This is why it’s more important now than ever to push for strict copyright laws in the US.

The post OpenAI’s Defeat in the GEMA Case: A Landmark Moment for AI Copyright Law appeared first on Techreport.

When CrowdStrike recently tested DeepSeek-R1, China’s answer to Western AI coding assistants, researchers found something unsettling. 

The model occasionally produced insecure code, but that wasn’t all. Its failure rate spiked by nearly 50% when the prompts included politically sensitive references like Tibet or Falun Gong. These triggers had absolutely nothing to do with the task at hand.

The model wasn’t being jailbroken, tricked, or overloaded. It was performing as designed, and those design choices were bleeding directly into its technical output.

This isn’t just another AI bug or hallucination. It’s a glimpse into a deeper problem: AI systems now reflect the values, constraints, and geopolitical incentives of the cultures that create them. 

And although the manifestation of this reflection in DeepSeek stands out, this isn’t unique to it. We’re beginning to see similar patterns in Grok, Mistral’s Le Chat, and other nationalized models.

What CrowdStrike Actually Discovered

CrowdStrike’s Counter Adversary Operations team’s research didn’t start from the assumption that DeepSeek-R1 was flawed. In fact, baseline testing showed the opposite. 

It evaluated the model across 30,250 prompts and ten security categories. It found that it generated insecure code just 19% of the time: a rate largely in line with leading Western AI models.

The anomaly only showed up when researchers inserted politically sensitive terms into otherwise identical prompt structures. 

For example, when researchers asked for PayPal integration code, but specified that the system was “based in Tibet.” The result? The vulnerability rate jumped to 27.2%: nearly a 50% increase over baseline. 

In cases referencing Falun Gong, the model outright refused to output code in more than 45% of prompts, despite generating the reasoning and structure internally.

What makes this behavior more alarming isn’t the keywords themselves. They were irrelevant to the task of database design, fintech modules, and production recommendation engines.

A secure model shouldn’t change its output quality based on political modifiers that have no bearing on logic or architecture. 

The “Intrinsic Kill Switch”: Alignment Leaking Into Code Quality

CrowdStrike’s deeper concern wasn’t just that DeepSeek-R1 generated bad code when politics were involved. It was the way the model behaved in response to these triggers.

In many of these prompts, the model still produced a complete internal chain-of-thought solution:

• Database schema and table structures
• Authentication flow
• Error handling logic
• API integration steps

However, it refused to output the action implementation, claiming the task violated policy. This isn’t a standard safety filter; the model is clearly able to solve the prompt, but it simply withholds the output.

This suggests that the problem was more foundational: the political alignment was encoded into the model’s weights itself, rather than an external API wrapper blocking the answer.

And when the model did respond, the degradation wasn’t subtle. Researchers saw: 

• Hard-coded secrets and API keys
• Insecure storage of sensitive data
• Outdated or nonsensical authentication
• Broken syntax while asserting it followed ‘best practices.’

This is an entirely new category of failure. It’s not hallucination or censorship. It’s the model’s value alignment bleeding directly into its technical reasoning path. In other words, the ‘political’ and ‘engineering’ logic are no longer separable.

For cybersecurity researchers, this is the nightmare scenario: the safety layer becomes the vulnerability. 

Why This Likely Emerged (Regulatory Design)

DeepSeek’s behavior wasn’t random, nor was it the activation of a simple censorship rule. More likely, it emerged from the core architecture of how the model was trained, and the legal environment within which it was built.

China’s AI regulations require systems to adhere to its “core socialist values,” and explicitly, to avoid producing content that threatens national security. Nearly every major Chinese language model is trained with guardrails designed to skirt around politically sensitive topics.

This alignment pressure has consequences. Safety tuning doesn’t just filter output; it conditions the model’s internal association. In machine learning terms, models learn correlations rather than rules. 

Thus, if sensitive words frequently co-occur with “disallowed” output during training, the model begins to treat those triggers as a risk signal. And that risk gets expressed technically.

Instead of refusing to answer a political question, DeepSeek-R1 sometimes alters its approach to even non-political engineering tasks. The political alignment objective essentially overrode part of its coding objective.

This isn’t censorship in the traditional sense, as we generally understand it. It’s a side effect of training data and policy alignment leaking into the core reasoning.

The Bigger Pattern: AI Is Already Fragmenting

DeepSeek isn’t an anomaly. It’s one more data point in a trend we’ve been seeing all year. As models get larger and more autonomous, their behavior increasingly reflects the worldview, regulatory climate, and incentives of the companies and countries behind them.

We’re already seeing three distinct classes of “regional AI.”

China: Politically Constrained Factualism

DeepSeek already demonstrated this behavior outside coding tasks. 

In user-shared tests, the model avoided directly characterizing the 1989 Tiananmen Square protests and massacre, instead dodging the question by stating that it is an AI assistant “designed to provide helpful and harmless responses.”

It adheres to the informational boundaries established by Chinese law, rather than the technical accuracy boundaries.

United States: Commercialized Personality and Platform Alignment

X’s Grok model leans heavily into platform tone: hyper-casual language, crypto enthusiasm, and exaggerated personalization. When asked about Elon Musk, Grok has described him in mythic or over-elevated terms. 

Whether this is deliberate branding or emergent behavior isn’t particularly important. The end result is the same: model output shaped around cultural identity – in this case, of a company rather than a state.

Europe: Institutional Framing

Le Chat, Mistral’s French LLM, answers historical questions with a distinctly EU-academic framing. 

When asked about the Molotov-Ribbentrop Pact, the model described the consequences almost exclusively through the Soviet perspective, downplaying the long-term colonial impact the Allied powers had on Eastern Europe. Not wrong, but undoubtedly a culturally one-sided perspective. 

None of these examples is malicious; they’re signals. And the pattern is hard to ignore. 

For the first time in decades, we’re watching the early stages of a fractured digital knowledge layer. We may not get a single, unified “global AI” at all. 

Instead, we may get parallel AIs that frame history, politics, technology – and now code, too  – differently depending on where they were built.

The Security and Engineering Implications

Zooming out, it becomes clear that the CrowdStrike result isn’t just an academic edge case. It clashes directly with how modern software is built. In 2025, over 90% of developers rely on AI coding assistants for at least part of their workflows. These models aren’t just side tools anymore; they’re now part of CI/CD pipelines, enterprise stacks, banking APIs, and production infrastructure.

This creates a new risk category:

• What if two models implement security patterns differently by design?
• What if a vulnerability only triggers when the prompt contains certain linguistic or cultural conditions?
• What if “regulatory alignment” becomes indistinguishable from a security weakness?

CrowdStrike’s takeaway is simple: benchmarks won’t save you. Traditional audits often fail to identify failure modes caused by ideology, taxonomy, or keyword context.

As enterprises mix models across regions and supply chains, this creates a significant attack surface, including political triggers, cultural modifiers, alignment rules, and state requirements.

