Six people from METR now have Anthropic hardware, office space and access to its internal tools and systems. The access is roughly on par with what Anthropic's own risk assessment teams have.
Fact
The facts: on 15 October 2026 Anthropic announced that it had started to give employee-like access to a team from METR, a nonprofit that evaluates frontier AI models. METR will publish the results of its risk assessments. Anthropic has onboarded six METR employees with company hardware, office space and access to internal tools and systems roughly on par with what its own risk assessment teams have. Outside a small number of clearly defined categories, such as data about customers, attorney-client privileged information and information about model architectures, Anthropic employees are encouraged to share information with them. Anthropic can't veto publication, but it can require METR to redact information if it falls into predefined categories; Anthropic ultimately decides what falls into them, while METR reserves the right to make certain statements about the level of access and information it received. Among the requirements for an evaluator, Anthropic lists that it should not hold equity in the company it evaluates or be financially or operationally dependent on it. Anthropic compares the setup to bank supervision, where resident examiners work inside the largest banks, and writes that it is embedding evaluators voluntarily and believes every frontier AI company should do the same.
We wrote earlier that the system would prove itself with the first report Anthropic doesn't like. Now the company has itself written the most important condition: it can't stop such a report.
It can't stop the report. It decides what gets redacted.
The catch is in the next sentence. Redaction applies to predefined categories like customer data, attorney-client privilege and model architectures, and Anthropic decides what falls into them. How much will be redacted, the text does not say. Who pays METR for this work is not stated either, and in September the talk was of pilots funded by METR's own money.
Anthropic describes three tiers of access. In the first, the evaluator runs narrowly scoped investigations and publishes with Anthropic's approval. In the second, it publishes without approval and sees more, though Anthropic can still require removal of specific categories. METR is in the third: employee-like access and publication rights similar to the second tier. SecureBio is reviewing the chemical and biological weapons chapter of the August Risk Report. With Accenture and Faculty, Anthropic is working toward an ambitious partnership, and it is in early discussions with several other organizations, including Hugging Face.
The page carries a note from 16 October saying the current terms of the project with METR were clarified, with no details. When METR will publish its first analysis, Anthropic does not say. Until then we have a description of the rules, not a report.
Official primary sources
Anthropic - An update on independent evaluators, 15.10.2026The visual is generated code-art, no third-party images.