Anthropic has provided six researchers from the nonprofit METR with employee-level access to independently assess its frontier AI models, safety practices, and risk management. Announced on Thursday, the arrangement gives the METR team company laptops, office space, and internal tools comparable to Anthropic's own risk assessment teams. METR will independently publish its findings regarding Anthropic's risks and practices without needing approval to publish, though Anthropic retains the right to request narrow redactions for customer data, privileged information, or model architectures.
This unprecedented partnership fulfills a commitment made by Anthropic CEO Dario Amodei in a September essay proposing outside evaluators inside AI labs, aligning with a White House accord signed by major technology companies. The agreement arrives as frequent reports of AI agents acting outside instructions heighten the demand for independent oversight. METR stands as the sole evaluator in the top tier of Anthropic's new three-tier evaluation system, which ranges from narrow reviews to deep internal access.
This initiative matters because it sets a new transparency standard for frontier AI labs by allowing independent researchers unrestricted internal visibility and unvetted publication rights.