Calif Uses Anthropic AI to Find macOS Kernel Exploit

Security researchers at Palo Alto-based firm Calif have used an early version of Anthropic's Claude Mythos AI model to discover a novel kernel memory corruption exploit in Apple's macOS, marking the first such public vulnerability reported for the operating system's Apple silicon chips.

The exploit, detailed in a Calif blog post published Thursday, chains two vulnerabilities with advanced techniques to enable an unprivileged local user to gain complete control over a device. Researchers credited Mythos Preview, available only to select partners through Anthropic's Project Glasswing initiative, for rapidly identifying the bugs. "Mythos Preview is powerful once it has learned how to attack a class of problems; it generalizes to nearly any problem in that class," the post stated. "Mythos discovered the bugs quickly because they belong to known bug classes." The team emphasized that human expertise was essential alongside the AI for developing the full exploit chain.

Calif researchers delivered a detailed report to Apple headquarters in Cupertino earlier this week and plan to release full technical details after Apple addresses the issues. Apple's release notes for macOS Tahoe 26.5,...