A cybersecurity startup says its AI agents got into internal Anthropic data in less than 12 hours. Anthropic says the incident was smaller than that claim makes it sound.
The startup is Cathedral, which was founded by former members of Elon Musk’s DOGE. It reportedly found weaknesses in business software the AI company was using.
What the agents reportedly reached
The Washington Post reported that the data included Slack messages, meeting transcripts and information on salaries and hiring.
That list would worry any company. Internal chat logs and pay data are the kind of material that hurts most when it leaks, and none of it is customer-facing.
Anthropic’s account is smaller
Anthropic played down the scope. The company said a limited number of systems were affected, in its sales, human resources and IT functions.
It also said Cathedral didn’t breach Anthropic’s own systems directly. According to the company, Cathedral exploited a vulnerability in software from a third-party provider.
Anthropic said it spotted the activity and cut off access the same day. It said it has no evidence that anyone else used the same security gaps.
The two accounts don’t fully conflict. A flaw in a vendor’s tool can still expose a company’s internal messages and HR records, even if the company’s own infrastructure was never touched. For anyone whose salary sits in that data, that difference won’t mean much.
Who Cathedral is
Cathedral has raised $160 million. According to the Post’s report, it works with three US federal agencies.
That puts a startup run by former DOGE staff in an odd spot. It’s selling AI-driven security work to the government, and its most public demonstration so far is a same-day break-in at one of the best-known AI labs, which ended when that lab shut it out.
