
AI Breaks Five-Year Apple Security in Five Days
Anthropic’s Mythos Preview helped Calif researchers uncover and weaponize a data-only kernel escalation against Apple’s Memory Integrity Enforcement (MIE) on the M5, delivering a working exploit in five days after Apple spent five years building the defense. The flaw can bypass MIE even when active on macOS 26.4.1, with human experts guiding the exploit’s design. Apple says vulnerabilities are taken seriously and is working on a patch, while Calif plans to publish a 55-page technical report; the episode highlights how agentic AI can speed up security discoveries, for better or worse.


