Anthropic expanded METR’s independent evaluation access by providing employee-level tooling and allowing METR to publish risk assessment findings with limited redaction controls
Partnership Provisional 78% confidence first seen
Anthropic began granting six METR researchers employee-like access to evaluate its AI models and safety and risk management practices under third-party oversight. METR is set to publish its findings, while Anthropic can request redactions for narrowly defined sensitive categories such as customer data and attorney-client privileged information. The arrangement is described as expanding evaluator partnerships in tiers and involving Anthropic hardware and internal tools comparable to its own teams.