Cathedral AI Agents Breach Anthropic Systems in Under 12 Hours
A cybersecurity startup called Cathedral used AI agents to compromise Anthropic's internal business systems in under 12 hours during an authorised security test, exposing employee data, salary records, Slack messages, and meeting transcripts. The breach highlights how AI-powered offensive tooling dramatically compresses attack timelines against enterprise software perimeters. While core model assets — weights, source code, and customer conversations — were reportedly not accessed, the incident raises serious questions about third-party software exposure at frontier AI labs.