A cybersecurity startup founded by former DOGE staffers, Cathedral, says it got into Anthropic's internal systems in under 12 hours using AI agents, according to an exclusive from The Washington Post based on a draft of the company's report.
The way in was outside business software that Anthropic uses for administrative and sales tasks, with vulnerabilities the vendors fixed after being notified. Neither the software nor the vendors were named.
What was reached, according to each side
Cathedral's draft says its agents found paths to meeting transcripts, salary and job-offer details, and Slack messages about the company's advanced models. Anthropic acknowledges access to a limited set of sales, HR and IT systems, including business contacts for some customers and partners and basic user account records. It says Claude conversations, source code and model weights were not reached.
The two sides agree on the entry point but diverge on the scope. Anthropic says its monitoring flagged part of the activity before Cathedral reported it, and that access was shut off the same day. The software stayed offline until vendors confirmed fixes.
Anthropic also says Cathedral overstated the national security implications and did not fully follow the bug bounty rules because it did not disclose all of its activity inside the systems. Cathedral's report says data access was kept to the absolute minimum. The startup's spokeswoman declined to comment to the Post.
Why Washington is paying attention
The report has circulated inside the government. A senior Trump administration official, speaking anonymously, said that if a group of kids who just started a company can do this, the Chinese may not be far behind, and was struck by how quickly they got into sensitive systems at a company handling near-nuclear-grade material.
Cathedral is valued at $1.4 billion and has raised $160 million, according to PitchBook as reported by the Post. Its board includes Marc Andreessen and Shaun Maguire, and three federal agencies recently contracted with it. The backdrop matters: the Pentagon barred Anthropic from federal systems after it refused to allow its technology for autonomous weapons or domestic mass surveillance.
In its report, Cathedral writes that its findings reveal a serious gap between emerging offensive capabilities and the defenses protecting some of America's most vulnerable technological assets.
A problem every company shares
Set the politics aside and the technical point is simple and uncomfortable. Alex Stamos, now chief information security officer at Cognition, told the Post the findings were significant, and that connecting third-party applications to company systems is a huge problem for everyone. Anthropic, he said, could have done much better.
A similar pattern emerged in September, when researchers reached OpenAI through a single connected account: again the weak point was a tool tied to the company, not its core systems. The lesson is to know every outside app that touches your systems, grant the narrowest access it needs, and monitor what it does.
What to watch
Cathedral's full report, since the Post read only a draft. The names of the vendors and whether other companies using the same software were exposed. And a public statement from Anthropic that settles what was reached.
No comments yet. Be the first!