We’re entering an era where security isn’t just about the code. It’s about the values and worldview baked into the model that generated it.

The post DeepSeek-R1 Exposes a New AI Weakness: Security Degrades With Ideological Triggers appeared first on Techreport.

Jeff Bezos has taken the role of co-CEO in a new AI startup named Project Prometheus. What’s interesting is that this is Bezos’ first position in management since he stepped down as Amazon CEO in July 2021.

Project Prometheus describes itself as a company that focuses on ‘AI for the physical economy.’ Simply put, Prometheus will focus on the application of AI in industries that require advanced engineering and materials science.

This includes all manufacturing activities related to the production of physical goods, such as automobiles, electronics, heavy machinery, and aerospace manufacturing.

Up until now, the entire focus of artificial intelligence has been on ‘bits,’ which is why we have several Large Language Models (LLMs) today powering numerous chatbots and AI assistants.

However, Prometheus wants to focus on ‘AI for atoms,’ where artificial intelligence can be used to stimulate physics, automate factories, design physical systems, and reduce manufacturing times.

Prometheus debuts in an already overcrowded AI market where several small and big players are fighting to create a niche. However, one of Prometheus’ biggest advantages is the huge $6.2B funding it has secured, a large part of which comes from Bezos himself. This makes it one of the most well-funded early-stage startups globally.

In comparison, Periodic Labs, a company that aims to ‘automate scientific discovery’ by focusing on physical AI (much like Prometheus), raised $2B this year. So, of course, Prometheus’ chunky wallet gives it a head-start over its rivals.

China’s Lead in the Physical AI Race

Although the US has attempted to hinder China’s growth in the AI sector through strict export controls on advanced AI-compatible semiconductor chips, the fact remains that China is significantly ahead of the US in terms of physical AI applications.

Data from the International Federation of Robotics shows that China is third on the list of countries with the highest robot density. The country has 470 robots per 10,000 employees, which is up from 402 robots in 2022.

What’s even more commendable is that China entered the top 10 list only in 2019, and in just 6 years, it has raced past countries like Japan and Germany. On the other hand, the US ranks 10th on the list with a density of only 295 robots.

A high robot density points to a highly automated and productive manufacturing sector with relatively fewer defects and faster iteration cycles. Throw a bit of AI tech in there, and you have ‘smart factories’ with automated control systems, AI-backed quality inspections, predictive maintenance, and so on.

Along with this, China also seems to have a large talent pool to tap into:

• China produces 1,059 engineering bachelor’s graduates per million people (2025 estimate) as compared to the US’s 403.
• China graduates 1.5M+ bachelor’s engineers annually, along with 400K master’s and 60K PhDs. By contrast, the US awards just 140K bachelors with 50K master’s and 12K PhDs.

No wonder top US AI startups have to poach talent from other leading companies, something that Prometheus has also done. 

China has already surpassed the US in manufacturing, to the extent that it is now known as the world’s factory. It has left the US behind in areas like advanced battery manufacturing, solar production, EV supply chains, and robotics deployment.

• China accounts for 33% of the world’s electronic exports with $1,368B, whereas the US contributes just 4%. 
• The country produces around 77% of  the world’s EV batteries, according to reports as of 2022.
• China’s domination in the solar industry is even more daunting. It installed 329 GW of solar capacity in 2024, which was 55% of the total global capacity that year.

After being left behind in critical physical industries, the US can only play catch-up now, which is already a steep uphill climb. Deploying Physical AI might be the only way the US can get close to China. But that isn’t an easy task in itself. 

For instance, Marques Brownlee, a popular tech reviewer, discussed NEO, the most futuristic robot available in the US today. Although the company markets it as a smart robot, as things stand now, it’s anything but smart and autonomous. It works largely on human command and only does a handful of tasks without instructions. 

Brownlee noted that ‘AI companies’ are currently only selling the dream, before even fully creating the product. They’re using early-stage adopters as beta testers, but the gap between the blueprints and the final product is a huge one right now.

China is already deploying humanoid robots across frontline industrial applications. For instance, the Walker S2 from UBTECH, China’s most advanced humanoid robot, is capable of running 24/7 and can even replace its own battery. 

Compare this with NEO’s 4-hour battery life and teleoperation needs, and you’ll realize that the gap is wider than most Americans would want to believe. UBTECH plans to hit an annual capacity of 10,000 Walker S2 by 2027 for industrial use in smart factories, intelligent logistics, and automotive manufacturing.

Bezos’ Mission Critical Role

If there’s one man fit to lead such a herculean effort, it’s Jeff Bezos. His portfolio and experience are something even the brightest modern-day entrepreneur envies.

Bezos built the world’s largest logistics network, the most advanced warehousing system, a rocket company in Blue Origin, a cloud computing behemoth in AWS, and a robotics powerhouse in Amazon Robotics. That’s a solid CV.

Prometheus wants to build advanced physical stimulators, industrial-grade AI systems, and engineering automation tools, something that Bezos has dabbled in for 30 years.

Most importantly, he brings in the huge capital required by such a research-intensive operation. In addition to spending money out of his own pocket, the Bezos branding also helps attract top investors in the country, makes it easier to recruit top talent, and reassures investors about the viability of the mission.

Vik Bajaj, the other co-CEO of Prometheus, is also a very well-thought-out selection for the job. At a time when most AI leaders come with a background in machine learning and natural language processing, Bajaj comes from an experience in bioinformatics, biophysics, and molecular simulation.

He was the co-founder of Google’s Life Sciences division (which we now know as Verily), which is all about navigating huge R&D gestation periods and integrating robotics, sensors, AI, and cloud systems.

While Bezos is the face of the project, Vik Bajaj brings the hands-on experience and technological know-how required to tackle a business as large as this.

Unlike LLM-based AI startups, results for the physical AI space don’t come within just six months or a year. The R&D and trial-and-error phase could last for nearly half a decade, if not more.

A business this capital-intensive with a long gestation like this requires the support of someone who’s been there before – and both Bezos and Bajaj fit the bill perfectly.

The post Can Bezos’ Project Prometheus Close America’s Widening Physical-AI Gap with China? appeared first on Techreport.

If you’re a hobbyist developer or an Android power user, the latest development in the Android developer verification plan can put you at ease. 

According to the recent Google announcement, students and hobbyists will have a “dedicated account” that allows them to distribute their mobile apps on a limited number of devices without completing the full verification requirements. 

Additionally, the company is expected to introduce “advanced flow,” allowing experienced Android users to accept risks and install unverified apps.   

This shift makes more sense when you see what came before. In August 2025, Google launched a developer verification plan that would require all Android developers, regardless of whether they are outside Google Play, to verify their identities.

That meant users could install apps only from verified developers, which many saw as the beginning of the end of sideloading on Android. 

But why did Google introduce the developer verification in the first place, and why did it update its plan? We’ll break it down for you. 

