They should stop data centre construction: bring in a moratorium.

Governments and civil society would then have control and time, where they currently have none.

When Jacob Coxon broke free from Anthropic last week and voiced fears of possible human extinction echoed by AI insiders, it flipped AI leaders into advocating a “pause”.  Such calls have been made before but the context is now different.  AI is unpopular, wildly successful at attracting investment, and increasingly seen as already out of control.  If they can grasp it, this gives politicians and civil society a chance to take back control of the AI regulation agenda.

Politicians and campaigners need to cut through the welter of media coverage of AI triggered by the proposed ‘pause’ on extinction risks, the increasing reports of rogue agents hacking ‘secure’ systems and fevered speculation about finances and the ‘real’ motivations of AI billionaires backing the ‘pause’, to decide two things:

• what is our immediate objective in responding to this moment of flux?
• how can we bring that about?

The immediate objective should be to slow the development of AI models, as Dario Amodei argues in his 3,400 word essay, only not on his terms.  Buried under layers of expressions of thoughtful concern, Amodei is proposing a set of voluntary actions, with lots of vague caveats, to be decided on by Big Tech itself, which could then get legal force.

Instead society should treat the acceptance of a “pause” as an opportunity to slow the industry right down so as to allow time for proper scrutiny and accountability in the public interest, in order to frame regulations.  To ensure the pause is real, data centre construction should be put on hold.

Don’t Take The Bait

It will be tempting for politicians to treat the industry’s call for a slow down as a change of course, and even be tempted to use Amodei’s essay as a to-do-list.  They shouldn’t take that bait.

Amodei’s piece is in line with the established industry narrative:

‘this magical technology is so powerful that it can only be developed and controlled by us.  We are the magicians, the sorcerers, alchemists and wizards and we are your best chance of steering an AI-society to heaven rather than hell.  So leave it to us, and write our desires into law’.

Under challenge, Big Tech’s justification expands to the seemingly pragmatic:

‘this is just moving so fast, your old-school politics, governance systems and legacy media simply can’t get up to speed.  You will never understand it so you cannot decide on questions of need or justification’.

In this AI political economy, the role of the non-wizard class is then to buy AI, pay to educate people to use it, and enable its expansion into every field of human endeavour.  Pure and irresistible (if mainly imaginary) bio-physics of technological innovation set the speed and direction of AI, so there is no point in you questioning it.  Miss out and your economy will tank.

Escape Velocity & Technical Fixes

Ever since OpenAI’s release of ChatGPT in November 2022, the default industry strategy has been to reach escape velocity, both Business-to-Business through competition within their sector, and Business-to-Consumer in escape from scrutiny or control by politics, government and civil society.  Until recently it has mostly achieved that.

Alarming proofs that it cannot control its own creations has somewhat eroded Big Tech’s magical teflon PR-coating but by accident or design – probably the latter – AI companies are still trying to draw any political scrutineers onto their ground: of technical fixes agreed amongst themselves, rather than public interest.

Simple communications dynamics has given them an advantage over critics: they dominate the media.  Just a handful or so of AI CEOs are the larger-than-life principals in the story: Elon Musk with his dark imagery and attacks on empathy, Mark Zuckerberg with his gross ‘yacht’, the hyperbolic Sam Altman, and Dario Amodei with his Harry Potter glasses and ethical pondering.  Consequently they are the primary AI news providers, with new models, astonishing feats, economic hype and spine chilling warnings. As a small number of ‘colourful’ personalities they personalise the story and set the agenda which politicians consume.

Against that there are thousands of much less entertaining politicians, who mostly follow rather than lead, and billions of often confused individual voters and consumers.  Their story is simply much harder to tell.

So civil society and politicians must not import Amodei-type lists of fixes (such as ‘Embedded Evaluators’) as the industry framing and focus of the terms, duration or purpose of a ‘Pause’.

