Anthropic and OpenAI leaders call to slow frontier AI race

This is the clearest sign yet that the frontier labs themselves feel they are getting uncomfortably close to capabilities they cannot reliably control. Amodei’s essay does not call for a moratorium, but it does propose a concrete framework for pacing frontier progress and, critically, invites embedded external evaluators into Anthropic’s operations. That is a big cultural step for a high growth AI lab and, if copied, would move the industry closer to something like bank style supervision rather than self attestation.

Altman and Musk lining up behind Amodei changes the political dynamics. For the past two years, calls to slow AI have mostly come from academics, safety advocates and a subset of employees. Now three of the most powerful commercial actors are explicitly saying that capabilities are advancing too fast relative to safety and that some form of coordinated slowdown is necessary. That will embolden regulators and give cover to more cautious boards and investors. At the same time, none of these firms is promising to stop training new models, so there is a real risk that “pacing the frontier” becomes a reputational shield while the underlying race continues.

For the race to AGI, the key question is whether this moment produces hard constraints on compute, model deployment and self improvement, or just more safety teams inside the same acceleration curve. If embedded evaluators gain real access and bite, they could meaningfully slow deployment of the most dangerous systems. If not, this may simply mark the point when labs started trying to manage public expectations about how far ahead they already are.