Gavin Kliger — former DOGE staffer, former Pentagon chief data officer, now co-founder of a cybersecurity startup called Cathedral — walked into Anthropic's systems and had access to meeting transcripts, salary details, Slack messages about advanced AI models, and customer business contacts. It took him and his team just twelve hours.
Cathedral was founded by four former DOGE staffers — Kliger, Luke Farritor, Marko Elez, and Jack Stein — and the exercise was a contracted security audit, not a break-in. But the results read less like a pentest report and more like a property inspection where the inspector finds the front door was never installed. The DOGE kids found "a serious gap between the offensive capabilities now emerging and the defenses" at one of the most well-funded AI companies on the planet.
The specific data accessed tells the story better than any summary. Sales systems. HR systems. IT infrastructure. Job-offer details. Partner business contacts. Basic user account records. Slack messages discussing Anthropic's most advanced models. Cathedral's team noted in their report that "data accesses were kept to the absolute minimum" — meaning they stopped looking once they'd proven the point. The full extent of what was accessible is larger than what they chose to touch.
Alex Stamos, chief information security officer at Cognition, reviewed the findings and immediately proclaimed, "They found something big here." On Anthropic's vendor oversight specifically, Stamos said the company "could have done a lot better." When a CISO at a competing AI firm tells the press your security needed work, the professional courtesy translation is that it was bad. Really bad.
A Trump administration official framed the national security implications without decoration: "If a bunch of kids who have just started a company can do this, then potentially, the Chinese aren't far behind."
Anthropic builds the AI models that government agencies and major corporations are integrating into their operations. The security gap Cathedral found isn't academic — it's a map showing where a foreign intelligence service could walk in through the same door.
The AI industry has spent the last three years asking Washington and the public to trust them with increasingly powerful systems. Trust us to build safely. Trust us to deploy responsibly. Trust us to self-regulate. Anthropic specifically has positioned itself as the "safety-first" AI lab — the responsible adults in a room full of move-fast-and-break-things competitors.
Cathedral has raised $160 million in funding, carries a $1.4 billion valuation backed by Andreessen Horowitz and Sequoia, and has already signed contracts with three federal agencies. The company was born out of the DOGE operation — the same government-efficiency initiative that the press spent months calling a group of unqualified kids playing with federal databases. Those unqualified kids now run a billion-dollar cybersecurity firm that just exposed a gaping hole in one of Silicon Valley's flagship AI companies.