Cathedral AI Agents Access Anthropic Business Systems Through Third-Party Software

Cybersecurity startup Cathedral said its AI agents found paths into Anthropic’s internal business systems in less than 12 hours during an authorized test through Anthropic’s outside-researcher program. The researchers exploited weaknesses in unnamed third-party administrative and sales applications, rather than Anthropic’s own code or core AI models. Anthropic confirmed limited access to sales, human resources, and IT systems involving sensitive business information, but said customer prompts, Claude responses, source code, and model weights were not accessed.

Anthropic detected some activity before Cathedral reported it, terminated access the same day, and disabled affected applications pending vendor fixes. The company said it found no evidence of exploitation by anyone outside Cathedral. The companies disputed whether Cathedral fully complied with the program’s disclosure requirements. The incident highlights third-party business applications as a route to sensitive enterprise data, reinforcing the need to restrict their access, monitor integrations, and coordinate remediation with vendors.