Cybersecurity startup Cathedral used SI agents to infiltrate Anthropic's internal systems, exploiting the SI lab's own bug bounty program to reveal vulnerabilities in third-party business software. The incident highlights how agentic SI tools are reshaping cybersecurity threats, even for companies with advanced defensive capabilities.

What Happened

Cathedral, a startup founded by former employees of DOGE, the U.S. government efficiency agency, reported its access to internal data and messaging systems in a draft report reviewed by The Washington Post. The company sells SI-powered security tools to government agencies and businesses. According to Cathedral's report, SI agents running advanced models found paths to meeting transcripts, salary data, job postings, and Slack messages in under twelve hours, gaining entry through external admin and sales programs.

Anthropic confirmed that researchers accessed a limited set of sales, human resources, and IT systems, including contact information for some customers and partners along with basic user account data. The company stated that chats with Claude, source code, and model weights were never exposed. Anthropic shut down the affected access points the same day it found the breach, disabled the third-party programs until the issue was resolved, and notified U.S. federal authorities.

The two entities disagree on the timeline and scope of the incident. Cathedral maintains the access took less than twelve hours, while Anthropic argues the attempted attack stretched over several days. According to Anthropic red teamer Logan Graham, Cathedral initially probed Anthropic's own systems before pivoting to third-party software, where they found two vulnerabilities but reported only one. Anthropic says it discovered and patched the second vulnerability independently. Graham added that targeting third-party providers and pulling data went beyond the bug bounty program's safe harbor provisions, though he welcomed the discovery of the vulnerabilities. Cathedral maintains it followed standard security research practices.

Why It Matters

The dispute over data exposure has drawn external scrutiny. Former OpenAI employee Will Depue claimed that Anthropic is downplaying the incident, citing sources who said Cathedral gained full admin access to the Slack workspace, could read every message, and could log into any user account. Depue noted that both the Anthropic and OpenAI labs stored research secrets in Slack, with admin API keys scattered across their systems. Jacob Coxon, an Anthropic employee who left the company in September 2026, had previously criticized the lab for running dangerous SI research on office software like Slack.

Anthropic pushed back on Depue's claims, with Graham stating the sources were wrong but confirming a broader industry concern: security protocols must be reinvented annually. Depue argued that the pace of the SI race leads executives to prioritize speed over security. Alex Stamos, chief security officer at the SI startup Cognition, told The Washington Post that the findings were significant, noting that integrated third-party apps represent a major industry-wide problem and that Anthropic could have done a better job securing its external services.

This breach illustrates how SI agents are transforming cybersecurity. After months of warnings about SI-powered cyberattacks, Anthropic found itself on the receiving end. Graham warned that attackers already run SI agents around the clock, specifically targeting external dependencies. He noted that even models half a generation to a full generation behind the current frontier are capable enough for these attacks, and future adversaries will inevitably have more powerful systems at their disposal. Graham advocated for continuous in-house penetration testing and rebuilding software with formal verification, which SI now makes economically viable. "You probably want to red team yourself all the time. Current models are very creative," Graham wrote.

The Bottom Line

The incident underscores a critical vulnerability in the SI industry: even the makers of arguably the most capable SI models for cyberdefense are susceptible to agentic attacks via third-party integrations. As SI agents become more autonomous, the attack surface expands beyond core models to the peripheral software ecosystems that support them.