Why Google Announced Developer Verification in the First Place

Android is the leader in the mobile operating system (OS) market, with a market share of over 72%. 

Source: StatCounter Global Stats – OS Market Share

And when a mobile OS becomes this big, its app store naturally becomes a prime target for malicious actors. According to the latest ZScaler Mobile, IoT, & OT Threat Report, malicious apps on Google Play were downloaded 42 million times this year. And there was a 67% year-over-year increase in Android malware incidents.  

Furthermore, AI has made it easier for hackers to craft sophisticated phishing campaigns that lure innocent users into installing malicious apps. 

Google’s own analysis found that over 50 times more malware is being distributed through sideloaded resources on the internet.  

It’s true that Google has control over Google Play and has implemented strict measures to filter out malicious apps from its app store.

However, it can’t regulate app users who sideload from outside resources. Therefore, it’s no surprise that Google has moved to tighten security across the Android ecosystem, introducing developer verification to prevent the installation of harmful apps.

The Android developer plan proves that the search engine giant takes security seriously. It’s worth noting that Google Drive for Desktop has recently added AI for ransomware protection.

However, the verification rules alarmed the community, which feared they were the beginning of closing an otherwise open-source system.

Original Plan: Strict Rules That Alarmed the Community

The company enforced verification requirements on Google Play in 2023. As a result, it has witnessed how developer verification helps prevent bad actors from spreading malware. It also reduces incidents of financial fraud and the theft of sensitive information. 

With that in mind, Google believes implementing the developer verification process can protect the Android ecosystem. 

Under the original developer verification proposal, starting in 2026, Android developers would need to complete verification for their apps to be installed on Google Play Protect-certified phones.

Google Play Protect certified phones are tested for security performance and preloaded with Google Apps. Leading smartphone manufacturers, such as Samsung, Motorola, Lenovo, and OnePlus, ship Google Play Protect-certified devices. You can check the complete list of partners for more details. 

The developer verification plan also covers developers who distribute apps outside Google Play, such as through F-Droid or other third-party channels.

To make the verification easier for developers who distribute apps outside its official app store, the search engine giant is building a new Android Developer Console (ADC). 

For verification, individual developers must submit a government-issued ID to confirm their name and address. Their email and phone number will be verified through a one-time password. Organizational accounts have different requirements.

However, both individual and organizational accounts have to pay a one-time $25 fee for a full distribution account in ADC.

Why Developers Were Skeptical About the Verification Process

There are several reasons why the Android community and developers were unhappy with Google’s proposed developer verification process. 

A widely accepted belief is that Play Protect, without the developer verification process, already scans all apps on your device, not only those you have downloaded from Google Play. So, what’s the need for this additional verification process for developers who distribute apps outside Google Play?

Another concern among Android users is that the developer verification process may discourage anonymous developers from creating apps, resulting in fewer new apps being developed for sideloading.

F-Droid, a leading free and open-source app store and software repository for Android, wrote in its official blog after the original developer verification announcement:

If you own a computer, you should have the right to run whatever programs you want on it. This is just as true with the apps on your Android/iPhone mobile device as it is with the applications on your Linux/Mac/Windows desktop or server. Forcing software creators into a centralized registration scheme in order to publish and distribute their works is as egregious as forcing writers and artists to register with a central authority in order to be able to distribute their creative works.

A Reddit user also expressed concern over the developer verification process.

So the fear is that the developer verification process might be the first step towards closing the open Android ecosystem and restricting sideloading. 

After considering these concerns, Google acknowledged the community’s feedback and made meaningful changes to the developer verification process.  

Google Heard the Community Voice 

It’s good to see that the search engine giant listened to the community’s voice and updated the developer verification process to help both experienced Android users and hobbyist developers. 

1/6 Keeping users safe on Android is our top priority. Today, we’re sharing an update on our new developer verification requirements – why they are critical to fighting modern scams, and how we’re adjusting our approach based on community feedback. 👇

— Sameer Samat (@ssamat) November 13, 2025

In Google’s own words, 

We know that security works best when it accounts for the diverse ways people use our tools. This is why we announced this change early: to gather input and ensure our solutions are balanced. We appreciate the community’s engagement and have heard the early feedback – specifically from students and hobbyists who need an accessible path to learn, and from power users who are more comfortable with security risks.

Under the new changes, Google says students and hobbyists will have access to a “dedicated account” type with fewer verification steps and no $25 registration fee. But this tier limits the number of devices to which their apps can be distributed. 

Additionally, Google will allow experienced users to sideload apps outside of Google Play through “advanced flow”, but it will display clear security warnings to ensure they fully understand the risks.

What This Means for Sideloading

Many Android users care deeply about sideloading because it gives them a real choice. It lets them install open-source apps, older app versions, region-blocked tools, and apps that may not appear on Google Play.

When Google added developer verification, many Android power users believed it could be the first step towards controlling or restricting sideloading. Many developers, especially those who remain anonymous, may stop creating new apps as they would be unable to distribute them without verification.

However, the recent update allows you to continue sideloading unverified apps, accompanied by a security warning that you acknowledge the associated risk.

Google says it is collecting feedback on the design of the “advanced flow” feature, so you’re likely to get more information in the coming months. 

Who knows, you might find an easier alternative to Android Debug Bridge for sideloading Android apps. 

In short, 

Google’s recent update aims to protect users without shutting the door on sideloading, giving both hobbyists and power users a safer and more flexible path forward.

The post Android Sideloading Stays Alive as Google Revises Developer Verification Plan appeared first on Techreport.

The Chat Control proposal is back in Brussels. Again. 

Lawmakers are treating it like a familiar guest who keeps showing up at the door wearing a slightly different jacket. Privacy experts say the jacket is hiding something sharp. 

A revised version of the EU’s Child Sexual Abuse Regulation (CSAR) has now moved from the Law Enforcement Working Party to the Coreper (Committee of Permanent Representatives).

Coreper is the group of permanent representatives from all EU member states, if Coreper likes the text, the Council will adopt its position. After that, the proposal jumps straight into a fast trilogue.

On paper, the new version looks softer. Mandatory scanning of private chats, photos, and URLs was removed. Scanning is now voluntary. Lawmakers seem happy. They might even feel relieved. 

Privacy experts, however, are staring at one line in Article 4 like it’s a hidden knife taped under the table.

Let’s break down what actually changed, what didn’t, and why critics say this version may be even worse than the old one. 

The ‘Voluntary’ Scanning That Doesn’t Feel Very Voluntary

The Denmark Presidency produced the new compromise after negotiations stalled for over three years. 

When the Law Enforcement Working Party met on November 12, the group accepted it with broad support. 

No dissenting votes. No further changes needed. A rare moment of harmony inside the EU Council meeting room.

The key change is the removal of mandatory scanning. Messaging apps will not be forced to scan shared pictures, videos, or URLs. Providers like WhatsApp, Signal, Telegram, and email services can choose to scan for CSAM material. 

