AI agents accessed Anthropic systems via third-party software
Original title: AI agents hacked Anthropic, maker of arguably the most capable AI models for cyberdefense
Anthropic says sales, HR, and IT data were exposed, but Claude chats, source code, and weights were not.
Cybersecurity startup Cathedral says AI agents entered Anthropic through external admin and sales programs during a bug-bounty engagement and reached internal business data. Anthropic confirmed limited access to sales, human resources, and IT systems, including some customer and partner contact information, then shut the paths the same day, disabled the third-party programs, and notified US authorities. The sides dispute how long access lasted, whether a second flaw was withheld, and how much Slack data was readable.
Unconfirmed: Duration, Slack admin access, and whether Cathedral broke bounty rules remain disputed.
- Cathedral says agents reached transcripts, salaries, job posts, and Slack in under twelve hours.
- Anthropic says probing lasted days and that it found and patched a second flaw itself.
- Will Depue claims full Slack admin access; Anthropic’s Logan Graham disputes that.
- Chats with Claude, source code, and model weights were not exposed, Anthropic said.
Impact: Limited sales, HR, IT, and contact data exposed
- Products
- Slack
- Victims
- Anthropic
- Organizations
- Cathedral
- Countries
- United States
- AI models
- Claude
From the original article
3 min read at the-decoder.com
Cybersecurity startup Cathedral used AI agents to break into Anthropic's internal systems through third-party software as part of a bug bounty program. Anthropic confirmed access to sales, HR, and IT data but disputes how deep the breach went. The incident shows how vulnerable even leading AI labs are through their external software dependencies.
By Matthias Bastian · the-decoder.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.