GoKawiilCathedral says AI agents breached Anthropic's HR and sales software in 12 hours
Security startup Cathedral says it ran an authorized test on Anthropic's systems and found autonomous AI agents could gain access within 12 hours by exploiting third-party sales, HR and IT software rather than Anthropic's Claude model itself. Cathedral says the agents reached compensation data, internal communications, meeting transcripts and Slack discussions about Anthropic's models, and disclosed the findings on October 26, 2026. Anthropic confirmed limited access to some sales, HR and IT systems but says customer conversations, Claude outputs, source code and model weights were not reached.