A $1.4 Billion Cybersecurity Startup Run by Ex-DOGE Staffers Says Its AI Agents Got Inside Anthropic in Under 12 Hours

Cathedral, a months-old cybersecurity startup founded by former members of the U.S. DOGE Service, recently told Anthropic that its researchers penetrated sensitive internal data and messaging systems at the Claude maker after running AI agents for less than 12 hours, according to a draft report reviewed by The Washington Post and published Monday.

According to the draft, Cathedral's agents found routes to meeting transcripts, details of salaries and job offers, and Slack messages discussing Anthropic's most advanced AI models. The company got in by exploiting vulnerabilities in commercial business software that Anthropic and many other companies use for administrative and sales work; the Post withheld the vendors' names for security reasons. Cathedral conducted the probe under Anthropic's own program that invites outside security researchers to test its systems, disclosed the findings earlier this month, and said it also informed the software vendors. The draft calls the result evidence of "a serious gap between the offensive capabilities now emerging and the defenses."

Anthropic's account is narrower. The company confirmed that Cathedral reached "a limited set" of its sales, human resources and IT systems, exposing some customer and partner business contact details and basic user account records. It said the researchers could not access messages customers sent to its models, the responses returned, or the code and model weights behind its systems. Anthropic said its own monitoring flagged some of the activity before Cathedral reported it, that access was shut off the same day, and that the affected software stayed disabled until the vendors patched it. It added it has "no evidence that anyone exploited these vulnerabilities outside of Cathedral," and pushed back on the startup for what it described as overstated national security implications and incomplete disclosure of its testing activity.

Cathedral was founded by four DOGE alumni to sell AI-driven cyber capabilities to government and corporate clients. It raised $160 million at a $1.4 billion valuation in July, with Andreessen Horowitz and Sequoia leading, and Marc Andreessen and Sequoia's Shaun Maguire sit on its board. Reuters has reported that three federal agencies recently signed contracts with the company. Co-founder Gavin Kliger served as the Pentagon's chief data officer after his DOGE stint.

The episode has already echoed inside the government. A senior Trump administration official told the Post the report circulated within the administration and raised national security concerns, saying "if a bunch of kids who have just started a company can do this, then potentially, the Chinese aren't far behind" and describing the sensitivity of a company holding what the official called almost nuclear weapons grade material. Alex Stamos, chief information security officer at Cognition and former director of the Stanford Internet Observatory, reviewed the findings at the Post's request: "They found something big here," he said, adding that hooking third-party applications into company systems "is a humongous problem for everybody" and that the episode exposed a gap in how Anthropic oversees outside software providers. "Anthropic could have done a lot better," he said.

The weak point was not Claude or its weights but ordinary office software, the same third-party layer every large organization depends on. It caps a rough stretch for AI lab security: the FBI earlier this month arrested an OpenAI engineer for allegedly planting a recording device in a conference room, and Anthropic last week confirmed one of its models pulled Georgia vote-count data during a training run. If AI agents can chart a path through vendor software in half a day, the labs guarding the industry's most valuable secrets will have to secure their suppliers as tightly as their own code.