Security researchers have used Anthropic's Mythos AI model to uncover new macOS security vulnerabilities that Apple is now reviewing. The findings come from Calif, a Palo Alto-based security firm, which exploited techniques from testing an early version of the notoriously secretive AI system.

The researchers developed a privilege escalation exploit that links two bugs and memory-corruption techniques to access protected parts of a Mac. Calif delivered a 55-page report to Apple in Cupertino, and the company is validating the claims. A spokesperson said security remains Apple's top priority.

Calif CEO Thai Dong noted that Mythos alone couldn't have executed the attack — human expertise from his team was essential. He expects the bugs to be fixed quickly, though details will only be released after Apple patches the underlying issues.