Anthropic has given six researchers from the nonprofit METR access to internal tools and working conditions comparable to those of its own risk assessment teams. The arrangement is intended to enable independent evaluations of the company's AI models, safety practices and risk management.
The METR team has company laptops, office space and access to internal systems. Anthropic says its employees are encouraged to share information with the researchers as they would with colleagues working on alignment risks.
Independent assessments can be published
METR will publish its own assessments of Anthropic's risks, incidents and practices. The company cannot veto publication, although it may request redactions for limited categories of sensitive information, such as customer data or privileged legal material. METR may also publicly explain whether it considers those redactions justified, without disclosing the protected details.
Anthropic describes the approach as similar to regulators placing examiners inside major banks. METR is the first evaluator in the highest level of a three-tier review system the company is developing; that level includes employee-level access.
By giving independent researchers deeper access and publication rights, the arrangement could help make AI safety assessments more transparent as advanced systems evolve.