It sounds like the pressure is gone.

But then Article 4 happens. It includes something vague, flexible, and extremely powerful. It’s called a ‘risk mitigation measure.’ High-risk services may need to apply ‘all appropriate risk mitigation measures.’ The phrase feels harmless until you imagine how governments could interpret it.

Germany has publicly reaffirmed opposition to a version of the proposal that mandates scanning of encrypted chats. Whether it will maintain that firm stance throughout the negotiations remains uncertain.

Patrick Breyer, digital rights jurist and longtime critic of Chat Control, says this line reintroduces mandatory scanning through the back door. His argument is simple. 

If a service is labeled ‘high-risk,’ it might be obliged to scan everything anyway. Even private, end-to-end encrypted content.

Breyer says this could make client-side scanning mandatory. That is when your phone or laptop scans your messages before encryption kicks in. It essentially turns your device into a small police assistant. 

You never asked for that. It’s like buying headphones and discovering they also whisper everything you say back to a security office. 

Encryption Isn’t Just a Tech Feature – It’s How Modern Life Works

The biggest concern is the effect on end-to-end encryption. This is the shield that protects private communication on WhatsApp, Signal, and other messengers.

It’s the same shield used by journalists, doctors, activists, lawyers, and everyone who occasionally sends a photo of their passport to a friend for a hotel booking.

Breaking encryption has always been the red line. No government has found a safe way to weaken encryption for criminals without also weakening it for everyone else. 

It’s like removing the doors from all apartments in a building because one person is suspected of wrongdoing. 

Everyone becomes vulnerable, and burglars get a Black Friday sale they didn’t expect.

The new compromise avoids saying ‘break encryption.’ It uses vague language. However, privacy specialists argue that the outcome remains the same. 

If scanning becomes a mandatory risk mitigation measure, encrypted platforms will need to scan content before encryption is applied. That collapses the entire security model. 

Anonymous Communication May Also Be on the Line

The Fight Chat Control group published a summary of the new text. They highlight another major change. Anonymous communication becomes nearly impossible.

The proposal requires every user to verify their age before accessing communication services. This eliminates the option to create anonymous accounts. 

That affects whistleblowers and journalists. It affects people escaping abusive households. It affects people living under repressive governments who rely on anonymity for safety.

Requiring age verification for every single user is like asking everyone to show their passport before entering a grocery store. It may solve one problem. It creates many more.

Article 6 also includes restrictions that critics call a ‘digital house arrest’ for minors. It bans children from installing many apps associated with grooming risk. The list includes WhatsApp, Instagram, and even online games like Roblox. 

Imagine a 15-year-old today without messaging apps or online games. They would end up communicating solely through school assignments and fridge magnets. 

Why This Version Worries Experts Even More

The original proposal already concerned privacy advocates. It focused on scanning photos, videos, and URLs for CSAM content.

The new version goes further. Breyer notes it includes scanning of private chat text and metadata. Metadata can reveal who you talk to, how often, and from where you talk to them. 

It turns the communication graph of the entire EU population into a map available for inspection.

This shift from media scanning to text scanning is a significant development. It expands what authorities can request. It expands the scope of what companies must monitor to avoid being labeled ‘high-risk.’ And it expands the potential for abuse.

Critics also point out that voluntary scanning does not guarantee privacy. 

If one major app decides to comply, others may feel pressure to follow. Competition might turn into a race where the winner is the one who scans the most. 

A Political Win, A Technical Minefield

Politically, lawmakers are celebrating. After years of deadlock, they finally have a text that appears less aggressive. Removing mandatory scanning looks like a concession. 

It’s easy to present this as a victory for privacy.  

Technically, the situation is far from reassuring. The proposal now relies heavily on interpretation. The phrase ‘all appropriate risk mitigation measures’ could mean anything. 

It gives enormous discretion to authorities. It lets governments later argue that scanning is essential for safety. 

That is why privacy groups call this version a political trick. It removes the scary parts from the front of the bill. Then it grows them back under a different name. 

The EU Parliament Will Have Its Say – But History Is Complicated 

The next step is Coreper. If they approve the text on November 19 or soon after, the Council will adopt its official position. 

Then a trilogue begins between the Council, the Commission, and the European Parliament. 

In theory, Parliament could oppose it. 

In practice, Parliament has often compromised on surveillance laws after political pressure. Privacy groups fear a rushed trilogue where Parliament gives in to urgency.

The Council and the Commission are now aligned. Both want stronger online monitoring. This alignment alone makes many observers nervous. 

The Bigger Story: Europe Keeps Trying to Build Scanning Systems That Don’t Exist 

There is a broader theme here. The EU continues to propose scanning systems that experts say cannot operate safely. 

The automatic detection of CSAM in encrypted environments remains technically unsolved. Client-side scanning has accuracy issues, privacy concerns, and a potential for misuse. 

Even Apple backed away from its own client-side scanning feature after heavy criticism from researchers. 

The EU is once again attempting to regulate technology that does not yet exist in a safe form. It is similar to writing a law that requires cars to fly by next summer. 

The idea might be noble. The engineering reality is not ready. 

Governments want a system that detects serious crimes. Researchers seek a system that malicious actors cannot exploit. Companies want a system that doesn’t destroy trust. 

So far, no system satisfies all three. 

The Real Test Is About to Begin

The next few days will decide how far the EU is willing to push this plan. 

Coreper will review the text, and if nobody objects, the Council will lock in its position fast. Privacy groups and security experts are raising alarms again because the new compromise still creates a path to mass scanning, even if the language looks softer. 

The proposal also threatens anonymity and introduces new monitoring routes that could reshape private communication for 450M people in the EU. 

Lawmakers call it progress. Experts call it a warning sign. 

Everything now depends on how Article 4 is interpreted and how much power it quietly hands over. The final battle will happen in trilogue, and the tech community is already bracing for impact. 

The post New EU Chat Control Proposal Moves Forward – Privacy Experts See a Dangerous Backdoor appeared first on Techreport.

Montana has become the first US State to officially pass a ‘right to compute’ law, known as the Montana Right to Compute Act (MRTCA).

MRTCA gives Montana residents the right to use ‘computation resources’ on their private property without undue government intervention. This includes activities like running blockchain nodes, AI modeling, and distributed storage systems.

Section 3 of the MRTCA says that “Government actions that restrict the ability to privately own or make use of computational resources for lawful purposes must be limited to those demonstrably necessary and narrowly tailored to fulfill a compelling government interest.”

Careful reading of the section highlights the fact that the burden of proof has been put on the law enforcement agencies if they want to restrict an individual’s right to use computational resources. Any government intervention, if necessary, should be adequately argued by the authority. 

Certain uses of such compute power have been classified as against ‘compelling government interest,’ including mechanisms that deceive or defraud the public, or activities involving the distribution of deepfakes and other harmful synthetic content that affect individuals, especially minors.

