Cathedral AI agents breached Anthropic systems in under 12 hours

The disclosure shifts the security conversation from model behavior to the software supply chain: the entry points were third-party tools, not Anthropic's core systems, which is where most enterprise exposure actually sits. The absence of named vendors and any stated notification to Anthropic leaves the claim difficult to verify independently, and that gap matters for a startup whose founders' government background invites scrutiny of its methods. Whether Cathedral discloses the vendors or Anthropic confirms the findings is the open question; until then, the precedent of public breach-style announcements without coordinated disclosure is the more consequential signal for AI vendors and their customers.