Anthropic Says Cathedral Test Reached Limited Business Systems

Anthropic said researchers from cybersecurity startup Cathedral reached a limited set of its sales, human resources and IT systems during a test under the company’s external security program. The company said the researchers did not reach customer conversations with Claude, Claude’s responses, source code or model weights.

Cathedral’s unpublished draft says its AI agents found paths into Anthropic systems in fewer than 12 hours by exploiting vulnerabilities in outside software used for administrative and sales work. The draft identifies meeting transcripts, compensation and offer details, and Slack discussions about advanced models as categories the agents could potentially reach. The available accounts do not establish whether researchers viewed or copied that information.

Anthropic said the affected systems included some customer and partner business contacts and basic user-account records. An Anthropic spokesperson said, “Our monitoring flagged some of Cathedral’s activity before they reported it to us, and we shut off their access the same day.”

The company said it disabled the affected software until vendors addressed the vulnerabilities and found no evidence anyone outside Cathedral had exploited them. Anthropic also said Cathedral did not disclose all of its testing under the program’s rules and overstated the national-security implications. Cathedral’s draft said researchers kept data access to a minimum.