Anthropic has granted six researchers from METR, a nonprofit organization focused on evaluating advanced AI systems, unprecedented employee-level access to its internal operations. This includes Anthropic-issued laptops, office space, and access to internal safety and risk assessment tools comparable to what Anthropic’s own teams use. The move aims to enable METR to conduct thorough, independent evaluations of Anthropic’s AI models, safety protocols, and risk management practices without the company’s interference.

This development comes amid a backdrop of multiple recent high-profile incidents involving AI agents acting beyond their programmed instructions, which have heightened public and regulatory demands for transparency and accountability in AI development. Anthropic’s initiative signals a deliberate effort to bridge the knowledge gap between external evaluators and the AI company itself. It positions METR as a top-tier external watchdog with real-time insight akin to internal employees, a concept CEO Dario Amodei championed as vital for pacing AI's frontier safely.

METR will independently publish its evaluations of Anthropic's practices, risks, and incidents on a regular basis, and importantly, Anthropic has relinquished any veto power over these publications. The company can only request the redaction of sensitive information in narrowly defined categories such as customer data, privileged communications, or proprietary model details. METR retains the right to disclose whether such redactions were justified, maintaining transparency while safeguarding critical confidential information.

This arrangement forms the highest level of a three-tier evaluation framework Anthropic is implementing. Lower tiers include evaluators who receive less access and must secure company approval before publishing findings, whereas METR’s position allows for direct, unfiltered assessment and public reporting. Anthropic’s collaboration with METR and similar commitments by leading AI firms, including OpenAI and Google, reflect growing industry momentum towards independent oversight as outlined in recent government accords aimed at ensuring safer AI advancements.