Instead they should listen to and involve the growing number of former AI insiders and watchers who both understand the workings of Big Tech and have a wider social perspective, and set their own agenda for a public negotiation of what a ‘pause’ should consist of.  Until now that would have probably been a futile exercise – why would the AI industry take any notice ?  Now the context [1] is different.

Losing The Justification for Data Centres

With current computing technology the ‘brute force’ model of LLM AI development which became the ‘race’ to ‘super-intelligence’ triggered by Open AI’s ChatGPT requires ever bigger and faster and more powerful data centres.

Now the likes of Anthropic, OpenAI and Elon Musk are backing a ‘pause’ in the race, the same logic supports a pause [2] in construction of data centres. Their justification was based on the belief that there would be an ever escalating need for increased data crunching capacity.  That justification has been lost.

The social significance is obvious.  Unlike developing, training, testing and building AI models which is done secretly in private digital domains, and in esoteric terms hard for the public or politicians to understand, data centres are a place where the AI race plays out in familiar and understandable real world terms.  Earlier this year Enrique Dans called them‘a structural friction point in the expansion of the AI economy’.

Politicians have been foisting the air, water and noise pollution with its impacts on health and property prices onto communities because they have adopted the industry narrative that the AI race is a strategic national imperative. Now that industry itself is saying more capable models are currently too dangerous to develop, data centres lose their imperative.

Data centres are physical concrete, steel, silicon chips, cooling systems and electronics.  Data centres for the AI race are worse neighbours than others because of their huge demands for continuous compute ‘density’ leading to vast requirements of energy inputs, and clean cooling water.  That and their noise is a major reason for local opposition across the world.

Data Centre construction, and with it the vast investment in Nvidia’s high performance chips, is also at the centre of the financial flywheel which has convinced politicians that AI is bringing an economic miracle.  The case for net economic growth caused by the actual use of AI, especially once social and economic costs are accounted for, is far weaker.

Data Centres were already a potential choke point for the commercial and geopolitical AI ‘race’ but now refusing permissions  for more would give voters and politicians real leverage in a reset of the political-big-tech relationship. In strategy terms, thanks to Musk, Amodei and Altman espousing a ‘pause’, opposition to data-centres has turned from a bottom-up ‘bushfire’ tactic to one with far more strategic potential, at least for now.

Civil society and governments should seize this moment, whether or not a ‘pause’ also deflates the investment bubble, which would make the task easier.

***
[1] Context is one of the seven key ‘message’ factors in CAMPCAT – see also pp25-6 in How to Win Campaigns: Communications for Change 

[2] A pause is also likely to make the public and politicians more aware of conversations which have been ongoing in AI-world for decades about the fundamentals of the current LLM-based models, which effectively assume that intelligence can emerge from increased ability with language.  Others say that humans learn and communicate not via words but with concepts.  That could make LLMs a dead-end technology, with big implications for existing investments. Whether concept-capable models would be safer or require more or less ‘compute’ is not clear to me.

***

This blog was first published on 14 September 2026 as Campaign Strategy Newsletter 127

Subscribe to these free newsletters at  www.campaignstrategy.org

Chris Rose

Quite where they go up and how they will work is not predictable but once there is a boundary where there was none before, finding yourself on the wrong side of it may have real consequences, even without regulations.  Once you are ‘on the wrong side’, moving may be painful, maybe impossible.

August 1961 the beginnings of the Berlin Wall – East German soldiers put up barbed wire to start with. Photo retronewser.com 

It reminds me of a story I heard some years ago on the BBC, about the erection of the Berlin Wall.  It featured an eye witness account from a British military officer* who went to assess reports that a barrier had started to be put up between the Soviet and British sectors of Berlin, in the early hours of 13 August 1961.

It went something like this:

“so I went to investigate and at [place?] I  found a detachment of British troops manning a position along the dividing line, watching East Germans putting in concrete foundations and a barbed wire barrier.  I asked them if they knew what was happening and they said ‘no’ but they had been wondering.  I said “those chaps over there are building a wall between the British sector and the Russians. You are on the wrong side of it.  So you had better move””.

