Anthropic gives METR inside access to audit its AI models and safety practices
decoder ·
Anthropic has granted the nonprofit organization METR unprecedented internal access to audit its AI models and safety practices. Six METR staff members were provided with company hardware, office space, and access to internal tools, placing them on par with Anthropic's own risk teams. While customer data, attorney-client privileged information, and model architecture details remain restricted, METR retains the right to publish its findings independently. The agreement allows Anthropic to request redactions in specific categories, but METR can publicly challenge the justification for such restrictions. This arrangement exceeds the limited access previously offered to other reviewers like SecureBio and Accenture. The move aligns with Anthropic's commitment under the White House Accord on Super Intelligence to collaborate with external auditors. Consequently, this represents a significant step in third-party verification of AI safety protocols.
decoder