Cathedral, a cybersecurity startup founded by four former DOGE employees, claims its AI agents found paths into Anthropic's internal business systems in under 12 hours. The way in wasn't Claude: it was third-party admin and sales software that many other companies use too.
What the Washington Post Reported
According to the Washington Post, as picked up by Decrypt and Tech Startups, Cathedral tested Anthropic's systems through the company's vulnerability disclosure program, a "bug bounty" scheme that rewards outside experts for reporting flaws. Cathedral, which Decrypt says raised $160 million at a $1.4 billion valuation in July, says its agents found paths to meeting transcripts, compensation data and job offers, and Slack messages about advanced models.
Those claims come from a draft of Cathedral's report reviewed by the Post. That's an important detail: the broader version of the scope is asserted by the startup, not by an independent audit.
Anthropic's Version
Anthropic confirmed access to "a limited set of sales, human resources and IT systems" containing business contact data and basic account records. It says customer conversations with Claude, source code and model weights were not accessible. It also says its monitoring caught some of the activity before Cathedral reported it, that it cut off the access the same day, and that it disabled the affected software until the vendors (unnamed) confirmed fixes. It further argues that Cathedral overstated the national-security implications and didn't fully follow the program's rules, because it didn't disclose all of its activity. It has no evidence that anyone else exploited these flaws.
The Technical Lesson: The Weak Link Is the Vendor Chain
Neither Claude nor the models were the way in. The agents automated what a human team would do in days: reconnaissance, probing weak configurations and chaining access across corporate SaaS tools. The point isn't that an AI company is vulnerable, but that the cost of probing your external attack surface drops when attackers (or auditors) use agents. Any company that relies on third-party CRM, HR tools or IT platforms is in the same position.
What to Do at Your Company
- Inventory the SaaS tools with access to sensitive data (HR, sales, IT) and who administers them.
- Require MFA, least privilege, and a review of integrations and tokens in each one.
- If you run a disclosure program, set clear rules on the use of automated agents and full logging of their activity.
- Test your own external surface with agents before someone else does.
Sources: Decrypt, Tech Startups, Crypto Briefing (original Washington Post report, October 26, 2026).