Anthropic gives METR inside access to audit its AI models and safety practices

Anthropic has given six METR staffers company hardware, office space, and internal tool access, which the company says puts them roughly on par with its own risk teams.

METR will publish its findings independently, though Anthropic can request redactions in predefined categories. The deal goes further than the access offered to SecureBio, Accenture, and Faculty, and Anthropic has begun early talks with Hugging Face.