Anthropic Grants METR Staff-Level Access for AI Safety Audits

US-based artificial intelligence company Anthropic has announced that it granted “staff-level access” to six employees from the non-profit evaluation organization METR (Model Evaluation & Threat Research). This move fulfills previous commitments to strengthen third-party safety oversight amid growing concerns about catastrophic risks from frontier models.

According to the announcement, the METR team has officially taken up residence at Anthropic’s offices, gaining access to company hardware as well as internal tools and systems. Their permissions are roughly equivalent to those held by Anthropic’s internal risk assessment team. The company encourages its employees to share information with these external evaluators as they would with internal risk colleagues.

Regarding independence, Anthropic stated it will not pay METR for this collaboration. METR’s funding comes from individual donors, charitable organizations, and government bodies such as the UK’s AI Safety Institute, and it does not accept donations from AI companies or their employees. While Anthropic cannot veto the publication of analysis reports produced by the evaluators, it can request the redaction of specific categories of information, such as customer data, attorney-client privileged content, or model architecture details.

This initiative is seen as the first step in implementing a proposal made by Anthropic CEO Dario Amodei in September, which received support from OpenAI CEO Sam Altman and xAI CEO Elon Musk for stationing external auditors at leading AI labs. Although critics argue that such voluntary arrangements may still suffer from conflicts of interest and call for stronger government regulation, Anthropic likens the model to banking regulators stationing resident inspectors within large banks to proactively identify threats to public interest.