The point being that although tensions had been rising for some time, when The Wall started to go up it took most people by surprise, even the troops (in several places what became the Wall did not exactly follow the sector borders, which up to that point were only marked by some rusty metal posts).

Big Tech AI is already causing widespread social anxiety on top of the impact of social media on children.  From data-centres and their water and energy demands to the erosion of truth, the disabling of learning in education and undermining of law, journalism and science, and the loss of authenticity from the creative arts.

photo: cbsnews.com

The AI spiral of blind innovation sucking in vast investment spins political FOMO – Fear of Missing Out – which hypnotises politicians, and that inhibits regulation. Last year in AI’s War on Truth  I suggested ten factors which might bring the AI spiral down to ground: impact on mental health, economics (eg financial implosion), product safety, Sovereign AI, fire-walling (the fences), environmental damage (via data centres), Creative IP Rights, technological/market limits, the volume of harms, and the scandal of ‘hallucinations’. Already some items on that list sound rather quaint, so fast has AI developed.

On the other hand, grass-roots opposition to data-centres is also spreading like wildfire, from Chile to Scotland to Australia to Malaysia. 70% of Americans oppose them being built in their area.  As to Big Tech, in California, Meta was forced to settle for $16.7bn in a social media addiction court case against 29 States, and it has had to abandon a plan to replace up to 60% of its own staff with AI.  A survey found the more scientists have used AI, the less they trust it and the less capable they think it is, while as models become more ‘capable’, it turns out they hallucinate not less as was expected but more.

Perhaps most serious for Big tech, in the US AI is even more unpopular than ICE. Worse still, young people particularly distrust AI bosses.  Pew found US concern about AI rose from 37% to 52% in 2026, and for the first time a majority under 30 are more ‘concerned’ than excited about it.  The current global MIT survey found the country with the greatest increase in people agreeing ‘products and services using artificial intelligence make me nervous’ was tech-obsessed India.

These are just some of the indications that the magical aura of Silicon Valley has dimmed significantly in less than a year, while concerns about AI have started to crystallise and opposition to organise.  Even governments like the UK’s which imported US-style unadulterated enthusiasm for AI in 2024, will now be infused with some doubt.

Partly to stay in line with requirements of the EU’s AI Act (and California’s AI Transparency Act) to make AI images or audio identifiable but also because of customer concern, Spotify and TikTok have both enabled or undertaken labelling of AI content. Likewise YouTube, LinkedIn, and Substack have also moved to identify, label, or limit some AI-generated content.  Even if nearly all the ‘rules’ are mostly voluntary, they indicate an unmistakable direction of travel as AI ‘de-socialises’.

Many universities are still committed to embracing AI at a corporate level but many academics are in despair at its impact on real learning, and some are resorting to old-school analogue and human face to face assessment to avoid the polluted digital domains.   Businesses need people as customers – they probably won’t be selling to agents anytime soon – and people have agency. Some will be putting up “AI-free” fences.

Civil Society – NGOs and not for profits – need to read these tea-leaves and decide which side of those fences they will be on, and soon.

Where the lines are drawn on “AI-free” will be case specific and driven by grossness of famous cases, opportunity and many other factors.  There may be no good homologue from the past.

As a pollutant of truth, AI is already similar to plastic. Its explosive global adoption was driven by being super-convenient, and it escaped effective regulation before it was realised to also be toxic.  Pervasive AI would still leave Pioneers to start ‘free from’ businesses as happened with GM- or pesticide-free food.  In the UK and probably elsewhere, some young highly educated people are already seeking out AI-proof authentic craft livelihoods.  AI might go that way: stimulating a lot of AI-free start ups and lifestyle choices.