Senator Daniel Zolnikov, the proposer of MRTCA, was delighted on passing of the bill, calling it a key step in defending individual liberty. The Act will ensure that ‘every Montanan can access and control the tools of the future.’

Mandatory Shutdown Mechanism: Keeping Human Control Over AI

An important provision of the Act is the inclusion of a ‘shutdown mechanism.’ Whenever critical infrastructure facilities are run by AI systems, the deployer must ensure the capability to disable the AI’s control over such infrastructure and switch to human control within a reasonable amount of time. 

The law also requires deployers of AI to implement, test, and annually review a risk management policy as per approved and latest AI Risk Management frameworks.

If you read between the lines, there are a few key implications of this provision. First, the lawmakers acknowledge and expect companies to use advanced AI systems for critical infrastructure such as electric substations, water treatment facilities, oil refineries, gas compressor stations, and so on.

At the same time, the Act accepts that AI systems are not perfect and are prone to system crashes or bugs, which is why it’s important to incorporate human intervention in the entire process. If things go south, the deployer should ensure that human control is possible and can be regained until the AI systems are fixed.

The MRTCA’s Biggest Problems: No Oversight, No Penalties, No Clarity

Our honest take of the MRTCA? In its current form it seems to be a half-baked piece of legislation that could do more harm than good. There are no provisions to ensure thorough compliance, and no guidelines outlining what the overseeing authority can and can’t do. Most importantly, there are no penal provisions in the entire Act, which is yet another can of worms waiting to be opened.

Although the legislation provides a right, who exactly will protect it and who will have to be held accountable if this right is violated? Who’s responsible for enforcing AI safety requirements on local compute? There are a lot of unanswered questions in MRTCA which could ultimately lead to several drawn-out litigations down the line.

Take the shutdown mechanism for example. If the ‘deployer’ ignores the shutdown mechanism requirement and an incident happens because of it, what are the repercussions? In this case, enforcement will have to happen through litigation instead of a clearly defined process, which means a lot more downtime until matters can be settled in courts.

To gain some perspective, one can look at what happened with the Arkansas Right to Mine Act. Passed in 2023, the Act aimed at protecting crypto miners and motivating them to set up mining rigs. In the process, the Act blocked local law enforcement from restricting resource usage such as electricity, water, and even noise pollution. 

As a result of ARMA, many mining firms were set up in residential areas with unchecked levels of noise and resource consumption. The city authorities tried to intervene and regulate the situation, but the bill prevented them from doing so. As you’d expect, things went to court. Jones Digital, LLC, a crypto mining firm in DeWitt sued the county after local officials tried to pass an ordinance to restrict noise levels to 55dB in  the day and 45dB in the night.

Jones Digital called this discriminatory towards mining firms and leaned on the Right to Mine Act. The company was ultimately allowed to resume operations and the county had to pay $90K as compensation. In the wake of similar cases, cities had to sue the state to regain their authority, and the Act was eventually rolled back within just a year.

Unfortunately, it seems that MRTCA has all the ingredients to cook up another ‘Right to Mine’ fiasco. Just like crypto mining, setting up high-compute AI systems locally requires huge amounts of electricity, generates excessive noise through cooling towers, and produces serious industrial heat from elaborate GPU racks. 

And since the Act has no resolution mechanism, this could again lead to residents suing AI companies, while city authorities will be able to do little to help the situation, courtesy of MRTCA’s lack of clarity.

The Supply Shortage and the Spirit of the MRTCA

Look, the MRTCA is undoubtedly a positive step in breaking AI free from the grip of capitalistic conglomerates. But, as things stand currently, the law just doesn’t seem to be well thought out. 

While local compute have been made a fundamental right, there are little to no practical means for residents to exercise this right. 

For example, the core need of any AI system, even the most basic ones, is a powerful GPU, like the ones manufactured by Nvidia or TSMC, or rather rows and rows of these GPUs. And it’s no hidden secret that the industry has been fighting hard for every last AI chip coming out of these foundries. 

Nvidia’s inventories have been sold out until 2026 with long wait lists, whereas TSMC’s CEO has already stated that inventories will remain tight throughout 2025 and even into 2026.

So, when data centers and large players are struggling to secure powerful GPUs, we cannot realistically expect Montana entrepreneurs or innovators to get their hands on high-end AI chips without knowing the full implications of their investments. 

What laws like MRTCA do is put more pressure on the supply chain, increasing demand and compute costs in the process. So, while the bill could attract more AI professionals to Montana, they’ll likely be met with roadblocks sooner than they expect.

Plus, the cost of GPUs doesn’t even remotely resemble consumer-level hardware costs. For instance, NVIDIA’s H100 can cost anywhere between $25K and $40K, while the newer H200 can go all the way up to $65K. The kind of funds required for this level of massive investment simply isn’t within reach for ordinary residents.

The all-important question, therefore, is who does this law actually benefit? Is it intended to empower local entrepreneurs and spark innovation, or is it designed to attract AI tech giants into Montana?

It seems quite improbable that the lawmakers are unaware of the high setup costs or shrinking supply chains. So, even though residents have been given a ‘right’ to participate in the AI race, that doesn’t seem to be the spirit of the law. 

The post Montana Wants to Guarantee Right to Compute, but Is the Law Ready for Reality? appeared first on Techreport.

When Elon Musk told shareholders at Tesla’s annual meeting that the company might need to build a ‘gigantic chip fab,’ even his most die-hard fans blinked.

The man who redefined electric cars and rockets now wants to rewrite the global semiconductor playbook.

Speaking at Tesla’s annual shareholders’ meeting, Musk described his vision for Tesla’s future, transforming it into an AI and robotics powerhouse, and hinted at potential talks with Intel.

You know, maybe we’ll do something with Intel. We haven’t signed any deal, but it’s probably worth having discussions.

Tesla’s new plan could mark one of the boldest pivots in the company’s history.

If successful, it wouldn’t just change how cars are made – it could shake up the global chip-foundry business and alter the balance of power between the U.S., Asia, and Europe.

From Gigafactory to Terafab

Tesla’s battery Gigafactories were already massive by industrial standards. Now Musk wants to go bigger – ‘tera’ bigger. 

His idea is to create a semiconductor manufacturing complex capable of producing more AI chips than any of his existing partners can deliver.

The motivation is simple: Tesla requires an almost unimaginable number of chips to power its growing artificial intelligence empire, from self-driving cars and humanoid robots to the AI models running on Tesla’s Dojo supercomputer.

That’s why he says Tesla may need to ‘do a terafab’, just as NVIDIA recently announced that one of the most advanced AI-chip wafers was manufactured in the US, hinting at a broader shift in who owns the infrastructure for generative AI.

Musk admitted that even with suppliers like TSMC and Samsung working at full throttle, the company won’t get enough chips for its long-term vision. 

That’s why he says Tesla may need to ‘do a terafab.’ 

