Former Government Efficiency Staffers Breach Anthropic Internal Systems

Former Government Efficiency Staffers Breach Anthropic Systems – Cyber Startup Highlights Enterprise Software Vulnerabilities

Key Takeaways

• Ex-employees of the Department of Government Efficiency (DOGE) used AI agents to breach Anthropic’s internal systems in under 12 hours.
• The breach occurred through third-party enterprise tools, exposing administrative data and internal chats without compromising core AI models.
• Anthropic shut down access the same day, citing rule violations in how the cyber startup Cathedral reported the vulnerabilities.
• The incident coincides with Anthropic’s Nasdaq IPO filing, which targets a contested valuation of up to $2 trillion.

Cathedral Navigates Bug Bounty to Penetrate Administrative Systems

Four former employees of the Department of Government Efficiency (DOGE) successfully breached the internal network of AI company Anthropic. The government office, originally run by Elon Musk and Vivek Ramaswamy for President Donald Trump, focused on extensive federal cost-cutting measures. These 2025 efficiency initiatives ultimately resulted in more than 250,000 people leaving the federal workforce. The department officially closed its operations on July 4, 2026.

Less than three weeks after the department shut down, the four former staffers surfaced in the private sector. Gavin Kliger, Luke Farritor, Marko Elez, and Jack Stein established a military cyber startup called Cathedral. The newly formed company rapidly secured a $1.4 billion valuation following a $160 million funding round in July. Andreessen Horowitz and Sequoia Capital led the investment, and a senior Trump administration official confirmed that three federal agencies have already hired the startup.

Operating within Anthropic’s official bug bounty program, Cathedral utilized artificial intelligence agents to bypass primary defenses. They completed this infiltration within 12 hours. Cathedral researchers initially spent several days attempting to breach Anthropic’s proprietary software directly, but these direct attacks failed. The team then redirected their AI agents to target unnamed external software tools utilized for routine sales and administrative tasks.

This alternative route provided unauthorized access to sensitive internal records. Cathedral retrieved meeting transcripts, salary details, job offers, and internal Slack communications discussing advanced AI models. The startup circulated a draft report emphasizing that vulnerabilities in standard enterprise software can compromise the infrastructure of highly advanced organizations.

Anthropic Secures AI Models and Criticizes Disclosure Methods

Anthropic acknowledged the incident but immediately clarified the strict limits of the accessed data. The AI firm confirmed that Cathedral reached a limited set of its human resources, information technology, and sales systems. Core technical assets remained entirely unaffected by the intrusion.

The company stated that customer chats, source code, and the specific model weights required to operate its Claude AI systems were never compromised. Anthropic’s internal monitoring systems actually flagged the unauthorized activity before Cathedral submitted its official security report. The company subsequently terminated the startup’s system access on the exact same day.

Anthropic representatives strongly criticized Cathedral’s operational methods and public conclusions. The AI firm stated that the startup violated the established rules of the bug bounty program by failing to disclose the full extent of its hacking activities. Anthropic also argued that Cathedral deliberately overstated the potential national security risks associated with these specific administrative vulnerabilities.

Following a comprehensive internal review, Anthropic reported no evidence indicating that other malicious actors had exploited these same flaws. The company maintains that its primary AI infrastructure remains completely secure against similar third-party entry points.

IPO Valuation Faces Skepticism Amid Hacking Revelation

The timing of this bug bounty incident aligns directly with Anthropic’s aggressive move toward public financial markets. The AI giant filed its official paperwork for a Nasdaq initial public offering on the exact same day news of the Cathedral breach broke. This public filing now places the company under intense scrutiny from both cybersecurity analysts and institutional investors.

Reports indicate that Anthropic is seeking a highly ambitious valuation of up to $2 trillion for its upcoming stock market debut. Financial experts have already begun to publicly question the viability of this massive figure. Nigel Green, CEO of financial advisory firm deVere Group, characterized the proposed valuation as “nuts.”

The sudden juxtaposition of an internal system breach and a massive IPO filing highlights the distinct operational risks facing high-valuation technology firms. Companies operating at this scale must carefully balance rapid market expansion with the comprehensive security of their decentralized software environments.

Our Assessment

The breach of Anthropic’s administrative systems by Cathedral demonstrates the persistent vulnerabilities found in third-party enterprise software. While Anthropic successfully protected its core AI models and customer data, the exposure of internal communications highlights a structural weakness in corporate supply chains. The incident shows that advanced AI agents can rapidly identify and exploit external access points when direct attacks fail. The timing of this disclosure adds immediate pressure to Anthropic’s public offering, requiring the company to reassure investors regarding its comprehensive security protocols. The rapid transition of former government efficiency staff into a heavily funded military cyber startup underlines the fast commercialization of AI-driven cybersecurity tools.