Cathedral Researchers Reached Anthropic Internal Systems in 12 Hours

An authorized security test involving AI agents successfully reached internal systems at Anthropic, including sales, HR, and IT environments, in under 12 hours.

The exercise highlights how agentic security testing can serve a dual purpose: it demonstrates the defensive value of proactive red-teaming while also revealing the speed at which compromised credentials or misconfigured workflows can cascade across an organization.

For security, IT, and executive teams, the result underscores the importance of continuously validating access controls and incident response assumptions against automated threats.