Ex-DOGE Staffers Breach Anthropic via Bug Bounty

From Government Cutbacks to AI Cybersecurity

Four former employees from Elon Musk's Department of Government Efficiency, known as DOGEDOGE-0.29%, recently tested the security of AI powerhouse Anthropic. Using automated AI agents, the group accessed internal corporate files in under 12 hours.

This action was not an illegal hack. It took place through Anthropic's official bug bounty program, which rewards researchers for uncovering security flaws.

What Was the DOGE Initiative?

President Donald Trump appointed Elon Musk and Vivek Ramaswamy to lead DOGE. Their primary mission was cutting federal spending and streamlining government operations.

During 2025, the office introduced major spending cuts. These changes led to more than 250,000 workers leaving the federal workforce.

The DOGE office officially closed on July 4, 2026. Shortly after, several key team members launched their own private enterprise.

The Rise of Cathedral

Less than three weeks after DOGE dissolved, four alumni launched a military cybersecurity startup named Cathedral. The company was founded by Gavin Kliger, Luke Farritor, Marko Elez, and Jack Stein.

Cathedral quickly raised $160 million in July 2026, reaching a valuation of $1.4 billion. Venture capital firms Andreessen Horowitz and Sequoia Capital led the funding round.

A senior Trump administration official told the Washington Post that three federal agencies have already hired the startup for cybersecurity services.

How AI Agents Found the Weak Spots

Cathedral researchers spent several days targeting Anthropic's core software directly. Those initial attempts failed to break through.

The researchers then directed their AI agents to target external third-party tools. Anthropic used these outside software systems for everyday administrative and sales operations.

Within 12 hours, the AI agents bypassed these secondary tools. This allowed access to internal meeting transcripts, salary data, job offers, and internal Slack chats about advanced AI models.

The fact that weaknesses in ordinary enterprise software can compromise the infrastructure of even the most advanced organizations should be a wake-up call for every company in America.

On October 26, 2026, prediction platform Polymarket highlighted the incident on social media, noting how automated agents penetrated enterprise defenses.

Anthropic Responds to the Security Finding

Anthropic confirmed that Cathedral accessed a limited set of sales, HR, and IT systems. However, the core intellectual property remained completely safe.

Anthropic emphasized that customer chats, source code, and model weights—the core files that power its Claude AI—were never touched.

Anthropic stated that internal monitoring flagged Cathedral's activity before the team submitted its report. Security teams revoked Cathedral's access on the exact same day.

Anthropic also pushed back against Cathedral. The AI company stated Cathedral broke program rules by failing to report all activities promptly and by overstating potential national security risks. Anthropic confirmed no outside malicious actors utilized these flaws.

Anthropic IPO Context and Market Reaction

The news surfaced on the same day Anthropic submitted its public filing for an Initial Public Offering (IPO) on the Nasdaq exchange.

Some market reports suggested potential IPO valuations reaching up to $2 trillion. Nigel Green, CEO of financial advisory firm deVere Group, publicly described such valuation figures as nuts.

Investors should remember that initial valuation estimates carry significant market uncertainty. Past performance of digital assets or corporate valuations is not a guarantee of future results.