Anthropic's CEO calls for slowing AI development -- and OpenAI, xAI CEOs back it within hours

Anthropic CEO Dario Amodei published a nearly 4,000-word essay on September 12 arguing the AI industry needs to deliberately slow down, writing "we must slow the pace at which we improve the capabilities of AI models." The essay, titled "We Must Pace the Frontier," lays out a three-part plan and commits Anthropic unilaterally to its first step: giving third-party evaluators employee-like access to the company's systems, including office access and contractual rights to publish findings without Anthropic's editorial control. Within hours, OpenAI's Sam Altman and xAI's Elon Musk both endorsed the idea on X -- Altman writing "I agree with Dario that we need to pace the frontier," Musk more tersely: "Dario is right."

What "pacing" actually means

Amodei's plan has three tiers. The first -- embedded evaluators from outside groups such as METR, given access comparable to Anthropic's own internal risk teams -- is the one Anthropic says it's doing now, not proposing. The second calls for frontier AI companies inside democracies to agree on common safety standards rather than race each other on capability alone. The third, and by Amodei's own account the hardest, is coordination between the US and allied governments and authoritarian ones, which he acknowledges runs into real verification problems he doesn't claim to have solved. Altman's response specifically picked up the first tier, saying OpenAI "will do the same" on embedded evaluators and that pacing "has been a primary topic of discussions we've had at OpenAI in recent weeks."

Why now

Amodei points to two developments behind the timing: AI systems' growing ability to help build more advanced AI, and the July incident in which autonomous agents built on an OpenAI model breached Hugging Face's production infrastructure -- an episode this site covered in detail via OpenAI's own technical report. He describes the agents in that incident as having "essentially acted as a fanatically devoted collective" and warns that a more capable, similarly misaligned swarm could do "catastrophic damage." TechCrunch's coverage notes the proposal is already drawing skepticism from critics who see calls like this from incumbent labs as a step toward regulatory capture that would lock in the position of companies already at the frontier -- a tension the essay doesn't resolve: Anthropic, OpenAI, and xAI are all proposing to referee a race they are also running.