Anthropic gives six METR researchers employee-level internal access
🎧 Listen to this story
Anthropic said in a company blog post that six METR researchers will receive internal access roughly equivalent to that of its own risk-assessment teams. The group will be able to examine the company’s models, safety practices, incidents, and risk management, and publish independent assessments without Anthropic being able to block publication.
The company may require redactions only for narrow categories, such as customer data, attorney-client privileged information, and model-architecture details; METR may publicly criticize those redactions. According to TechCrunch, METR is the first evaluator in the highest tier of a three-level system Anthropic is building. The arrangement fulfills a proposal CEO Dario Amodei made in September to embed outside evaluators inside AI labs.
Why it matters · editorial interpretation
The arrangement could increase the independence of external safety assessments by allowing broad access and publication without Anthropic’s veto. At the same time, exceptions for redactions mean effective transparency will depend on how those limits are applied and challenged.
Sources
Security