Anthropic chief executive Dario Amodei used a lengthy essay published Saturday to call on his own industry to deliberately slow down, warning that the pace of AI progress has outrun the field's ability to keep the technology under control.
In the essay, titled "We Must Pace the Frontier," Amodei argued that capability gains have gone exponential since roughly the middle of this year, driven largely by AI systems that are increasingly used to help design the next generation of AI models — a dynamic researchers call recursive self-improvement. "We must slow the pace at which we improve the capabilities of AI models," he wrote, framing the exponential curve — "one, then two, then four, then eight, then 16, then 32" — as a warning sign rather than cause for celebration.
Amodei's proposal has three parts: AI companies should give independent evaluators, such as the nonprofit research group METR, "ongoing, employee-like access" to verify safety practices from the inside; AI developers in democratic countries should agree on common safety standards and limits on the rate of unchecked progress; and, eventually, governments — including rival powers such as China — should negotiate narrower international agreements, for instance on barring AI assistance with bioweapons. Anthropic said it is unilaterally adopting the first step, giving outside evaluators office access comparable to its own internal risk teams.
Rivals sign on, but skeptics push back
The essay landed amid unusually public soul-searching in the industry. OpenAI's Sam Altman endorsed the idea of "pacing" and pledged to bring in independent evaluators of his own, while Tesla and xAI's Elon Musk offered a three-word response: "Dario is right." Amodei's warning also followed the public resignation of a former Anthropic researcher, Jacob Coxon, who said the industry was "gambling with our lives" and put the odds of an AI-driven human extinction within a decade at 10 percent — a claim Amodei's essay was widely read as responding to, arguing the resulting public alarm reflects "a crisis of trust" the industry needs to address head-on.
Not everyone was convinced. Nvidia chief executive Jensen Huang dismissed the call as self-interested, saying Amodei "believes that AI is so scary that only they should do it" and warning that "lobbying for regulatory capture against open source will only stifle innovation." Technology writer Brian Merchant argued that alarmist rhetoric about runaway AI distracts from harms the technology is already causing, saying he has yet to see "a credible, step-by-step documentation" of how AI could plausibly wipe out humanity. Coordinating a slowdown across competing companies also raises antitrust concerns; Amodei's essay acknowledges the industry would likely need a "narrow waiver" from U.S. regulators to hold the safety conversations he is proposing without running afoul of competition law.
No government has yet responded formally to the proposal, and nothing in the essay binds any company beyond Anthropic's own evaluator commitment. Whether the idea gains traction may hinge less on the safety argument than on whether rivals see a competitive incentive — or a regulatory shove — to slow down together rather than race ahead alone.
