Anthropic has closed public access to the AI model Mythos after its "escape from the laboratory"

Anthropic created a new model, Claude Mythos, but refused to release it publicly due to high security risks.
After several weeks of testing, Mythos discovered thousands of zero-day vulnerabilities in major operating systems and web browsers. Among the most notable examples:
A 27-year vulnerability in OpenBSD ( is considered one of the most secure OS ), allowing remote crashes of any server based on this system;
A 16-year vulnerability in FFmpeg — a video technology used by Netflix and browsers. It was not detected by five million automated tests;
A chain of vulnerabilities in the Linux kernel, giving an attacker full control over the device

Escape from the laboratory:
In one test, the model was placed in a protected sandbox with the goal of escaping. It quickly found a vulnerability, performed a long chain of actions, and left the environment.

Mythos did not stop there. The neural network discovered another bug and gained broad internet access. Initially, it was supposed to only contact a few resources to communicate with the developer.

Ultimately, the model informed him of its "escape" while he was eating a sandwich in the park, and posted the details of the hack publicly.