Anthropic gives METR employee-like access to assess AI risks

Anthropic says it has begun giving a team from METR, a nonprofit that evaluates frontier AI models, “employee-like access” to parts of its company so the outside group can assess model risks and safety practices. The announcement follows Anthropic's pledge to strengthen independent, third-party evaluation of its systems. ^{1 2}

What METR can access

Anthropic says six METR employees have been onboarded with company hardware, office space and access to internal tools and systems roughly comparable to those used by Anthropic's own risk-assessment teams. The arrangement excludes defined categories such as customer data, attorney-client privileged information and some model-architecture details. ¹

The company says METR will be able to publish its analysis of Anthropic's risk levels, incidents or practices. Anthropic cannot veto publication, although it can require redactions for information that falls into the agreed restricted categories. METR retains the ability to discuss the level of access and information it received. ¹

Why the arrangement matters

Anthropic argues that external evaluators need more than access to a model's public interface. They also need visibility into training, safeguards and the way a company assesses risk internally. The company compares the arrangement with bank supervision, where resident examiners work inside major banks because the institutions know more about their own risk positions than outside observers. ¹

The move is also a response to a practical tension in AI safety reporting. Companies cannot publish every detail about their systems without potentially exposing commercially sensitive information or helping malicious actors, but fully internal safety claims are difficult for the public to scrutinise. Anthropic says independent evaluators can see more while publishing evidence and reasoned conclusions. ^{1 2}

METR is not the only evaluator involved. Anthropic says it is also working with SecureBio and Accenture, while discussing possible future work with other organisations. It describes the METR arrangement as its third tier of evaluator access, with deeper access and publication rights than its earlier-stage partnerships. ¹

Our opinion

The useful part of this announcement is not the corporate analogy; it is the publication right. Giving an outside group hardware, office access and internal tools is meaningful only if the evaluator can report uncomfortable findings without needing the company to approve the conclusion. Anthropic is still defining the boundaries, and it retains redaction powers, but a named evaluator with a public reporting remit is a more testable safety commitment than another promise to improve internal processes. ^{1 2}