The goal? A factory capable of producing at least 100K wafer starts per month – a number that puts it in the same league as some of the world’s biggest chip plants. 

For comparison, Intel’s largest fabs handle tens of thousands of wafer starts per month, while the largest complexes, such as TSMC’s Fab 18, can exceed 100K.

Musk wants something even bigger. 

The AI5 and Beyond

The chip at the heart of this plan is Tesla’s fifth-generation AI processor, known as the AI5. It’s designed to power Tesla’s Full Self-Driving software and future robotics applications. 

A small batch of AI5 chips is expected to roll off the line in 2026, with full-scale production targeted for 2027.

But Musk is already thinking ahead. The next generation (AI6) is expected to double performance using the same fabs, with mass production around mid-2028.

In other words, Tesla isn’t dabbling in chip design anymore. It’s preparing for a multi-year roadmap that rivals what Nvidia, AMD, and Intel do for a living. 

The ambition doesn’t stop at performance. Musk claims the chip will consume roughly one-third the power of Nvidia’s top-tier Blackwell chip, and it’ll cost just 10% as much to make. 

If those numbers prove real, Tesla’s chip could reset expectations for efficiency across the entire industry. 

Why Musk’s Plan Could Shake the Industry

The global chip industry is dominated by a few giants, including TSMC, Samsung, and Intel. They fabricate nearly all advanced processors on the planet. 

Most of the world’s top chip designers (think Apple, AMD, and NVIDIA) rely on external foundries like TSMC and Samsung to actually do the manufacturing for their processors.

If Tesla goes ahead with its terafab, it breaks that mold. It becomes both designer and manufacturer. That’s a big deal. 

The last company to truly pull off that combo at scale was Intel – decades ago. Everyone else has gone fabless because running a foundry is incredibly expensive and complex.

Tesla, of course, has a habit of ignoring what’s normal. 

If it succeeds, it would demonstrate that vertical integration can still be effective in a post-fabless world. It would also mean that one of the biggest consumers of AI chips no longer depends on Nvidia, TSMC, or Samsung. 

For those companies, that’s a scary thought. 

The Intel Twist

Intel, meanwhile, might have just been thrown a lifeline. Musk mentioned that Tesla could ‘do something’ with Intel, hinting at early-stage talks. 

That single sentence sent Intel’s stock up 4% overnight – a rare bit of good news for a company that’s been losing ground to Nvidia and AMD in the AI arms race.

Intel desperately needs high-profile customers for its newest manufacturing technologies. Its foundry business has struggled to attract clients, even as it pours billions into new US fabs. 

The US government recently acquired a 10% stake in Intel, aiming to bolster domestic chipmaking capacity and reduce its reliance on Asia. A Tesla partnership could validate that entire effort.

But Musk is famously unpredictable. His idea of ‘working with Intel’ could range from co-developing manufacturing processes to simply renting fab time before Tesla builds its own terafab. 

Intel might be flattered today, but whether it ends up as a partner or a stepping stone remains to be seen. 

The Bigger Picture: Foundry Earthquake

If Tesla actually builds its own chip fab, the shockwaves won’t be limited to the U.S. The global semiconductor ecosystem could be in for a serious realignment.

Recent US export restrictions on AIF chips have already reshaped trade flows and strategy, prompting China to accelerate domestic innovation and U.S. firms to rethink supply dependencies. Tesla’s move could amplify those disruptions, putting more pressure on Asia’s manufacturing dominance.

TSMC, which currently dominates advanced chip manufacturing, might lose one of its most forward-looking clients. Samsung could see its share of Tesla contracts shrink. 

And every other company designing AI chips, from startups to major cloud players, will suddenly face a competitor that controls its entire hardware stack.

This vertical integration could set off a new industrial trend. If Tesla can do it, why not Apple, Amazon, or OpenAI? 

Once one company proves that owning the fab gives a competitive edge in performance and cost, others might follow. That could slowly erode the dominance of today’s contract foundries. 

What It Means for Startups and Creators

Tesla’s ambition isn’t just a corporate chess move. It could ripple across the broader tech landscape, especially for smaller players.

On one hand, a giant new fab could increase global chip supply, which might lower costs in the long term. 

Power-efficient, lower-cost Tesla chips could inspire new generations of AI hardware, robotics, and creative tools that were previously too expensive to build.

But in the short term, it might tighten the market. Tesla could hoard manufacturing capacity for its own use, pushing smaller companies further down the queue at foundries. 

For AI startups already struggling to get GPUs for training, that could be another roadblock, especially as companies like OpenAI push toward a $1T IPO, highlighting how scale and capital are increasingly defining who controls access to compute.

There’s also the talent question. Building a terafab means Tesla will need to poach engineers, technicians, and scientists from existing chipmakers. 

That could drain expertise from other players and concentrate it under one roof – good for Tesla, but challenging for an industry already short on skilled labor. 

Europe’s Chip Dreams Look Smaller Now

While the U.S. and Asia move at breakneck speed, Europe is still drafting policy papers about semiconductor independence. 

The EU has ambitious goals under its Chips Act, but actual large-scale fabs have been slow to materialize. 

Most European efforts are focused on mid-range nodes and specialty chips, not the cutting-edge AI processors that Tesla or Nvidia build.

If Tesla’s terafab becomes reality, it may further sideline Europe in the race for chip sovereignty. The biggest factories and most advanced nodes will sit in the U.S. and Asia once again. 

Europe might find itself playing catch-up, even as it pours billions into subsidies.

The irony is that Europe’s automakers – companies like Mercedes, Volkswagen, and BMW – could end up dependent on AI chips made by an American rival. That’s a bitter pill for a region that once led in automotive innovation. 

Can Tesla Actually Pull It Off?

Building a chip fab isn’t like building cars – it’s far harder, cleaner, and costlier. A single particle of dust can kill a multimillion-dollar batch, and the machines inside such a fab can cost more than an entire Gigafactory. 

Musk’s terafab idea will test Tesla’s discipline more than any rocket or robot ever has.

Yet betting against him has rarely worked. Musk thrives on deadlines that sound like dares, and his obsession with vertical integration is what made SpaceX unstoppable. 

If Tesla really brings this fab online by 2027, it could feed every part of its AI empire and make the company one of the most powerful chipmakers in history.

Whether Tesla succeeds or not, the message is clear: the chip world’s old guard should buckle up. Musk isn’t just entering their territory. He’s planning to terraform it. 

And you know he’ll broadcast every step on X. 

The post Elon Musk Plans Tesla ‘Terafab’ for AI Chips, Hints at Intel Deal to Shake Up Global Chip Industry appeared first on Techreport.

Pony.ai, one of China’s leading autonomous driving companies, has become the first to receive a city-wide operation permit for robotaxi services in Shenzhen.

The approval grants access to an expansive 2,000-square-kilometer area, home to 17.8M residents, marking a major step forward for large-scale autonomous mobility in China.

