Anthropic Gives METR Employee-Like Access to Audit Its Models

Anthropic said on October 15, 2026 that it has begun giving a team from METR, a nonprofit that evaluates frontier AI models, employee-like access to the company. According to Anthropic’s post, six METR employees received company hardware, office space and access to internal tools and systems roughly comparable to those of its own risk assessment teams.

What changed?

Anthropic described three tiers of evaluator access. In the first, Anthropic decides what evaluators can see and approves publication. In the second, evaluators can publish without approval and get more information. The third, which METR has, adds employee-like access with publication rights similar to the second tier. METR will publish its analysis of Anthropic’s risk levels, incidents or practices.

Anthropic says it cannot veto that publication, but it can require redaction of certain categories, such as customer data, privileged information and model architectures. It is also working with SecureBio, Accenture and Faculty, and says it is in early discussions with other organizations, including Hugging Face. SecureBio is reviewing the chemical and biological weapons chapter of Anthropic’s August 2026 Risk Report.

Why is Anthropic doing this?

Anthropic says models are more powerful and need stronger safeguards and harder risk assessments. It says it wants to be “publicly accountable” but cannot share all relevant safety information, because some could help malicious actors and some is commercially sensitive. It compares the model to bank supervision, where examiners sit inside large banks, and says “every frontier AI company should do the same.”

The company ties the step to its signing of the White House Accord on Super Intelligence, in which companies pledged to work with independent evaluators. The announcement also arrives as the FTC investigates leading AI labs, though Anthropic’s post does not link the two.

What has not been shown?

The arrangement is described by Anthropic itself, and METR’s findings have not yet been published. How much access evaluators really have, and how often redaction is used, will only be clear once reports appear.

Why it matters: For buyers of AI tools, independent evaluation is one of the few ways to test a vendor’s safety claims. Use it as one input in your AI vendor due diligence, alongside contract and data terms.