Anthropic's tightly restricted cyber model Mythos has reportedly been reached by an unauthorized group through a third-party vendor.

"The group has been using Mythos regularly since gaining access to it， and provided evidence to Bloomberg in the form of screenshots and a live demonstration of the software."

The technical issue is not that Anthropic's own core systems were reportedly breached， but that access control around a partner environment may have exposed a powerful security model whose value comes from helping experts find weaknesses faster than humans can.

Interestingly Anthropic's Project Glasswing was specifically built around limited distribution to stop misuse.

But this whole incidence shows that model secrecy is only as strong as the weakest contractor， endpoint， credential， and naming pattern around it.