Anthropic is allowing the external AI safety evaluator METR to access its models at a level described as “employee-like,” according to multiple reports. The arrangement is intended to enable METR to more closely monitor and evaluate risks related to Anthropic’s AI systems.
The Financial Times and other outlets frame the move as part of Anthropic CEO Dario Amodei’s broader safety plan, which includes bringing in outside evaluators to increase transparency and independent scrutiny. Seeking Alpha and Channel NewsAsia report similar details, describing METR’s access as more direct and operational than typical third-party evaluation.
Across the coverage, the core point is consistent: Anthropic expands the role of an outside evaluator and grants heightened access to support ongoing AI risk monitoring. Outlets do not present differing conclusions about the purpose of the access, though they emphasize it through slightly different wording—such as “employee-like” or “employee-level” access—to convey the extent of METR’s involvement.
