Anthropic Update on Independent Evaluators

Anthropic announced it is granting deep access to its AI models and safety practices to independent evaluators, including a team from nonprofit METR, to increase public accountability while protecting sensitive information that cannot be publicly shared. The company is providing METR employees with employee-like access to internal systems and tools, and is also working with SecureBio and Accenture as evaluators, with plans to partner with additional organizations. This approach mirrors bank regulatory supervision, aiming to close the gap between what external evaluators and the company itself understand about frontier AI risks.