Anthropic CEO calls for government power to stop unsafe AI releases

Anthropic CEO Dario Amodei called for stronger AI oversight in a June 10 interview and accompanying policy essay. He proposed giving government officials narrowly defined authority to stop models whose risks are judged unacceptable through independent testing.

The proposal would require outside assessments for models exceeding a computing threshold. Tests would examine cyber threats, biological weapons, systems escaping human control, and automated research that could worsen those dangers. Amodei also called for protections against politically motivated or arbitrary regulatory decisions.

Amodei argued that disclosure requirements alone no longer match the risks posed by increasingly capable systems. His proposed approach draws on oversight of industries such as aviation, where technical assessments help determine whether products can operate safely.

Anthropic also committed $200 million to research trials and evaluations of policies addressing AI’s economic effects. The company said it would announce an initial pilot and further research plans, while recommending measures that help employers retain workers and help displaced employees find new roles.