Cathedral Says Its AI Agents Accessed Some Anthropic Systems

Cathedral said its agents found a route into Anthropic systems through vulnerabilities in third-party business software. Anthropic confirmed limited access to some internal systems, but said users’ conversations, source code and model weights were not accessed. The company also said Cathedral violated the rules of its vulnerability-testing program.

• Cathedral said its agents found access paths to sensitive information in under 12 hours by exploiting third-party business software vulnerabilities. Anthropic said the researchers had first spent several days unsuccessfully attempting to breach its software directly.· 1 source
• Anthropic confirmed access to a limited set of sales, human resources and IT systems, including some business contact information and basic user account records. It said the researchers did not access Claude conversations, source code or model weights.· 1 source
• The access occurred during a cybersecurity testing program that rewards researchers for finding vulnerabilities. Anthropic accused Cathedral of not disclosing the full extent of its activity and violating the program’s rules.· 1 source
• Cathedral is described as an AI cybersecurity startup focused on military applications, founded by former DOGE staffers. The article says it raised $160 million and was valued at $1.4 billion.· 1 source
• An unnamed senior Trump administration official reportedly said the incident raised national security concerns, including the possibility that similar software vulnerabilities could be exploited by Chinese hackers.· 1 source