Or, as some AI advocates hope, it will be like writing or the invention of printing, ultimately ethically and morally seen as in itself, neutral.  Given the centralised production and accrual of benefits as wealth, which is already causing all sorts of real world effects, personally I doubt that’s how it will go. The strategies (Peter Thiel’s dominant first mover) and the mentality (hyper-scaling domination) are more like the ‘Kings’ of the seven oil ‘Super Majors’ of the C20th than Johannes Gutenberg.

One of the worst cases for well intentioned AI users and developers is that it comes to be seen like tobacco and smoking, and thence progressive expansion of smoke-free-zones.  The dynamics of the child-harm social media court cases are already analogous: starting with places (the home, the child’s bedroom) which should be ‘sacrosanct’, and centred on ‘blameless’ victims (children and uninformed/ deceived parents).

BBC report

Meanwhile the financial markets are wondering about the growing gyre of investment circulating between tech supplier Nvidia and its AI firm customers, and asking who will survive when, rather than if, the bubble bursts, as it did with railroads and dot.com companies.  When that happens the effect on credit could reach far beyond ‘tech’ itself, similar to the 2008 credit crunch which triggered recession affecting all of society.

On ‘X’, analyst @HedgieMarkets posted on 25 August that ‘Anthropic is expected to tell IPO [its impending public stock market flotation] investors that its total addressable market exceeds $30 trillion. That’s roughly the entire GDP of the United States’. Anthropic reaches this astronomical sum by assuming it will take all possible business to be done with AI models. As Hedgie points out, institutional investors will recognise this as financial hyperbole  but ordinary retail investors who buy after them, may not.

A recent FT report ‘The Multiplying Risks of Financing Datacentres‘ noted ‘By 2030, tech companies are expected to pour $7tn into data centres — enough money to feed every person in China for three years’.

Although they control permitting of data centres and so could curtail or stop the entire spiral, few politicians have yet said much about whether this is actually a good use of investment.  Might it for instance, be better spent on hastening the real-world transition from fossil fuels to renewables to help fix climate change?  When the AI wonder-product hysteria started just a couple of years ago it was often suggested that AI could “help solve problems like climate change”.  You don’t hear that now, both because it’s patently obvious that we already know what’s needed, and because Gen AI has proven itself so unreliable in practice.

***

And finally, if you are in a public interest cause group, do think about the fence issue.  I had a feeling that the fence between human-made and AI made may turn out to be, like the Berlin Wall, the sort you can’t really sit on.

***

(Throughout this and my previous blog I’ve referred to ‘AI’ in an unqualified way, which is what almost all non-technical commentary does but in reality there are non-generative AI systems which often pose very little social risk.  Such as specialised medical applications. They however, usually have tiny requirements for data and processing compared to the ambitions of the ‘super intelligence’ chasers whose vision is for AI to run entire economies and societies instead of humans).

____________________________________________________________

*The story was in a BBC radio programme but I can’t find it online. There is one obvious candidate, Brigadier L F Richards in charge of the Military Police but it’s not in his published recollections.  If anyone knows where the recording is to be found, please let me know.

Modi was playing politics by convening a global ‘AI Summit’ and maybe didn’t care about the optics beyond that but it might turn out this was the moment it will be remembered for

Visual language is powerful – when it goes right or wrong.  On 19 Feb India’s PM Modi got surprised tech’ leaders at his AI Summit to join a ‘raised hands’. Sam Altman of OpenAI & Dario Amoedi of Anthropic awkwardly raised fists instead, visualising not just rivalry but the safety split in BigTech (Altman is right of Modi and Amodei to the right of Altman).

On Feb 18  @Ric_RTP reported on X that OpenAI, Google and X had agreed to Pentagon demands they remove safety restrictions for military use but Anthropic had refused:

“They don’t want Claude used to build fully autonomous weapons that fire without a human in the loop, and they don’t want it used to mass surveil American citizens”.

