Cybersecurity startup Cathedral accessed sensitive internal systems at AI company Anthropic in under 12 hours, exposing employee data and internal communications during a sanctioned security vulnerability test.
Key Points
- Cathedral researchers used AI agents to exploit weaknesses in third-party administrative and sales software used by Anthropic.
- The breach exposed employee salary details, job offers, internal Slack messages, and business contact information for partners.
- Anthropic confirmed that its core AI model weights, source code, and customer chatbot conversations remained secure during the incident.
- The company disabled the affected software immediately after its monitoring systems flagged the unauthorized activity.
- Anthropic reported the incident to federal officials, though it disputes Cathedral's claims regarding the severity of the national security implications.