AI's biggest bosses agreed to slow down. Almost nobody believes them.

Key points

• Sam Altman (OpenAI), Dario Amodei (Anthropic), Demis Hassabis (Google DeepMind) and Elon Musk (SpaceX) verbally agreed over the weekend to slow down frontier AI development.
• Amodei's written proposal calls for third-party auditors embedded inside AI labs, domestic regulation for large AI companies, and a global slowdown agreement.
• An Anthropic researcher's public resignation letter, viewed more than 170 million times on X, helped push the issue into mainstream view.
• Critics say the proposal could benefit large labs at the expense of smaller rivals and open-source developers, who build AI tools anyone can freely use and modify.
• No binding deal exists yet, and the Trump administration has shown no appetite for AI regulation.

On paper, it looks like a turning point. Four of the most powerful people in artificial intelligence sat down and agreed, at least loosely, to stop racing quite so fast. In practice, almost nobody's sure what that means.

Sam Altman of OpenAI, Dario Amodei of Anthropic, Demis Hassabis of Google DeepMind and Elon Musk of SpaceX reached a verbal agreement over the weekend to "pace the frontier," meaning they'd deliberately slow the pace of developing the most powerful AI systems. Amodei put the plan in writing: bring outside auditors into AI labs, create domestic rules for large AI companies, push for a global agreement to slow development across the board. As we reported the day Amodei published his proposal, he's also offering to let independent evaluators into Anthropic itself.

So why aren't people celebrating?

Because a verbal agreement between competitors who all stand to benefit from fewer rivals isn't the same as a law or a binding standard.

Critics called it a cartel move almost immediately. The concern, reported first by The Verge, is straightforward: rules targeting only the biggest AI labs could squeeze smaller startups and open-source projects, which let developers share and build on AI models freely, while protecting the companies already at the top. Calls for slowdowns have been aimed almost exclusively at large frontier labs defined by exact size metrics, multiple industry experts told The Verge, which blunts that criticism somewhat.

Daniel Kokotajlo, a former OpenAI employee who now leads the AI Futures Project, a nonprofit focused on AI research, said the CEOs are "bowing to that pressure and also claiming credit for it, not rightfully." He's been watching this pressure build since well before this weekend.

More than 1,000 AI lab employees signed a public letter in July 2025 calling for slower development, following an incident involving OpenAI and Hugging Face, a platform for sharing AI tools.

What actually pushed this into public view?

A resignation letter. Jacob Coxon, a researcher at Anthropic, quit and published a public explanation. He 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." The letter has been viewed more than 170 million times on X, and Coxon's story has since appeared in newspapers and on television.

That kind of number pulls a niche argument onto the front page.

What do independent experts actually think?

Guardedly positive, but nowhere near convinced.

Apollo Research CEO Marius Hobbhahn called the proposal "one of the best things for safety in a long time if it actually happens." Redwood Research CEO Buck Shlegeris said he was "cautiously optimistic." Nick Reese, a former director of emerging tech policy at the US Department of Homeland Security and now a professor at New York University, said he doesn't think it's "all hot air." Every one of them added that a lot of work remains to make this real and enforceable.

Kokotajlo's concrete suggestion is giving outside auditors access to the companies' compute budgets, the raw spending on the chips that train AI. Double your chip spend quietly and an auditor would see it.

The biggest obstacle, though, is a single word: China. AI leaders and politicians routinely argue that slowing US development simply hands an advantage to Chinese labs. One widely shared post on X put it bluntly: "If I am going to die at the hands of killer AI, I want it to be American, not Chinese."

President Trump's position is clearer still. He posted Monday that the only guardrails AI needs are "a STRONG AND SMART (High IQ!) PRESIDENT," called recent safety concerns a "hoax," and said "the robots will not be taking over."

With no government action coming soon, a voluntary deal among competitors is what safety advocates have to work with. The people who built the problem volunteering to fix it, on their own terms, with no enforcement mechanism: that's the part worth watching most closely.