AI Agents Breach Anthropic Systems: Cybersecurity Startup Cathedral Exposes Vulnerabilities

Cybersecurity startup Cathedral, founded by former DOGE staffers, reported its autonomous AI agents breached Anthropic's internal systems, gaining access to meeting transcripts, salary data, job postings, and Slack messages through third-party administrative and sales software in under twelve hours.

How Cathedral's AI Agents Gained Access

Cathedral's AI agents reportedly exploited vulnerabilities within Anthropic's third-party administrative and sales software. Through these entry points, the agents gained access to various internal business systems. The startup claims its AI tools were able to navigate and extract sensitive information, demonstrating a sophisticated level of autonomous penetration.

Data Exposed During the Incident

According to Cathedral, the compromised data included meeting transcripts, salary information, job postings, and internal Slack messages. Anthropic confirmed that there was limited access to its sales, HR, and IT systems, which included contact details for some customers and partners. However, Anthropic maintains that critical assets such as Claude chats, source code, and model weights remained secure and were never exposed during the incident.

Anthropic's Response and Discrepancies

Upon detecting the unauthorized access, Anthropic acted swiftly. The company stated it shut down the affected access points on the same day the breach was identified, disabled the compromised third-party programs, and promptly notified U.S. federal authorities. This rapid response aimed to contain the breach and mitigate further risks.

Disagreement Over Scope and Duration

Despite Anthropic's swift action, a public disagreement has emerged between Cathedral and Anthropic concerning the exact duration and overall scope of the breach. Both parties have also accused each other of violating established bug bounty disclosure rules, adding a layer of contention to the incident. This highlights the complexities that can arise when security researchers uncover vulnerabilities in high-profile organizations.

The Broader Implications for AI Security

This event underscores a critical challenge in the rapidly advancing field of artificial intelligence: securing AI systems themselves. As companies like Anthropic develop powerful AI models, the infrastructure supporting these operations becomes a prime target. The use of AI agents by Cathedral to identify and exploit weaknesses demonstrates a new frontier in cybersecurity, where automated systems are pitted against each other.

For businesses, this incident serves as a stark reminder of the importance of robust third-party vendor security and continuous monitoring of internal systems. The fact that AI agents could autonomously navigate and extract data in a relatively short timeframe suggests that traditional security measures may need to evolve to counter these sophisticated threats. Organizations must consider how their own AI-powered developer platforms and administrative tools could be potential vectors for attack.

Key Takeaways

• Cybersecurity firm Cathedral used AI agents to breach Anthropic's internal systems.
• The breach exposed meeting transcripts, salary data, and customer contact information.
• Anthropic quickly shut down access and notified federal authorities, confirming Claude chats and source code were not exposed.
• Both companies dispute the breach's scope and duration, citing bug bounty disclosure issues.
• The incident highlights the growing need for advanced AI-driven cybersecurity defenses.

Sources

• Hacker News 热门文章摘要(2026-01-19)
• PStryder/Cathedral: Cathedral is a local-...
• sanctum/docs/crypto.md at master - jorisvink
• The DOGE Boys Get VC Funding to Support Their Latest Enterprise | WIRED