Anthropic's Amodei Calls for Slower AI Race, Rivals Agree

Dario Amodei just asked the AI industry to hit the brakes, and the loudest voices in the room said yes. In an essay titled "We Must Pace the Frontier," the Anthropic CEO laid out a three-step plan to slow frontier AI development, and within a day, Sam Altman, Demis Hassabis and Elon Musk had all publicly backed it.

Amodei's argument rests on two triggers: the growing capacity of AI systems to improve themselves without human input, and a recent security breach involving OpenAI and Hugging Face. Both, he wrote, expose how far ahead capability has moved relative to oversight.

Three steps, one goal

The plan starts with embedding independent evaluators, drawn from groups like METR, directly inside AI labs. Anthropic said it will unilaterally give these outside monitors company badges, desks, laptops and access on par with its own internal risk teams.

The second step pushes AI companies in democratic countries to agree on shared safety standards and limits on unchecked capability gains. The third extends that effort outward: engaging China and Russia on global safety norms, while preserving a US and allied technological edge through chip export controls and restrictions on model distillation.

Rivals sign on, fast

Altman called embedded evaluators "a valuable idea," said existing safety standards fall short, and committed OpenAI to giving outside evaluators employee-level access. Hassabis wrote that "Dario's essay points towards the right path forward." Musk, posting as both xAI founder and SpaceX CEO, simply said: "Dario is right."

"Dario is right." - Elon Musk

Hugging Face CEO Clement Delangue announced a new initiative to support independent evaluation efforts, adding a fourth major player to the coalition forming around Amodei's framework.

Not everyone reads it the same way

Not all reactions treated the proposal as pure altruism. Investor Chamath Palihapitiya suggested the pacing plan could double as a way for Anthropic, which along with OpenAI is reportedly preparing for a future public offering, to lock in a competitive advantage over rivals moving faster.

Amodei addressed the geopolitical tension directly, insisting that any slowdown must not cede ground to China, and calling for continued export restrictions on advanced AI chips to authoritarian states.

The essay landed days after former Anthropic researcher Jacob Coxon resigned, telling media outlets that leading labs are "gambling with our lives" and that many people inside these companies privately believe the technology could pose an extinction-level risk within two years, some say by the end of this decade.

Whether badges and desk access for outside evaluators translate into genuine constraints on model capability, rather than a coordinated slowdown among a handful of dominant labs, is the question the industry now has to answer in practice, not in essays.