Amodei calls for a slower AI race — and rivals sign up fast

Anthropic just set a competitive benchmark rivals cannot easily ignore: safety transparency. On 12 September, CEO Dario Amodei published an essay, "We Must Pace the Frontier", arguing the industry should slow capability gains by one to two years relative to today's trajectory, and committing Anthropic unilaterally to let third-party evaluators inside its systems with permanent, employee-level access.

The roughly 3,800-word essay lays out a three-part plan: independent evaluators embedded inside AI labs, targeted US regulation covering every frontier company rather than volunteers alone, and eventually global coordination, including with China, on the highest-stakes risks. Amodei ties the urgency to recursive self-improvement — AI accelerating AI research itself — and to this summer's agent incidents.

Anthropic is not waiting for regulation or rivals to move on step one. On X, Amodei wrote: "We'll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models' alignment during training." He separately pressed for chip export controls on China to stay in place and called for a ban on distillation of US models.

The market reaction was immediate. Within hours, OpenAI's Sam Altman wrote "I agree with Dario that we need to pace the frontier" and said OpenAI will match the evaluator-access commitment. Musk, DeepMind's Demis Hassabis and Microsoft's Satya Nadella all publicly backed the first step too. Tech Policy Press pushed back, questioning whether a lab CEO should be the one setting the industry's pace.

The essay follows OpenAI chief scientist Jakub Pachocki's 6 September piece hoping for voluntary slowdowns, and builds on the July "Pacing the Frontier" letter that Amodei and more than 1,100 AI workers signed.

Amodei has converted a safety argument into a standard his competitors now have to meet in public. By committing Anthropic first and naming the exact access it will grant, he has made refusal look like evasion — which is why Altman, Nadella and Hassabis fell in line within hours rather than days. That is a cheap win for Anthropic: it costs transparency, not capability, and it buys reputational cover while regulation is still just "targeted" and hypothetical.

The harder test is step two. Voluntary evaluator access is easy to match; binding US rules that apply to every frontier lab, plus coordination with China, are not something any CEO can commit to alone. Watch whether Anthropic's rivals' enthusiasm survives contact with actual legislation, and whether Amodei's distillation ban and export-control stance — clearly aimed at Chinese competitors — draws pushback from labs with more exposure to that market.