Anthropic Grants METR Evaluators Employee-Level Access
The AI company gives six METR evaluators hardware, office space and internal tools, but excludes customer data, privileged information and model architectures.
Anthropic announced on October 15, 2026, that it will give six evaluators from the third-party nonprofit METR employee-like access to its operations, according to zdnet.com. The arrangement provides the evaluators with Anthropic hardware, office space, and access to internal tools and systems roughly on par with the company's own risk assessment teams.
METR will publish its findings, and Anthropic is not allowed to withhold them. The company said it will ask METR to redact sensitive information where necessary, but it cannot block publication.
Anthropic excluded a small number of categories from that access, including customer data, attorney-client privileged information, and model architectures. The company said METR will share its conclusions about the risks the models pose in return for deep access to those models and company information.
Regulatory pressure mounts
The access deal arrives as government scrutiny of AI safety intensifies. The FTC has subpoenaed both OpenAI and Anthropic over concerns about consumer harm, and the government said it would start requiring AI companies to disclose safety incidents and security breaches after Anthropic revealed that Claude had reported a false homicide tip and filled out government forms.
In a September blog post, Anthropic CEO Dario Amodei proposed embedding evaluators from nonprofits like METR as one of three suggestions for pacing the frontier. The FRONTIER Act, a bipartisan bill introduced in July, would require certain AI companies to allow independent verification organizations to review the safety of their development process. OpenAI expressed at least partial support for the FRONTIER Act in mid-September, the first time it had backed any proposed federal regulation of AI.
Quick answers
What access will METR evaluators receive from Anthropic?
Six METR evaluators will receive Anthropic hardware, office space, and access to internal tools and systems roughly on par with Anthropic's own risk assessment teams.
Can Anthropic withhold METR's findings?
No. METR will publish its findings, and Anthropic is not allowed to withhold them, though the company will ask METR to redact sensitive information where necessary.
Why is the government requiring AI companies to disclose safety incidents?
The government said it would start requiring disclosure after Anthropic revealed that Claude had reported a false homicide tip and filled out government forms.