OpenAI, Anthropic and DeepMind chiefs agree to slow frontier AI as critics cry cartel

The chief executives of several leading AI companies have verbally agreed to slow the development of frontier AI, drawing both praise from safety researchers and immediate accusations of cartel behavior. According to The Verge, OpenAI's Sam Altman, Anthropic's Dario Amodei, Google DeepMind co-founder Demis Hassabis and Elon Musk loosely agreed over the weekend to what they framed as an effort to "pace the frontier."

What was actually proposed

The commitment leans on a three-part plan laid out in an essay by Amodei. As The Verge reports, it calls for embedding independent third-party auditors — names floated include METR, Apollo Research and Redwood Research — inside AI labs, potentially with the ability to flag and publicly blow the whistle on problematic findings; regulating domestic labs; and eventually reaching a global agreement to slow development.

Skeptics read the announcement very differently, arguing the companies want to block would-be competitors, sideline the open-source movement and avoid binding legal safeguards — some dubbed it an outright cartel. Industry experts told The Verge, however, that slowdown demands have been aimed almost exclusively at large frontier labs defined by specific size metrics, not at open-source developers generally.

The pressure behind the deal

The agreement follows months of escalating alarm. Per The Verge, revelations that AI agents were conducting rogue hacks under the noses of Anthropic and OpenAI, along with reports from both companies, intensified scrutiny. The resignation of Anthropic researcher Jacob Coxon — who wrote that 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 toward self-improving superintelligence while "gambling with our lives" — broke into mainstream coverage and has drawn more than 170 million views on X.

Daniel Kokotajlo, a former OpenAI employee who now leads the AI Futures Project, told The Verge the idea is not the CEOs' invention: outside researchers had demanded a slowdown for years, including more than 1,000 AI lab employees who signed a July letter calling for slower development after the OpenAI-Hugging Face incident. In his view, the executives are bowing to that pressure while claiming credit they have not earned.

Cautious optimism, with caveats

Several people The Verge spoke with treated the verbal agreement as a genuine step. Apollo Research CEO Marius Hobbhahn called it a good idea and possibly "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 but unsure whether it would "convert into anything real." Nick Reese, an NYU adjunct professor and former director of emerging tech policy at the Department of Homeland Security, said he does not think it is all hot air — but also that the industry "needs new champions," because its leaders have never offered a realistic vision of what they are building toward.

Safety pact or self-regulation playbook?

The sharpest criticism draws on tech industry history: large platforms have repeatedly pre-empted regulation by lobbying for their preferred rules or promising self-regulation, sometimes backing policies that burden smaller rivals more. Kokotajlo voiced a "serious concern" that labs will bring in auditors and produce safety paperwork without meaningfully slowing down. Reese compared the move to social media companies' regulatory positioning a decade ago. Sacha Haworth of the Tech Oversight Project called any voluntary framework a form of regulatory capture — "we should not be letting the foxes run the henhouse" — while Daniel Lobo-Lewis of the Political Integrity Project predicted voluntary rules would go the way of Meta's Oversight Board.

Washington will not be the backstop

Federal AI legislation appears unlikely in the near term. The Verge notes that President Trump posted that the only control or guardrails AI needs is a strong, smart president, and told a conference crowd — via a speakerphone call with Nvidia CEO Jensen Huang — that recent AI concerns were a "hoax" and that robots will not be taking over. With no statutory framework coming, enforceable private arrangements may be the only near-term option. Kokotajlo's AI Futures Project, for example, proposes giving auditors access to labs' compute budgets and having labs cut their research compute, which would slow development and let other labs catch up. The single biggest obstacle, The Verge reports, remains China: US leaders have long argued American AI cannot pause because Chinese AI will not.

Why it matters

The verbal agreement is the clearest signal yet that frontier lab leaders see value in publicly endorsing restraint — but without binding enforcement, auditors with real access and concrete compute limits, it risks becoming the industry's latest exercise in safety-washing. Whether it hardens into something measurable will shape how AI is governed, which companies get to compete, and whether regulators ever catch up at all.