Four former employees of Donald Trump's Department of Government Efficiency (DOGE) used autonomous AI agents to breach internal enterprise systems at Anthropic in under 12 hours through the company's authorized bug bounty program.
How Cathedral Circumvented Anthropic's Defenses
The Department of Government Efficiency, led by Elon Musk and Vivek Ramaswamy, closed on July 4, 2026, after federal cost-cutting measures pushed over 250,000 people out of the workforce. Less than three weeks later, former DOGE staffers Gavin Kliger, Luke Farritor, Marko Elez, and Jack Stein launched Cathedral, a $1.4 billion military cyber startup. Cathedral raised $160 million in July in a funding round led by Andreessen Horowitz and Sequoia Capital.
According to a Cathedral draft report, researchers initially attempted direct attacks on Anthropic's primary software for several days without success. The team then instructed AI agents to target third-party enterprise tools used for administrative, HR, and sales operations. These entry points gave the researchers access to employee salaries, job offers, meeting transcripts, and Slack chats discussing advanced AI models.
Anthropic Pushback and IPO Filing Context
Anthropic confirmed that Cathedral reached a limited set of sales, HR, and IT systems, but emphasized that customer chats, source code, and model weights for its Claude models were never accessed. Anthropic stated its internal monitoring detected the activity before Cathedral formally reported it, allowing the firm to revoke access on the same day. Anthropic also stated Cathedral violated program guidelines by failing to disclose all activity and overstating the national security implications.
The disclosure surfaced on the same day Anthropic filed to go public for a Nasdaq listing. Valuations for the AI developer's market debut are reported up to $2 trillion, a figure described as "nuts" by Nigel Green, CEO of financial advisory firm deVere Group. Three federal agencies have already retained Cathedral's services, according to senior administration officials.
Key Takeaways
- Cathedral researchers accessed Anthropic's enterprise systems within 12 hours using AI agents.
- Claude model weights, core source code, and customer communications remained secure.
- Cathedral raised $160 million at a $1.4 billion valuation led by Andreessen Horowitz and Sequoia Capital.
- Anthropic is pursuing a Nasdaq IPO with potential valuation estimates reaching $2 trillion.
Why It Matters
This incident highlights how corporate enterprise tools remain the primary vulnerability for advanced artificial intelligence firms, even when core model weights and proprietary code bases are heavily fortified. As state actors and private contractors deploy autonomous AI agents to scan for enterprise supply chain weaknesses, security perimeter management is shifting from months-long penetration testing to real-time automated defense. With Anthropic moving toward a historic public listing, regulatory and institutional scrutiny surrounding third-party software risks will likely intensify across the entire artificial intelligence sector.
