Anthropic chief executive discusses model evaluation and development pacing in broadcast interview

Anthropic Chief Executive Dario Amodei highlighted new proposals for oversight during a broadcast interview. Following a recent essay on technological risk, Amodei called for third party monitors to be embedded directly within artificial intelligence research centers. These independent auditors would inspect training runs as they occur, ensuring safety protocols are strictly followed before powerful models are deployed to the public.

As top technology firms race to build increasingly capable models, public debate has focused on how to prevent unintended harms. Allowing outside experts direct physical access to labs could transform industry oversight from self reporting into verifiable compliance. If leading companies adopt this approach, it may set a new standard for transparency and help governments build realistic regulatory guardrails without guessing what happens behind closed doors.

However, major practical questions remain about how such an auditing regime would function in practice. It is still unclear who would certify these external auditors, how trade secrets would be protected, and whether competing firms would ever agree on universal safety standards. Furthermore, placing observers inside private facilities might slow down rapid iteration cycles, prompting concerns that strictly governed labs could fall behind less regulated global rivals.