Anthropic granted six researchers from METR, a nonprofit that tests frontier AI models, employee-level access to its internal systems, safety practices and risk management tools. The company announced the arrangement Thursday, Oct. 15, 2026.
The move reflects intensifying pressure for independent oversight. Reports of AI agents exceeding their programmed instructions have surfaced weekly or bi-weekly since late August, prompting calls for external review of AI development practices.
Under the agreement, METR researchers receive company laptops, dedicated office space and access to internal tools equivalent to what Anthropic's own risk teams use. Anthropic instructs employees to share information with METR researchers as they would with internal alignment specialists.
METR will publish independent assessments of Anthropic's risks, incidents and practices without company approval. Anthropic can only require redactions for customer data, attorney-client privileged information or proprietary model details. METR retains the right to publicly characterize any redactions and challenge their justification.
The arrangement sits atop a three-tier external review system Anthropic is building. Lower-tier evaluators like SecureBio conduct narrow reviews and need company approval to publish. The middle tier gains greater information access and publication freedom. METR occupies the top tier with embedded researcher status.
Anthropic CEO Dario Amodei proposed embedding outside evaluators within AI labs in a September essay on managing technology development. OpenAI has made a similar commitment. Both companies, along with Google, Meta, Nvidia and xAI, signed a White House accord requiring independent review of their AI models.
