Startup Founded by Former DOGE Staff Breached Anthropic's Internal Systems in 12 Hours
nashnova research
Cathedral, a cybersecurity startup founded by former DOGE staffers, says its AI agents found a path into Anthropic's internal systems in under 12 hours — pulling meeting notes, pay details, and Slack messages about its most advanced model. The entry point was not Claude itself but the third-party office software Anthropic runs on.
What exactly did they get into?
Cathedral's AI agents accessed Anthropic's meeting notes, salary and hiring details, and Slack messages referencing its most advanced model.
The breach came not through Claude but through third-party business software — the administrative and sales tools most companies rely on daily.
This means → what broke was not the AI's "brain" but the office infrastructure wrapped around it. The door was unlocked; the brain was fine.
How does Anthropic frame it?
Anthropic sharply narrowed the scope: Cathedral accessed only a "limited set" of sales, HR, and IT systems, including some client contact information and basic user-account records.
The company stressed that core assets were untouched — Claude user conversations, source code, and model weights remained secure. Its own monitoring flagged some anomalous activity before Cathedral reported in.
Anthropic also accused Cathedral of violating its bug-bounty rules — failing to fully disclose all activities and overstating "implied national-security implications." It shut down affected software and notified federal officials.
What do outside experts and officials say?
Security expert Alex Stamos — Cognition's CISO, invited by the Washington Post to review the findings — said plainly: "They found something big." He added that Anthropic "could have done better policing outside software vendors."
A senior Trump-administration official told the Post the report had circulated inside government and raised national-security concerns: "If a bunch of kids who just started a company can do this, the Chinese are probably not far behind."
The official also disclosed that three federal agencies have already contracted with Cathedral. This reflects how fast Washington's demand for AI-security offense-and-defense capability is converting into actual procurement.
Who is Cathedral?
The founding team — Gavin Kliger, Luke Farritor, Marko Elez, and Jack Stein — are all former DOGE staffers.
Reuters reported in July that Cathedral closed a $160 million round at a $1.4 billion valuation, backed by Andreessen Horowitz and Sequoia.
In plain terms = this is not a garage hack. It is a top-tier VC-backed, billion-dollar company running AI agents for offensive security testing.
Why did this surface on the same day as the IPO filing?
The disclosure landed on the same day Anthropic publicly filed its S-1, formally applying to go public.
For a company whose core narrative is "responsible AI," a supply-chain security gap points straight at governance capability — the exact issue institutional investors press hardest during an IPO roadshow.
This means → even if the core model was never touched, the question "you can't even secure the office software you run on" will come up again and again on the roadshow.
市场有风险，内容仅供研究参考，不构成投资建议。