Key facts
- METR researchers will have access comparable to Anthropic's own risk assessment teams.
- METR can publish findings without prior approval, with limited redactions for sensitive information.
- The move is part of Anthropic CEO Dario Amodei's proposal to pace AI development and coordinate safety standards.
Anthropic announced on Thursday that it has granted a team from the nonprofit AI evaluator METR employee-level access to its systems. This initiative allows METR to vet Anthropic's unreleased AI models and agents more thoroughly. The six chosen researchers will have access to company hardware, office space, and internal tools. Anthropic stated the goal is to close the gap between what evaluators understand about frontier AI companies' systems and what the companies themselves understand. This step fulfills a promise made by Anthropic CEO Dario Amodei last month to pace AI development and coordinate safety standards. METR can publish their findings with limited redactions for customer data, model architecture, and attorney-client privileged information, and Anthropic reserves the right to require redactions.