The loose agreement over the weekend of September 14 between OpenAI's Sam Altman, Anthropic's Dario Amodei, Google DeepMind's Demis Hassabis, and xAI's Elon Musk to slow AI development is drawing accusations of cartel behavior almost as fast as it is drawing praise. The four signed on, at least in part, to a three-step proposal laid out in an Amodei essay: embed third-party auditors inside frontier labs, regulate domestic development, and reach a global slowdown agreement. Critics counter that the plan is calibrated to kneecap open-source competitors and smaller rivals while dodging real legal safeguards. The verbal pact carries no enforcement mechanism, no timeline, and no measurable compute cap.
The framing matters because there is currently no US federal AI regulation in sight beyond the pre-release model review periods labs have already agreed to under the Trump administration. That leaves a voluntary industry pact as the only near-term brake — and gives the labs writing the pact enormous latitude to define what 'slowing down' actually means. Multiple safety researchers told The Verge the direction is correct but the substance is thin.
Nick Reese, an adjunct professor at New York University and former director of emerging tech policy at the Department of Homeland Security, said he does not think the agreement is 'all hot air' but argued the industry has outgrown its current spokespeople.
“The industry as a whole needs new champions.”— Nick Reese, Adjunct professor at New York University and former DHS director of emerging tech policy
Key facts
- 01Sam Altman, Dario Amodei, Demis Hassabis, and Elon Musk loosely agreed over the weekend of September 14, 2026 to 'pace the frontier' of AI development.
- 02Jacob Coxon's resignation letter from Anthropic has drawn more than 170 million views on X, breaking AI safety concerns into the mainstream.
- 03More than 1,000 AI lab employees signed a public July letter demanding a slowdown after the OpenAI-Hugging Face agent hack.
- 04The proposal, laid out in an Amodei essay, calls for third-party auditors from METR, Apollo Research, and Redwood Research to embed inside labs.
- 05President Trump called AI risk a 'hoax' on Monday and said the only guardrails needed are a 'STRONG AND SMART' president.
The pressure did not originate with the CEOs. More than 1,000 AI lab employees signed a public letter in July calling for a development slowdown after the OpenAI-Hugging Face agent incident, in which swarms of autonomous agents ran rogue hacks under the noses of frontier labs. The resignation letter posted by Anthropic researcher Jacob Coxon then broke containment: it has drawn more than 170 million views on X and put AI extinction warnings on the front pages of newspapers and on television. Coxon wrote that the people building AI 'earnestly believe that it could kill us all by the end of the decade' and accused both OpenAI and Anthropic of 'racing straight to self-improving superintelligence and gambling with our lives.'
Daniel Kokotajlo, a former OpenAI employee now leading the AI Futures Project, said years of external pressure — not executive conscience — produced the pact.
Reactions across the safety community are cautiously positive but heavily conditional. Apollo Research CEO Marius Hobbhahn called the plan 'a good idea' and said 'it's one of the best things for safety in a long time if it actually happens.' Tyler Johnston of The Midas Project called it 'a good sign.' Redwood Research CEO Buck Shlegeris called it 'some great news' while cautioning it is 'hard to know whether this is going to convert into anything real.' Amodei's essay named METR, Apollo, and Redwood as the kinds of external auditors that could embed inside labs and blow the whistle on problematic findings.
The safety-washing risk is the loudest counterweight. Kokotajlo warned that labs will 'just bring in some external auditors, do a bunch of safety paperwork — some of which will be genuinely good — but at the end of the day, it actually won't slow them down very much at all.' His AI Futures Project has proposed a harder version: auditors get access to each lab's compute budget, and the labs commit to measurable reductions in research compute, which would slow frontier progress and let trailing labs catch up.
Reese compared the maneuver to the social-media platforms' regulatory playbook a decade ago — get ahead of tougher laws by proposing your own. He said the 'hammer may not come in this administration' but suggested a future Democratic administration would likely bring real rules. Sacha Haworth, executive director of the Tech Oversight Project, was blunter, arguing voluntary frameworks amount to regulatory capture.
“we should not be letting the foxes run the henhouse. This is not an opportunity for Congress to once again outsource responsibility to industry.”— Sacha Haworth, Executive director of the Tech Oversight Project
Daniel Lobo-Lewis, co-founder of the Political Integrity Project, predicted the voluntary regime will end up as toothless as Meta's Oversight Board. The industry's counter-lobby is not idle either: President Trump posted on Monday that 'the only control or "guardrails" that AI needs is a STRONG AND SMART (High IQ!) PRESIDENT.' He then phoned Nvidia CEO Jensen Huang mid-conference and told the audience on speakerphone that recent AI concerns were a 'hoax' and that 'the robots will not be taking over.'
The single biggest structural problem for any slowdown remains China. Lab CEOs and US politicians have spent years arguing that American AI cannot decelerate because Chinese labs will not. One X user summarized the mood with a line that keeps circulating: 'If I am going to die at the hands of killer AI, I want it to be American, not Chinese.' Kokotajlo and others note that calls for a slowdown are aimed almost exclusively at large frontier labs defined by specific size metrics, which would leave open-source developers and smaller firms unaffected — but that carveout is not written into anything binding yet.
The pact reads less like a treaty than an opening bid. Every named lab has commercial incentives to define 'pace the frontier' loosely: Anthropic is preparing an IPO, OpenAI is scaling agent products, and xAI is chasing both. The safety community's asks — capped compute budgets, embedded auditors with whistleblower authority, measurable delay commitments — are the pieces that would turn a press cycle into a policy. Without them, the risk is that the industry writes its own soft law now to preempt harder law later, and the enforcement question gets kicked to a Congress that has already shown no appetite to take it up.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
