Safety Pact or Cartel? The Fight Over Big Tech’s AI Slowdown

A loose weekend promise from four of tech’s most powerful men has triggered a fundamental debate within 48 hours: are OpenAI, Anthropic, Google DeepMind, and SpaceX responsibly slowing down AI development — or building a wall against smaller competitors under the guise of safety? While critics call it a cartel, US President Donald Trump calls the underlying concern a hoax.

Key takeaways

• Sam Altman, Dario Amodei, Demis Hassabis, and Elon Musk agreed over the weekend to a three-step plan to ‘pace’ frontier AI development, sparked by an essay from Anthropic CEO Amodei.
• Cohere CEO Aidan Gomez called the move ‘a cartel by another name’ in a blog post, arguing that independent auditors would ultimately be funded by the same companies they’re meant to police.
• Legal analyst Dirk Auer compares the plan to the EU’s AdBlue cartel case (2021, a €875 million fine against Daimler, BMW, and VW), calling a joint output slowdown ‘a textbook restriction on output.’
• Donald Trump called the AI warning a ‘hoax’ on Truth Social on September 14, 2026, comparing it to the climate change debate.
• China is warning against ‘malicious competition’ in AI development; Xi Jinping proposed a separate regulatory framework for the community of nations at the BRICS summit in New Delhi.

Who’s actually on board — and who isn’t

The trigger was an essay by Anthropic CEO Dario Amodei titled ‘We Must Pace the Frontier,’ building on an open letter of the same name signed by more than 1,000 AI lab employees in July 2026. Amodei’s plan lays out three steps: first, AI companies grant independent ’embedded evaluators’ direct access to their most advanced systems — Anthropic says it has already committed to this. Second, the rest of the industry is meant to follow. Third, an international agreement should follow, explicitly including China. Sam Altman agreed within hours (‘I agree with Dario that we need to pace the frontier’), Elon Musk offered a terse ‘Dario is right,’ and Demis Hassabis signaled agreement as well. What’s notably missing: neither Google DeepMind nor Musk’s own xAI has made binding commitments beyond verbal agreement so far.

The cartel accusation

Cohere CEO Aidan Gomez pushed back publicly: the required infrastructure of computing capacity, monitoring, and safety teams would realistically only be affordable for the largest labs, he argued, and the supposedly ‘independent’ auditors would be funded by the very companies they’re supposed to evaluate. Legal analyst Dirk Auer of Truth on the Market goes further in his September 14 analysis: ‘The second step of Amodei’s plan does not merely look like collusion. It is collusion,’ he writes, drawing a comparison to the EU’s Irish Beef case, in which a joint capacity reduction was ruled an anticompetitive restriction ‘by object.’ A close European precedent is the AdBlue case: in 2021, the European Commission fined Daimler, BMW, and Volkswagen a combined €875 million — the first EU antitrust decision based solely on slowed technical progress rather than price-fixing. Unlike the secret AdBlue meetings, an AI pact would play out publicly and under a safety rationale, which could theoretically trigger an exemption under Article 101(3) of the EU treaty for technical progress benefiting consumers — but competition lawyers describe that as a gray area, not a free pass.

Trump’s hoax claim and China’s counter-proposal

Trump didn’t respond with a speech but with two Truth Social posts on the evening of September 14: the claim that ‘AI is going to take over the world’ is a hoax, he wrote, ‘just like global warming,’ lumping the AI warning in with a list of what he called scandals pushed by the ‘Radical Left.’ A more formal White House-adjacent position came shortly after from David Sacks, the administration’s AI adviser until March 2026 and now chair of the President’s Council of Advisors on Science and Technology: labs should ‘pace’ themselves, he said, but without demanding an antitrust exemption or a Washington approval regime — liability should run through product-liability lawsuits after harm occurs, not pre-approval. Beijing offered a different kind of criticism. Foreign ministry spokesperson Guo Jiakun warned that ‘threat narratives’ and ‘malicious competition’ were disrupting global AI governance and served ‘no one’s interest.’ President Xi Jinping proposed at the BRICS summit in New Delhi that the community of nations should quickly reach consensus on a joint regulatory framework — a standalone counter-proposal to the US companies’ informal self-regulation.

What safety researchers actually say

Caught between the political camps is a third group: researchers who have called for a slowdown for years and are now watching warily who claims their idea as their own. ‘A lot of people are thinking of this as Dario’s idea, or it’s coming from the CEOs, but that’s false,’ Daniel Kokotajlo, a former OpenAI employee who now leads the AI Futures Project, told The Verge. People outside the companies have been calling for this for years, he said, and the CEOs are ‘now bowing to that pressure and also claiming credit for it, although not rightfully.’ Apollo Research CEO Marius Hobbhahn still called the move ‘one of the best things for safety in a long time if it actually happens,’ while Kokotajlo warned of ‘safety-washing’: labs could bring in external auditors and produce safety paperwork, he said, ‘but at the end of the day, it actually won’t slow them down very much at all.’ The debate was amplified by the resignation of Anthropic researcher Jacob Coxon, whose public farewell letter — accusing both OpenAI and Anthropic of recklessly racing toward self-improving superintelligence — was viewed more than 170 million times on X alone, according to The Verge.

Conclusion

That the very four men who benefited most from AI’s breakneck pace are now publicly urging caution naturally breeds suspicion — and recent security incidents at OpenAI and Anthropic give plenty of reason for genuine concern, not just PR. Whether this loose weekend promise turns into a binding, verifiable agreement, or ends up producing little more than expensive consulting contracts and a few seals of approval, remains an open question. What the past few days have already shown, though, is that the moment market leaders start talking publicly about slowing down together, antitrust law inevitably enters the conversation — and, as Trump’s hoax claim shows, the political response in the US now falls along partisan rather than technical lines. For background on what ‘pacing’ technically means, see our earlier explainer on the original proposal.