Anthropic gives METR researchers employee-level access to assess its models
Anthropic said in a blog post on Thursday that it has begun giving six researchers from METR, a nonprofit that evaluates frontier AI models, "employee-like" access — including company-issued hardware, office space and access to internal tools and systems roughly on par with its own risk assessment teams. METR can publish its own assessments of Anthropic's risks, incidents and practices; Anthropic cannot veto publication, but can require redaction of narrow pre-defined categories such as customer data, attorney-client privileged information and details about model architectures, and METR may publicly characterize those redactions. Anthropic said it is also working with SecureBio (currently at the first tier), Accenture and Faculty, is in early discussions with several other organizations including Hugging Face, and is building a three-tier access system in which METR is so far the only top-tier evaluator.
Embedding outside evaluators inside a frontier lab, if copied by peers, could become an industry norm for AI safety oversight.