Anthropic grants METR researchers employee-level access

In a Thursday blog post, Anthropic announced it will give six researchers from the nonprofit METR employee-level access. They receive Anthropic-issued company laptops, office space, and access to internal tools roughly equivalent to those used by Anthropic's own risk assessment team. Anthropic encourages employees to share information with them as they would with colleagues working on alignment risk. METR will publish its own evaluations of Anthropic's risks, incidents, and practices; Anthropic has waived its right to block publication, can only request redaction in narrow predefined categories such as customer data, attorney-client privileged information, or model architecture, and METR may publicly judge whether any redaction is justified. This is the top tier of a three-layer evaluation system Anthropic is building. Lower-tier evaluators like biosecurity nonprofit SecureBio get limited review and need Anthropic's approval to publish; the middle tier gets more information and can publish without approval; only the top tier includes METR's employee-level access. The move fulfills CEO Dario Amodei's September pledge to embed external evaluators inside AI labs, a commitment OpenAI later matched. For developers and researchers tracking frontier model safety practices, it means METR can now publish deeper independent evaluations grounded in internal information.