Anthropic pushes for supervised AI development with embedded external watchdogs

Anthropic CEO Dario Amodei published a detailed blueprint for decelerating AI development, responding to mounting safety concerns across the industry. The proposal centers on embedding external evaluators—vetted auditors from organizations like METR—inside AI companies to monitor compliance with safety standards and ensure incident reporting. Amodei said Anthropic is unilaterally committing to this approach, and OpenAI's Sam Altman confirmed his company will follow.

Amodei's strategy includes two additional pillars: coordination among leading AI firms on common safety standards, and global agreements limiting uncontrolled progress. He acknowledged antitrust risks but urged the U.S. government to broker these discussions. The proposal arrives amid heightened tension over AI safety—Anthropic researcher Jacob Coxon recently resigned over concerns the industry is recklessly advancing toward existential risk.