Anthropic gives METR employee-like access to assess AI risks
🎧 Listen to this story
Anthropic has onboarded six employees from METR, an independent risk-evaluation organization, giving them company hardware, office space and access to internal tools and systems roughly on par with what its own risk-assessment teams can use. The company says employees have been encouraged to share information with the evaluators as they would with colleagues in risk. Anthropic is not paying for the arrangement, and METR says it does not accept donations from AI companies or their employees.
METR will be able to publish its analyses without Anthropic’s approval or veto. The company can, however, require the removal of customer data, attorney-client privileged information and details about model architectures. Anthropic also said it is continuing to work with Accenture and SecureBio on safety evaluations.
Why it matters · editorial interpretation
The arrangement combines broad internal access with METR’s ability to publish analyses without Anthropic’s approval, while retaining limits to protect customer data, attorney-client privileged information, and architecture details. This design could enable more informed evaluations without removing restrictions on what can be disclosed.
Sources
Security