Anthropic gives non-profit evaluator internal access to monitor AI safety

Anthropic is granting independent researchers employee-level access to evaluate its artificial intelligence models for safety risks.

Artificial intelligence developer Anthropic has granted non-profit evaluator METR employee-level access to its AI models. The agreement allows external researchers to inspect safety risks and system capabilities directly from inside the company. Chief executive Dario Amodei has made independent safety testing a central part of Anthropic's risk management strategy.

• The arrangementMETR researchers receive direct access to systems and internal documentation to run safety tests.
• The objectiveExternal monitoring aims to identify potential safety hazards before new models are released.