So Hesgeth threatened to blacklist Anthropic as a “supply chain risk” meaning no company could get US military contracts if it made any use of Anthropics model Claude.

@Ric_RTP concluded: “whatever Anthropic decides in the next few weeks, it sets the precedent for how much control AI companies actually have over their own technology.

Turns out the answer might be: none”.

I terms of setting up photo-opp’s … I guess the lesson is, check that the participants are onside before you start, and, think about it from the viewpoint of the media (alternative and unwanted visual messages it may generate) , if you want to avoid unintended visualisations.

In terms of Big Tech and it’s ongoing internal culture wars, Anthropic is heavily outnumbered in the US with the Trump administration maxing out it’s Power play in values terms at home and abroad but the US is not the world market, just the biggest single lump.

It’s long been the case that American corporates got angry (and now the American Administration) when they conflated the fact that no other country or countries could force America to do something it didn’t want to, with America being able to force the rest of the world to do anything it wanted.  Trump is testing that theory on coal v renewables at the moment.

There are several avenues by which Anthropic might yet emerge the winner on the safety-v-forget-safety split but it could equally well be the fall-guy.

For those who might no longer be on X or ever were, here’s what @Ric_RTP wrote in full:

——————————————————- X ————————————————

@Ric_RTP  Feb 18

The Pentagon just threatened to BLACKLIST one of America’s most valuable AI companies.

Not Huawei or some Chinese chip maker…

It’s ANTHROPIC. The company behind Claude. $380 billion valuation.

And the reason is genuinely insane:

For months, the Pentagon has been pushing every major AI lab to remove their safety restrictions for military use.

The ask is simple: let us use your models for anything that’s technically legal.

Weapons development, intelligence collection, battlefield operations, mass surveillance of American citizens.

OpenAI said yes.

Google said yes.

xAI said yes.

Anthropic said no.

Not to everything tho. They were willing to negotiate.

But they held firm on two things:

They don’t want Claude used to build fully autonomous weapons that fire without a human in the loop, and they don’t want it used to mass surveil American citizens.

That’s it. That’s the line they drew.

But Pete Hegseth’s response was to threaten to designate Anthropic a “supply chain risk.”

Here’s why that matters:

That label isn’t a contract cancellation. It’s not a fine. It’s not a strongly worded letter…

It means every single company that wants to do business with the US military has to certify they don’t use Claude anywhere in their operations.

8 of the 10 largest companies in America use Claude.

Defense contractors, government suppliers, enterprise companies with any federal exposure…

ALL of them would have to cut ties with Anthropic overnight or lose their government contracts.

A senior Pentagon official told Axios:

“It will be an enormous pain in the ass to disentangle, and we are going to make sure they pay a price for forcing our hand like this.”

That’s a US government official threatening to financially destroy an American company because it doesn’t want its AI used to spy on American people.

And it gets WORSE.

Last week, Anthropic’s head of safeguards research resigned.

His parting message: “the world is in peril.”

Elon Musk – whose xAI already handed the Pentagon a blank check – is now publicly attacking Anthropic calling Claude anti-human.

And the Pentagon official told Axios they’re “confident” OpenAI, Google, and xAI will all agree to the “all lawful purposes” standard.

So what you’re actually watching right now is every major AI company in America quietly handing the government unlimited access to the most powerful technology ever built.

With no guardrails.

No limits.

No company-imposed restrictions on what it can be used for.

One company tried to hold a line.

But the government is about to make an example out of them.

If Anthropic folds, it’s over.

Every lab just learned what happens when you push back.

And every restriction, every safety policy, every ethical guardrail these companies spent years building gets negotiated away behind closed doors the second the government asks.

If they don’t fold, a $380 billion company gets made radioactive in its OWN country.

Watch what happens next.

Because whatever Anthropic decides in the next few weeks, it sets the precedent for how much control AI companies actually have over their own technology.

Turns out the answer might be: none.

————————————————– X —————————————————–

From Stars and Stripes