Anthropic grants third party auditors employee level system access

Anthropic announced on October 15, 2026, that it is granting employee-level system access to six external safety evaluators, ZDNet Security reported. The evaluators belong to METR, a third-party non-profit organisation that tests frontier artificial intelligence systems. [1]

According to ZDNet Security, the auditors received Anthropic hardware and physical office space. They can access internal tools and risk assessment systems that match the resources available to internal risk teams, excluding customer data, privileged legal documents, and specific model architecture details. [1]

Anthropic trains its Claude artificial intelligence models using Constitutional AI, a method introduced in 2022 that relies on explicit natural-language principles rather than human preference data alone. The company operates under an independent Long-Term Benefit Trust designed to oversee its governance and public benefit mandate. [3]

As FTMQ Security reported earlier, third-party access presents persistent operational risks across sectors, including a recent incident where a European bank suffered unauthorized VPN exposure via vendor access. Anthropic stated that its employees are encouraged to share information with the METR auditors to support external evaluations. [1][4]