Amodei calls for pacing frontier AI

On 12 September 2026 Anthropic's chief executive Dario Amodei published the essay "We Must Pace the Frontier": the rate at which model capabilities grow must be slowed so that risk prevention can keep up. As reasons he named recursive self-improvement, which since the summer had begun across the industry, including at Anthropic, and the OpenAI-Hugging Face incident. The first step of his plan, outside evaluators with employee-like access, Anthropic took on unilaterally.

The plan has three steps. The first is embedded outside evaluators (the essay gives METR as an example) with desks, badges and access close to that of internal risk teams, and the right to publish findings without Anthropic's editorial control. The second is coordination among companies in democratic countries on common standards and limits on pace, which needs government mediation or an antitrust waiver. The third is global coordination at four levels of agreement: a ban on narrow dangerous uses, testing of models before release, a speed limit on recursive self-improvement, and a full pause the author thinks unlikely. The Accenture partnership is led by its Faculty unit; both companies expect to invest at least a billion dollars each over five years. What the record does not claim: that other labs joined the plan, for the session read such responses only in secondary accounts, or that the pace of development actually slowed after the essay.