A months-old cybersecurity startup founded by former DOGE staffers says its AI agents found paths into Anthropic’s internal systems in under 12 hours, the Washington Post reported Monday, citing a draft report it reviewed. The company, Cathedral, says it reached meeting transcripts, salary and job-offer details, and Slack messages about Anthropic’s most advanced models. The breach was not in Claude itself – the hackers exploited flaws in third-party business software, the kind of administrative and sales tools most companies run. The Post withheld the vendors’ names.
Cathedral reported the findings to Anthropic earlier this month through Anthropic’s bug bounty program, which pays outside researchers to disclose flaws, and told the affected vendors too. Its draft calls the result evidence of “a serious gap between the offensive capabilities now emerging and the defenses,” and says its “data accesses were kept to the absolute minimum.” A Cathedral spokeswoman declined to comment to the Post.
Anthropic says the breach was much narrower – that Cathedral reached “a limited set” of its sales, HR and IT systems, including some customer and partner business contacts and basic user account records, and that it cut off the startup’s access the same day. It says Claude users’ conversations, source code and model weights were not touched, that its own monitoring flagged some of the activity before Cathedral reported it, and that it has no evidence anyone else exploited the flaws, Gizmodo and Tech Startups report. Anthropic also says the researchers spent several days failing to break into its software directly before turning to the vendors’ applications.
It accuses Cathedral of breaking the bounty program’s rules by not disclosing all of its activity, and of overstating the “implied national security implications.” It switched off the affected software until the vendors fixed it and briefed federal officials.
A security expert who reviewed the findings at the Post’s request said Cathedral “found something big.” Alex Stamos, chief information security officer at Cognition, reviewed them. “They found something big here,” he said, adding that Anthropic “could have done a lot better” in overseeing outside software vendors. Cognition uses models from Anthropic and other labs. A senior Trump administration official told the Post the report had circulated inside the government and raised national security concerns: “If a bunch of kids who have just started a company can do this, then potentially, the Chinese aren’t far behind.”
Cathedral is no obscure shop. Gizmodo lists its founders as Gavin Kliger, Luke Farritor, Marko Elez and Jack Stein, all DOGE alumni. Kliger was the Pentagon’s chief data officer and is reportedly tied to the Pentagon’s legal dispute with Anthropic. Reuters reported in July that Cathedral raised $160 million at a $1.4 billion valuation, backed by Andreessen Horowitz and Sequoia, and the senior official told the Post that three federal agencies have recently signed contracts with it.
The disclosure lands as Anthropic is courting public investors. On Monday the company dropped its S-1, showing $974 million operating profit against an $800 billion compute bill. The security question for a lab pitching itself as the careful one is now a concrete incident, and the two sides disagree on how big it was.
Keep reading
You must be logged in to post a comment.