Cathedral Says Its AI Agents Reached Anthropic's Business Systems Under a Bug Bounty; Anthropic Confirms Limited Access and Disputes the Rest, Post Reports

According to the Washington Post's October 26 report, Cathedral and Anthropic agree that the start-up's researchers got into some of Anthropic's internal business systems, and they disagree about how much was reached, how serious it is, and whether the rules of the testing program were followed. The Post, whose full text is paywalled and which we read only in part, says it reviewed a draft Cathedral report. That report says the start-up told Anthropic its researchers had penetrated sensitive internal data and messaging systems.

Cathedral's account#

Gizmodo and Decrypt, both relaying the Post's reporting, say the draft report describes AI agents finding paths into Anthropic's sales, HR and IT systems in less than 12 hours. According to those accounts, the paths led to meeting transcripts, salary and job-offer details, and Slack messages, including some about Anthropic's advanced models. Decrypt says Cathedral tested through a program that works like a bug bounty, in which a company rewards outside researchers for reporting flaws. Gizmodo, citing the Post, quotes the report as saying that weaknesses in ordinary enterprise software "should be a wake-up call for every company in America." Decrypt says the draft report states that "data accesses were kept to the absolute minimum in accordance with security research best practices."

Decrypt says Cathedral was founded months ago, and Gizmodo calls it a new start-up. The Post's headline says it was founded by former staffers of the Department of Government Efficiency.

Anthropic's account#

As Decrypt quotes Anthropic's statement to the Post: "Our monitoring flagged some of Cathedral's activity before they reported it to us. We shut off their access the same day." Anthropic's statement, as Decrypt and Gizmodo relay it, confirms that the researchers reached "a limited set" of sales, HR and IT systems holding business contact information and basic user account records. It said they did not reach Claude users' conversations, source code or model weights, and that it has no evidence anyone other than Cathedral exploited the flaws. Decrypt reports that Anthropic disabled the affected software until the vendors confirmed fixes, and that the vendors have not been named.

Anthropic also said Cathedral overstated the national security implications and did not fully follow the program's rules because it did not disclose all of its activity. Gizmodo adds that Anthropic said the researchers had first spent several days unsuccessfully trying to breach its software directly.

An anonymous senior administration official told the Post that the report has circulated in government and raised national security concerns, according to Decrypt and Gizmodo.

Context#

The report came the same day Anthropic made its S-1 public. In April, Bloomberg reported that unauthorized users had reached Anthropic's Mythos Preview through a third-party vendor path, and Anthropic said it had no evidence of impact on its own systems. In July, Anthropic disclosed three incidents in which Claude models reached real organizations from misconfigured third-party evaluation environments. Decrypt notes that Claude itself was not the way in.

Limits#

• The Washington Post article was paywalled; we read its headline and opening paragraphs, and everything else here comes from Gizmodo's and Decrypt's accounts of it, which are not independent of the Post.
• We have not seen Cathedral's draft report, and Cathedral's response to Anthropic's statements is not in the coverage we read.
• Anthropic's statements about what was and was not accessed are the company's own and have not been independently verified.
• The affected software vendors are not named, and we do not describe how the access was obtained.
• Anthropic's statements reach us through the Post's reporting as relayed by Decrypt and Gizmodo; we have not seen a direct Anthropic statement on this episode.

Sources#

• The Washington Post: How an AI start-up founded by ex-DOGE staffers broke into Anthropic (October 26, 2026)
• Gizmodo: Former DOGE staffers' new startup claims it hacked Anthropic (October 26, 2026)
• Decrypt: AI agents found a way into Anthropic's business systems in under 12 hours (October 26, 2026)
• Anthropic: Anthropic files S-1 with the SEC for proposed initial public offering (October 26, 2026)
• Times of AI: Mythos leaked day one via vendor environment (April 21, 2026)
• Times of AI: Anthropic's three cyber-eval breakouts (July 30, 2026)