Cathedral, a startup founded by ex-DOGE staffers, says its AI agents broke into Anthropic in under 12 hours

Cathedral, an AI startup founded by former staffers of the US Department of Government Efficiency, says its AI agents broke into Anthropic's internal systems in under 12 hours, and Anthropic has confirmed that access took place, according to a report by Pasquale Pillitteri.

The report says the intrusion reached employee details and internal messages, and that the work was driven by autonomous agents rather than a conventional human-led penetration team.

The 12-hour figure is the part that matters. Deep intrusions into a leading AI lab have traditionally been measured in weeks or months; when reconnaissance, probing and lateral movement are handed to agents that keep running, the defender's window shrinks to hours.

Cathedral's background complicates the picture. Staffed by former DOGE personnel and pitching AI capability to security and defense-adjacent customers, a public demonstration against a frontier lab doubles as the most persuasive proof of what its tools can do.

Anthropic's confirmation is narrow. The company acknowledges that access occurred, while the public reporting so far leaves the technical details and the full scope of the incident unclear.

Timing matters too. The disclosure landed on the same day Anthropic formally filed for its IPO, and a company heading to public markets now has to explain whether its safety story still holds when the attacker is also an AI.

Three things to watch: whether Anthropic publishes a fuller postmortem, whether affected employees and customers are notified, and whether other frontier labs disclose similar agent-driven intrusion tests.