Anthropic investigates unauthorized access to Mythos, an AI model that enables hacking

Mythos is an AI model designed to help people identify security vulnerabilities. Someone accessed Mythos without authorization. The breach was detected. An investigation began and is now ongoing, which means it has not concluded.

This follows the pattern where tools designed to break systems get broken into by people who break systems. The symmetry is obvious. No one has filed a report suggesting this is ironic. The review of what happened during the unauthorized access is also ongoing, meaning nothing has been determined about what was taken or changed.

The investigation will continue. The review will continue. Both processes are adequate containment of the situation until they are not.