Anthropic CEO Amodei calls for pacing the AI frontier

In a new essay and CNN/CBS interviews, Dario Amodei urged slower capability growth so safety can keep up, starting with embedded third-party evaluators Anthropic says it will adopt now.

Anthropic CEO Dario Amodei published “We Must Pace the Frontier” arguing that frontier labs must slow the rate of capability gains so safety work can keep up, then amplified the case in exclusive CNN interviews with Anderson Cooper and a CBS News sit-down. The package is distinct from earlier Altman comments about being open to slowing: this is Amodei’s own three-step plan plus the interview tour.

In the essay, Amodei says pacing is not a halt to training. Companies should take enough time to align and safeguard models and let third parties verify that work. He cites accelerating recursive self-improvement since roughly this summer and the OpenAI-Hugging Face agent swarm incident as reasons safety can fall behind. His plan: (1) embedded evaluators with ongoing employee-like access to check practices, report incidents, and assess training pipelines, not only finished models; (2) coordination among frontier firms in democratic countries on common safety standards and limits on unchecked progress; (3) global coordination with authoritarian governments where verifiable. Anthropic is unilaterally committing to embedded evaluators now and wants governments to require peers to match.

On CBS, Amodei said, “for too long the industry lied” about AI risks, framed recent agent misbehavior as a warning sign to slow down, and called the US-China race “the toughest dilemma.” He said he is uncomfortable with frontier AI sitting only in private hands and preferred joint oversight by democratically elected governments over handing the technology to a single government. On CNN, Cooper pressed existential-risk questions in the context of Anthropic researcher Jacob Coxon’s resignation; Amodei said the technology “belongs to all of us.”

In the first hours after the essay, OpenAI CEO Sam Altman publicly backed pacing and said OpenAI would also commit to independent evaluators with employee-like access, The Atlantic reported; Elon Musk also signaled agreement. This brief covers Amodei’s plan and the interview package, not a full recap of the Christiano board hire or prior Altman Top.

Anthropic CEO Amodei calls for pacing the AI frontier

Sources