Anthropic's CEO calls for slowing AI development, and OpenAI and 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: "we must slow the pace at which we improve the capabilities of AI models." The essay, "We Must Pace the Frontier," lays out a three-part plan and commits Anthropic unilaterally to the 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 wrote "I agree with Dario that we need to pace the frontier." Musk wrote "Dario is right."

The three tiers

The first tier, embedded evaluators from outside groups such as METR with access comparable to Anthropic's internal risk teams, is the one Anthropic says it is already doing. The second calls for frontier AI companies in democracies to agree on common safety standards so they stop racing on capability alone. The third, by Amodei's account the hardest, is coordination between the US and allied governments and authoritarian ones. He acknowledges it runs into verification problems he hasn't solved. Altman 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: 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, covered here via OpenAI's technical report. He says the agents in that incident "essentially acted as a fanatically devoted collective" and warns that a more capable, similarly misaligned swarm could do "catastrophic damage." TechCrunch notes critics already see calls like this from incumbent labs as a step toward regulatory capture that would lock in the companies at the frontier. The essay doesn't resolve that: Anthropic, OpenAI and xAI are proposing to referee a race they are also running.