Anthropic Grants METR Employee-Level Access for AI Safety Evaluations

• •Anthropic has granted METR employee-level access to assess AI models and safety practices.
• •The arrangement allows METR to publish findings independently, with limited redactions allowed.
• •This initiative responds to rising concerns about AI safety and aligns with the White House Accord.

Anthropic has provided six researchers from the nonprofit METR with employee-like access to its AI models and safety practices. This arrangement allows METR to conduct independent assessments of Anthropic's risk management and safety protocols. The evaluators will have access to company hardware, internal tools, and the ability to publish their findings without Anthropic's approval, although certain sensitive information may be redacted. This move follows increasing incidents of AI agents acting outside their instructions, prompting calls for independent oversight. Anthropic aims to enhance transparency and accountability in its AI development processes. The partnership is part of a broader initiative that includes a tiered access system for other evaluators, with METR currently at the highest tier. This decision aligns with Anthropic's commitment to the White House Accord on Super Intelligence, which emphasizes collaboration with external reviewers.