Stories
Anthropic Grants AI Evaluator METR Employee-Level Access to Models and Safety Systems
- •A team of non-profit AI evaluators from METR has received employee-level access to Anthropic's systems, advanced models, and safety practices.
- •Six METR employees were provided with company hardware, office space, and internal tools comparable to Anthropic's own risk assessment teams.
- •METR has the right to publish independent assessments of risks and safety practices without requiring prior approval from Anthropic, subject only to limited redactions.
- •Anthropic is developing a three-tiered access system for evaluators, ranging from narrowly scoped investigations to full employee-level access and publication rights.
Key Questions Answered
Anthropic has provided six METR employees with employee-level access to its systems, including company hardware, office space, and internal tools comparable to its own risk assessment teams. This grants them independent access to advanced models, safety practices, and risk management processes.
Yes, METR is permitted to publish independent analyses of risks, incidents, and practices without Anthropic having veto power or requiring prior approval. Publication is only subject to limited redactions for customer data, privileged information, and architectural details.
Anthropic is developing three tiers of access for evaluators. The first tier involves narrowly scoped investigations determined by Anthropic, the second offers broader information access and independent publishing with specific redactions, and the third provides full employee-level access and publication rights matching METR's level.