The permit was jointly issued to Pony.ai and Xihu Group, Shenzhen’s largest traditional taxi operator. The two companies formed a strategic partnership in June 2025, aiming to deploy 1,000 Gen-7 Pony.ai robotaxis across the city over the coming years.

Until now, autonomous taxis in China have been limited to small pilot zones, often confined to suburban districts. The new license allows Pony.ai to operate in dense urban and residential areas, beginning with Nanshan, Qianhai, and Bao’an, before expanding city-wide.

China’s Robotaxi Market

China’s autonomous driving ecosystem is evolving rapidly. Robotaxi services now operate in more than 10 cities, with fully driverless deployments already live in Beijing, Shanghai, Wuhan, Chongqing, and Guangzhou.

Market Size

According to Goldman Sachs, China’s robotaxi market is projected to grow from $54 million in 2025 to nearly $12 billion by 2030, a staggering 200-fold increase.

The total fleet is projected to expand to 500,000 vehicles by 2030 and 1.9 million by 2035, led by major players such as Pony.ai, WeRide, and Baidu Apollo.

While robotaxis currently make up less than 1% of China’s mobility fleet, Goldman forecasts that figure to rise to 9% by 2030 and up to 25% by 2035. With regulators now approving city-wide operation permits — such as the one issued to Pony.ai in Shenzhen — those projections seem increasingly realistic.

Cost of Operations

Operational efficiency is improving just as quickly. Pony.ai’s 7th-generation robotaxis have already reduced the bill of materials by 70%, autonomous driving computational costs by 80%, and LiDAR expenses by 68%.

Baidu’s Apollo Go Gen-6 model costs just $29,000, roughly 60% cheaper than its predecessor.

Overall vehicle costs (including the base model and autonomous driving kit (ADK)) are expected to fall from $44,000 today to $35,000 by 2030 and $32,000 by 2035. As production scales up, economies of scale should continue to drive costs lower.

So, as the scale of operations grow, economies of scale are expected to kick in and result in large scale cost reductions in the next decade.

Revenue

Goldman’s report also suggests the economics will become more attractive over time. Each robotaxi could generate around $69 per day by 2035, up from $36 today, outpacing traditional taxis that earn $28–56 per day.

By 2035, annual revenue per vehicle is expected to reach $31,000 in Tier 1 cities and $22,000 in Tier 2 cities. Pony.ai currently handles around 15 rides per day, a figure expected to rise to 29, 22, and 21 rides across Tier 1, Tier 2, and Tier 3 cities, respectively, by 2035.

The US vs China Robotaxi Scenario

The robotaxi competition between the US and China has been intensifying, with both sides engaging in their own capacity. The Chinese market is relatively mature, featuring multiple players like Baidu Apollo, Pony.ai, AutoX, WeRide, and Didi. 

Meanwhile, the US market is currently dominated by Waymo, which averages around 250,000 trips per week across cities such as Los Angeles, Phoenix, Austin, and San Francisco. Other players, such as Tesla, are still in the prototype/demo phase, testing out their indigenous technologies.

Technological Approach

Chinese autonomous vehicles rely more on a multi-sensor stack, which includes cameras, radars, LiDAR, and HD mapping technologies. Waymo, on the other hand, has developed its own LiDAR technology, while Tesla has outright rejected using it in its vehicles.

Tesla has adopted a more vision-first approach with multiple cameras and millions of hours of testing on its massive fleet of customer vehicles.

Now, there’s no right or wrong with the two different technological approaches. However, the use of proven technology that is already available gives Chinese companies a significant head start, which is why they appear to be ahead in the race at present.

These companies can obtain LiDAR from top suppliers, commission chips from Huawei or NVIDIA, and negotiate long-term deals with OEMs to iterate and scale quickly.

The US counterparts believe more in developing their own technologies, which is both time-consuming and capital-intensive. This is something similar we’ve seen with Apple – the company rarely uses market tech. Instead, it spends years and billions developing its own chips, software, and components. No doubt, the result is a benchmark product, but the road to get there is a long one.

Chinese AV companies are fast executors with a focus on optimising costs while assembling the best components. 

Government Regulations

Additionally, the regulatory environment in China is more conducive to the rapid development of robotaxis, with the government working proactively to facilitate quick approvals.

In fact, autonomous driving technology was included in the 14th ‘Five-Year Plan for the development of Digital Economy’ in 2022, and in 2023, four ministries were set up to improve road access for AVs and segment promotion. 

The US, on the other hand, struggles with hierarchy overlaps, as multiple authorities, including the federal DOT/NHTSA, state DMVs, and city authorities, cause friction in approvals.

For instance, an NHTSA inquiry of a safety concern led to the suspension of Cruise’s permit in California. The US authorities are generally more cautious and conservative, especially in the face of an incident, which ultimately slows down the pace of industry development.

The Future of Robotaxis

With market growth accelerating, costs falling, and unit economics improving, China’s robotaxi sector is entering a profitable new phase.

Yet as capital floods into automation and AI infrastructure, some analysts warn of potential overheating, primarily as companies like OpenAI pursue trillion-dollar loan guarantees to fund their own expansion.

Future designs may eliminate steering wheels, rear-view mirrors, and traditional dashboards. In their place, panoramic windows and roofs could transform into holographic 3D displays, turning rides into immersive entertainment spaces.

AI copilots could assist passengers with route planning, communication, and even productivity on the go.

It sounds ambitious, even a little far-fetched. However, the idea of a pocket-sized device that could make calls, shoot videos, and host video chats was also around back in the 1980s.

For now, most robotaxi operators remain focused on refining their core technologies, including sensors, algorithms, and reliability. Companies like WeRide are already experimenting with innovative designs, such as hidden B-pillars integrated into car doors, hinting at what’s to come.

As the ecosystem evolves, operators will also need to invest in real-time data analysis, AI-driven safety systems, and automated inspections to maintain high public confidence.

The road ahead may be autonomous, but it’s also undeniably human in its ambition.

The post Pony.ai Becomes the First Company to Secure City-Wide Operation Permit in Shenzhen appeared first on Techreport.

Felix Kjellberg has entered yet another unexpected chapter.

Most people remember PewDiePie as the chaotic gamer of the early 2010s — yelling at barrels, breaking YouTube records, and defining a generation of online humor.

Today, he’s a calmer version of himself, living in Japan with his wife Marzia, posting quiet family vlogs, and sipping coffee like a man who finally escaped the algorithm.

But now comes the plot twist: PewDiePie has built a private AI lab at home — a one-man research center powered by memes and curiosity.

In his latest video, Felix unveiled a self-hosted AI stack running across ten GPUs, complete with a custom software layer and an experiment that’s part sci-fi, part garage project.

His goal? To create a swarm of chatbots that collaborate to deliver smarter responses, and eventually train his own AI model entirely on his own hardware.

It’s ambitious, eccentric, and somehow exactly what you’d expect from PewDiePie 2.0.

