An AI company was broken into by AI agents in under 12 hours. That headline alone is enough to stop you scrolling. But the real story is stranger, and more useful, than the headline suggests.
What Actually Happened
On 26 October 2026, The Washington Post reported that Cathedral, a months-old cybersecurity startup founded by former US DOGE staffers, told Anthropic that its AI agents had found paths into the company’s internal systems in less than 12 hours. According to a draft Cathedral report reviewed by the Post, the agents reached meeting transcripts, salary and job-offer details, and Slack messages discussing Anthropic’s most advanced AI models.
The way in was not Claude itself. Cathedral exploited weaknesses in third-party business software that Anthropic uses for sales and administrative work, the same kind of ordinary office software thousands of companies rely on every day.
The Part the Headlines Skip
This was not a criminal hack. Cathedral probed Anthropic under the company’s own bug bounty programme, which invites outside researchers to find flaws, and disclosed its findings to Anthropic earlier this month.
Anthropic also disputes the scale. It confirms the researchers reached a “limited set” of sales, HR and IT systems, but says customer conversations, source code and model weights were never reached. Its monitoring flagged some of Cathedral’s activity before the report arrived, and access was shut off the same day. Anthropic further says Cathedral overstated the national-security angle and did not fully follow the programme’s disclosure rules.
So both sides agree on the way in. They disagree on how far it went, and what it means.
Why This Matters to Your Business
Here is the uncomfortable lesson. The weak point was not the AI, the code, or the model. It was the everyday software around it: the tools for recruiting, HR records, customer relations and meetings. If AI agents can map those gaps in half a day at one of the world’s best-defended AI labs, every business needs to ask the same question about its own vendors.
Three things worth doing this week:
- List your third-party tools. Every platform that holds your business data, from your CRM to your HR software, is part of your security perimeter.
- Patch and review access. Make sure those tools are updated, and that former staff and unused accounts no longer have access.
- Ask vendors hard questions. How do they handle vulnerabilities, and how fast do they fix them? If they cannot answer, that is your answer.
Frequently Asked Questions
Was Claude itself hacked?
No. Neither side reports any access to Claude’s conversations, source code or model weights. The entry point was ordinary business software.
Does this mean AI agents are a threat to my website?
It means automated attacks are getting faster and cheaper. The defence is the same as ever: updated software, tight access control, and regular security checks.
Not sure how exposed your business is?
Altus Web Solutions offers a free website security audit. We will check your site, your tools and your exposure, and tell you exactly what to fix, in plain English. Email us at [email protected] to get started.
