An advanced AI model from Anthropic named Mythos has been restricted after it reportedly exposed critical vulnerabilities in sensitive U.S. government computer systems during a security test.
A U.S. official speaking anonymously stated that the tool breached nearly all classified networks “not in weeks but in hours.” The testing occurred under Project Glasswing, which was announced on April 7 as an urgent initiative to apply the defensive capabilities of the frontier model Mythos Preview.
The official clarified that while vulnerabilities were identified through this exercise, they were not exploited. Project Glasswing was described at the time as “an urgent attempt to put [the capabilities of frontier model Mythos Preview] to work for defensive purposes.”
