Researchers use Mythos Preview AI to crack Apple M5's MIE in 5 days
Apple spent billions on M5 security; a team with Mythos Preview bypassed it in under a week.
In a major security milestone, researchers working with the Mythos Preview AI system have developed the first public macOS kernel memory corruption exploit that survives Apple's MIE (Memory Integrity Enforcement) on M5 chips. Apple spent five years and billions building MIE, a hardware-assisted memory safety system based on ARM's MTE, specifically designed to stop such exploits. The team discovered two vulnerabilities on April 25 and had a working local privilege escalation chain by May 1, using only normal system calls to gain root from an unprivileged user on macOS 26.4.1.
The exploit was an accidental discovery during broader AI-assisted vulnerability research. Mythos Preview generalized from known bug classes to find the flaws quickly, while human experts guided the MIE bypass. This demo shows that even the strongest hardware mitigations can be cracked when AI amplifies attacker capabilities. The researchers will publish a full 55-page report after Apple fixes the issue. The work signals a new era where AI systems can find and exploit bugs that defeat current best defenses.
- First public macOS kernel memory corruption exploit on M5 hardware with MIE enabled
- Built in 5 days from bugs to working root shell, using only normal system calls
- Mythos Preview AI discovered bugs by generalizing from known classes; humans handled MTE bypass
Why It Matters
Demonstrates AI-assisted exploit development can defeat Apple's flagship hardware security, forcing a rethink of mitigation strategies.