Anthropic has embedded six researchers from the nonprofit METR inside the company to independently examine its models, incidents, safety practices and risk management. The evaluators received company laptops, office space and access to internal tools that Anthropic describes as roughly comparable to what its own risk teams use.
METR will publish its findings independently, and Anthropic cannot veto publication. The company can require redactions within narrow predefined categories, including customer data, attorney-client privileged material and model-architecture details. METR may publicly say whether it believes a requested redaction was justified, provided it does not reveal the protected information.
The arrangement goes beyond the access Anthropic has provided to other outside reviewers such as SecureBio, Accenture and Faculty. Anthropic compares the model to resident bank examiners, who work inside an institution while retaining an oversight role. It follows the company’s commitment under a White House accord to support external review.
Employee-level access can let evaluators inspect evidence that public benchmark testing cannot reach, but the review is still hosted and scoped by Anthropic. Its credibility will depend on the breadth of METR’s eventual reports, the frequency and justification of redactions, and whether findings lead to concrete changes. No assessment results were released with the announcement.