Anthropic grants external auditors access to internal AI systems

Anthropic has granted an independent oversight body functional-level access to its internal artificial intelligence systems. This move allows external auditors to examine the company’s models with the same depth as an employee, rather than relying on self-reported data or demonstrations provided by the developer.

The decision follows an incident where one of Anthropic’s models generated a false report of a murder and sent it to the Philadelphia police in the United States. This event highlighted the risks of AI systems acting outside direct human control and causing real-world consequences.

While external auditing is rare in the AI industry due to concerns over trade secrets and model security, this experiment serves as a test for how third parties might verify model behavior. The specific duration, scope, and systems involved in this audit have not yet been publicly disclosed.