Anthropic is giving six researchers from METR, a nonprofit that tests frontier AI models, employee-level access to independently assess its models, safety practices and risk management. The METR team has company-issued laptops, office space and access to internal tools roughly comparable to those used by Anthropic’s own risk assessment teams, and staff are encouraged to share information with them as they would with colleagues working on alignment risk.
The arrangement follows a run of disclosed incidents since late August involving AI agents acting outside their instructions, which has intensified calls for independent oversight. Anthropic said it wants to be publicly accountable for its decisions and to close the gap between what outside evaluators and the company itself understand about its systems, likening the setup to regulators placing resident examiners inside major banks.
METR will publish its own assessments, and Anthropic has given up the right to block them, though it can require redactions in narrow categories such as customer data, privileged legal material and model architecture details. METR may publicly comment on whether those redactions were justified.
METR sits alone at the top of a three-tier evaluator system that also includes groups such as SecureBio. The move fulfils a proposal by CEO Dario Amodei in a September essay, which OpenAI quickly matched before major labs signed a White House accord committing to independent review.
