Cathedral, a cybersecurity startup founded this year, says AI agents it pointed at Anthropic found routes into the company's internal systems in under 12 hours by exploiting weaknesses in third-party administrative software. Anthropic confirms the researchers reached a limited set of systems and disputes Cathedral's account of what that access means.
The test ran under Anthropic's program for outside researchers to report vulnerabilities. Cathedral's draft findings, reviewed by The Washington Post, describe access to employee compensation and job-offer details, meeting transcripts, and internal Slack discussions about advanced AI models.
Anthropic's description is narrower. The company said the researchers reached a limited set of sales, human resources and IT systems holding business contact details and basic customer and partner account records, and that customer prompts, Claude responses, source code and model weights remained inaccessible.
Key facts
- 01Cathedral says its AI agents found paths into Anthropic internal systems in under 12 hours through flaws in third-party administrative software.
- 02Anthropic says the access covered a limited set of sales, HR and IT systems, and that model weights, source code and Claude responses were not reachable.
- 03Cathedral raised $160 million at a $1.4 billion valuation in July and holds contracts with three federal agencies.
- 04Co-founder Gavin Kliger was the Pentagon's chief data officer until April 2026.
- 05A federal appeals court ruled 2-1 on September 25 that the Pentagon could exclude Anthropic from its supply chain.
Anthropic also said its own monitoring flagged some of the activity before Cathedral disclosed it, and that it cut the researchers' access the same day. It then disabled the affected third-party software until the vendors shipped fixes, and said it found no evidence that anyone else had exploited the flaws.
The two sides disagree on both process and stakes. Anthropic said Cathedral did not fully follow its testing rules and overstated the national-security implications of the findings. Cathedral said its data access was limited to what the research required.
Cathedral raised $160 million at a $1.4 billion valuation in July. Co-founder Gavin Kliger served as the Pentagon's chief data officer until April. A senior Trump administration official said three federal agencies have contracted with the company, and that the government is working with Cathedral on possible adversarial AI scenarios.
The episode lands in the middle of a live dispute between Anthropic and the Pentagon over military use of Claude. Anthropic has declined to lift restrictions covering fully autonomous lethal operations and mass domestic surveillance. On September 25, a federal appeals court ruled 2-1 that the Pentagon could exclude Anthropic from its supply chain.
Anthropic has been running the same class of tooling on the defensive side of the problem. The company reported disclosing 6,157 vulnerabilities across 591 open-source projects by October 2 using its own AI-assisted security research.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
