Cathedral AI Startup Breaches Anthropic Internal Systems

AI cybersecurity startup Cathedral used AI agents to breach internal systems at Anthropic in under 12 hours. The intrusion targeted vulnerabilities in third-party enterprise software, including IT management tools, sales CRMs, and HR platforms, rather than Anthropic's core code or model weights.

Through these exploits, Cathedral accessed sensitive administrative data, including internal meeting transcripts, employee salaries, and Slack discussions concerning advanced AI models. Anthropic confirmed the breach occurred under its vulnerability-disclosure program but accused Cathedral of violating the program's terms by failing to fully describe the scope of its activity.

In response, Anthropic disabled the affected software and coordinated with vendors to apply security patches. A senior Trump administration official described the breach as a national security risk, comparing the potential for foreign adversaries to access AI lab internals to accessing nuclear weapons grade material. The federal government is currently working with Cathedral to evaluate hostile AI scenarios.