Anthropic CEO Dario Amodei Outlines Three-Part AI Pacing Plan
Original: Anthropic CEO outlines plan to ‘pace the frontier’
Why This Matters
A major frontier lab publicly committing to third-party embedded oversight sets a concrete precedent for AI accountability.
Anthropic CEO Dario Amodei published a blog post on September 12, 2026, calling to slow AI capability progress and outlining three strategies. Anthropic is unilaterally committing to one: embedding third-party evaluators inside the company with near-full internal access to verify safety commitments.
Amodei's post arrives amid a charged week for AI safety discourse. Researcher Jacob Coxon publicly resigned from Anthropic, writing that leading AI firms are 'gambling with our lives' and that builders 'earnestly believe it could kill us all by the end of the decade' — a claim others at Anthropic have also made. Amodei did not address Coxon by name, but cited two triggers for his shift toward caution: the OpenAI-HuggingFace hack, and AI's accelerating ability to design the next generation of AI systems. 'We must slow the pace at which we improve the capabilities of AI models,' he wrote. 'Progress will still seem fast, and we must make wise use of the time we gain.' His first proposal — and the one Anthropic is immediately adopting — is embedding evaluators from third-party organizations such as METR inside AI companies. These evaluators would receive company badges, desks, laptops, and access 'mostly comparable to what internal risk assessment teams have.' Amodei compared the model to regulators embedded within banks. His second proposal calls for leading AI companies in democratic nations to coordinate on common safety standards and limits on unchecked capability growth. He acknowledged antitrust obstacles, asking the U.S. government to issue a narrow waiver enabling those conversations. OpenAI CEO Sam Altman has also recently floated the idea of pacing development.