Cathedral says AI agents breached Anthropic in under 12 hours

Cathedral, an AI cybersecurity startup founded by four former Department of Government Efficiency staffers, says its AI agents found their way into Anthropic's internal business systems in less than 12 hours. The agents did not crack Claude; they exploited weaknesses in third-party administrative software the company uses for sales and operations, Gizmodo reported after the test was first revealed by the Washington Post.

A draft Cathedral report seen by the Post said the routes exposed meeting transcripts, salary information, job offers and Slack discussions concerning advanced AI models. Anthropic confirmed access to a "limited set" of its sales, human resources and IT systems, including customer and partner business contact information and some basic user account records, but said the researchers did not reach Claude users' conversations, source code, or model weights.

The breach was authorized. Cathedral participated through Anthropic's bug bounty program, and the two sides are now disputing what counted as fair play. Anthropic accused Cathedral of overstating the "implied national security implications" of its findings and failing to disclose all of its activity inside the company's systems. Cathedral, in its own report, said "data accesses were kept to the absolute minimum in accordance with security research best practices."

The context sharpens the stakes. Cathedral, founded in 2026 by DOGE alumni Gavin Kliger, Luke Farritor, Marko Elez and Jack Stein, raised $160 million at a $1.4 billion valuation this year from Andreessen Horowitz and Sequoia Capital to build AI tools for U.S. military cyber operations. A senior Trump administration official, speaking anonymously to the Post, said: "If a bunch of kids who have just started a company can do this, then potentially, the Chinese aren't far behind."