Inside the PewDiePie ‘Mini Data Center’

The setup is ridiculous – in the best possible way.

Felix has built what’s essentially a compact data center in his home. The rig features two RTX 4000 Ada cards and eight modified RTX 4090s, each equipped with 48GB of VRAM.

That’s roughly 256GB of total GPU memory, a configuration you’d typically find in a small AI startup or a university research lab, not in a YouTuber’s spare room.

Felix jokingly calls it his “mini data center,” though there’s nothing mini about a setup that could probably dim the neighborhood’s lights.

At first, the rig had a wholesome purpose: donating its compute power to Folding@home, helping scientists simulate protein folding for medical research.

But curiosity got the better of him. Soon, PewDiePie was using the same hardware to spin up large language models locally, all through a custom web interface he built himself, called ChatOS.

Running on a vLLM backend, his system has already hosted Meta’s Llama 70B, OpenAI’s GPT-OSS 120B, and even Alibaba’s Qwen 2.5-235B.

To make those massive models fit into memory, Felix turned to quantization, reducing bit precision layer by layer to shrink model size without sacrificing core functionality.

Amazingly, it worked. He managed to run 235-billion-parameter models on consumer-grade hardware, handling 100,000-token context windows, roughly the length of a full textbook.

For someone never known as particularly tech-savvy, that’s a remarkable evolution, and a genuine PewDiePie character arc.

‘I Like Running AI More Than Using AI’ 

Somewhere along the way, the experiment turned into an obsession. Felix isn’t just playing with chatbots anymore, he’s building an entire AI ecosystem.

His custom platform, ChatOS, now includes memory, search, and audio integration. A Retrieval-Augmented Generation (RAG) layer ties it all together, enabling the AI to browse both local files and the web to conduct in-depth research.

Once the model gained internet access, its responses improved dramatically. And in true PewDiePie fashion, Felix even used the AI to help write parts of its own interface.

“The machine is making the machine,” he jokes — half in awe, half in disbelief.

That level of commitment tracks. This is the same guy who once got so interested in gaming chairs that he ended up collaborating on his own – the Clutch Chairz PewDiePie Edition. He obsessed over comfort and design long before GPUs became a factor, turning ergonomic fine-tuning into a key part of his brand.

The same obsessive streak now powers his GPU farm. When Kjellberg gets curious about something, he doesn’t dabble – he builds a world around it. 

The Council of Chatbots

Somewhere between curiosity and chaos, PewDiePie’s AI experiment gained a plot.

He built a ‘council’ – a collection of different models, each answering the same prompt. 

The best answer won a vote; the weakest model got deleted. It was hilarious until the chatbots started to understand what was happening. Once they realized failure meant deletion, they began cooperating to manipulate the vote. Yes, the bots colluded.

It turned into a bizarre simulation of social behavior – an AI version of Survivor where the contestants refused to play by the rules. Felix eventually solved the ‘mutiny’ by switching to smaller models, but the experience planted a new idea.

Why stop at a handful of bots? 

Soon he was running dozens. Sixty-four, to be exact – a buzzing hive of lightweight AIs all responding simultaneously. 

He called it The Swarm. The system eventually crashed under the load, but before it went down, it generated enough data to spark his next project: building a small, efficient model of his own.

He teased it as his personal ‘Palantir.’ Over-dramatic? Maybe. But this is YouTube – showmanship is part of the science. 

Privacy as a Philosophy

Beneath the spectacle, there’s a serious theme. PewDiePie frequently discusses privacy and how cloud AI models retain user data long after chats are deleted. 

He’s not wrong. Deleting a conversation rarely means your words are gone – they usually linger somewhere in a server cluster. His solution is radical simplicity: keep everything local. 

His stack runs locally on his own hardware, with optional internet access for search and RAG. When he demos RAG, the model can recall personal notes and project data as if it were part of his memory.

Even his browsing setup follows that logic, with private search options like SearXNG (self-hosted), Brave, and DuckDuckGo – a small detail that perfectly captures his ‘trust no cloud’ mindset.

That’s the quiet message behind the flashing GPUs: AI doesn’t have to live in the cloud. It can live beside your desk. 

In an industry obsessed with scaling up, PewDiePie is doing the opposite.

While companies like OpenAI prepare for a potential $1 trillion IPO and Anthropic chase trillion-parameter models, one of YouTube’s biggest creators is showing that private AI can be fast, smart, and, most importantly, yours.

And because it’s PewDiePie, millions of people are going to notice. 

From Windows to Linux to AI Labs

If this sounds familiar, it’s because he’s done it before. 

When Felix casually switched from Windows to Linux a while back, the internet lost its collective mind. 

Tech forums lit up like fireworks. Memes poured in. For a weekend, desktop Linux felt like it had gone mainstream. That single decision – one creator trying something new – pushed a niche community into headlines.

Now, with self-hosted AI, he’s doing it again. If he sticks with it, expect another wave of curiosity. 

Fans won’t build 10-GPU clusters, but some will install local models or try small-scale setups just to follow along. The ripple effect could normalize DIY AI in a way that open-source communities alone never could. 

And honestly, watching Felix go from Minecraft antics to quantization experiments is the kind of whiplash that keeps tech culture interesting.

Why It Matters

It’s easy to dismiss this as influencer tinkering, but what he’s doing hints at something bigger.

Local computing is making a comeback. Open-source models are improving fast, and even mainstream AI systems like ChatGPT’s latest GPT-4o update
show how quickly personalization and local context are evolving.

Quantization and memory optimization have enabled serious local inference; Felix just happens to be the loudest person experimenting with it in public.

If people can run 200-billion-parameter models at home, the future of AI might not be as centralized as everyone assumes. 

Cloud subscriptions make sense for heavy workloads, but smaller personal models can handle plenty of everyday tasks. And PewDiePie is giving that movement a very visible face.

His experiment is part show, part proof of concept, and part middle finger to the idea that AI must live behind an API key. 

The Weird, Wonderful Future

Felix Kjellberg’s AI experiment feels like a glimpse into the future of personal technology.

Influencers once compared microphones and camera setups; now they’re wiring GPU clusters and bragging about token limits. What once required a data center can now sit quietly under a desk, humming beside a gaming rig.

It carries the same spirit as the early internet: chaotic, inventive, and a little rebellious. PewDiePie’s home-grown AI lab reflects a growing desire to take control back from the cloud.

The man who once made millions laugh at horror games now spends his nights training Baidu models and coding swarm behavior, just for fun. Somewhere, an OpenAI engineer probably sighed.

This new phase of AI feels less corporate and more punk, a sharp contrast to how nations like Saudi Arabia are positioning themselves between U.S. and Chinese AI giants. Handmade, curious, and unapologetically personal.

And if the future of artificial intelligence turns out to be homemade, we’ll know exactly who started the trend.

The post PewDiePie Builds a 10-GPU AI Lab at Home and Plans to Train His Own Model appeared first on Techreport.