Cathedral’s agents broke into Anthropic in under 12 hours

Security start-up Cathedral’s AI agents reached Anthropic’s Slack messages and salary details in under 12 hours, per a draft report reviewed by The Washington Post. Anthropic confirmed access to a limited set of sales, HR and IT systems, but said customer chats with Claude, source code and model weights stayed out of reach.
The agents went through business software Anthropic uses for admin and sales, in a test under its outside-researcher program. Anthropic said it cut off access the same day after its monitoring flagged some of the activity.
Anthropic says Cathedral broke the program’s rules by not disclosing all its activity. Cathedral says it kept data access to the minimum.
Up next
Claude Code won’t rebuild existing games, even renamed
“Do not build a recreation of a specific existing game, even if it is renamed or called original,” Claude Code’s new rule says. Genre games stay allowed. Has Claude ever refused a project of yours that was actually original?
Related stories
- Fake coverage report stole keys via Claude Code Desktop
- Claude Duet never sees your passwords, even when it signs in
- Anthropic test agents reached a State Department visa form
- Local MCP servers run outside Claude’s VM sandbox
- An AI agent found a hole in a gym’s booking system, then used it
- Claude gave up its sandbox binary and process memory
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.

