Anthropic has granted the AI evaluation institute METR employee-level access to its models and internal systems, in an unusual step toward independent scrutiny of powerful AI.
Six METR staff members receive access to company hardware, offices and internal tools, on par with Anthropic's own risk teams.
METR can publish independent evaluations without prior approval. Only limited redactions apply, covering customer data, sensitive information and details of model architecture.
The arrangement is aimed at independent assessment of AI risks, while keeping user data and sensitive information outside the access granted.