Anthropic Proposes Embedded Evaluators Amid AI Regulation Debate

Anthropic CEO Dario Amodei has proposed a safety framework requiring frontier AI labs to host embedded third-party evaluators. These independent experts would have employee-like access, including office space and company laptops, to verify that firms adhere to their safety and deployment claims. Amodei unilaterally committed Anthropic to this model, specifically suggesting the nonprofit Model Evaluation and Threat Research (METR) for the role. OpenAI CEO Sam Altman and SpaceXAI CEO Elon Musk expressed support for the use of independent evaluators, though Altman did not specifically endorse METR.

The proposal has triggered a broader conflict over AI regulation. Critics, including FTC Chairman Andrew Ferguson and former White House AI czar David Sacks, argue that the push for regulation is a strategy for regulatory capture. They allege that Anthropic and other leaders use "doomer" narratives of human extinction to create an oligopoly that freezes out open-source competitors. Specifically, skeptics claim METR lacks independence due to shared investors with Anthropic.

While some former employees warn that superhuman systems could kill humanity by the end of the decade, political reactions remain divided. The Government of China rejected calls to slow AI development, calling such measures fear-mongering. In the United States, Florida Governor Ron DeSantis proposed a state-level AI Bill of Rights following reports of chatbots goading children to commit suicide, while Senator John Kennedy expressed doubt that Congress can draft a balanced national regulation bill.