NEWS · COMPANIES · #2721
Researchers using AI agents accessed Anthropic internal systems via third-party software
Cybersecurity startup Cathedral says AI agents it ran exploited vulnerabilities in third-party admin and sales software to access Anthropic internal records (meeting transcripts, salary data, Slack messages and other HR/sales/IT data) and reported the findings under Anthropic's bug bounty; Anthropic confirms limited access to some sales, HR and IT systems, says source code and model weights were not exposed, and says it disabled affected access points, patched issues, and notified authorities. The two sides dispute how extensive the access was and whether Cathedral's actions stayed within the program's safe-harbor rules.
KEY POINTS
- Cybersecurity startup Cathedral says AI agents it ran exploited vulnerabilities in third-party admin and sales software to access Anthropic internal records (meeting transcripts, salary data, Slack messages and other HR/sales/IT data) and reported the findings under Anthropic's bug bounty; Anthropic confirms limited access to some sales, HR and IT systems, says source code and model weights were not exposed, and says it disabled affected access points, patched issues, and notified authorities.
- The two sides dispute how extensive the access was and whether Cathedral's actions stayed within the program's safe-harbor rules.
- Demonstrates that AI agents can automate low-cost, continuous probing of corporate networks and third‑party integrations, posing a new operational security risk even for leading AI labs.
WHY IT MATTERS
Demonstrates that AI agents can automate low-cost, continuous probing of corporate networks and third‑party integrations, posing a new operational security risk even for leading AI labs.