Anthropic Mythos AI Cracks Apple M5 Kernel in 5 Days
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Apple's Memory Integrity Enforcement (MIE) on M5 chips, a hardware‑assisted memory safety system built on ARM's MTE, was bypassed by a local privilege escalation exploit.
- Researchers Calif, Bruce Dang, Dion Blazakis, and Josh Maine built a working kernel LPE for macOS 26.4.1 (25E253) on bare‑metal M5 hardware, delivering a full root shell from an unprivileged account using only standard system calls.
- Anthropic’s Mythos Preview AI model helped identify the two underlying bugs and assisted the exploit development, allowing the team to go from discovery on April 25 to a working exploit by May 1.
- Apple received a 55‑page printed report delivered to Apple Park rather than a bug‑bounty submission, and is reportedly working on a patch; until then, M5 systems remain at theoretical risk.
- Memory corruption remains the most prevalent vulnerability class across modern platforms, and the exploit shows that even multi‑year, multi‑billion‑dollar mitigations like MIE can be overcome in days when AI assistance is used.
Why it matters: Apple must urgently patch its M5 hardware, while AI‑augmented security teams gain a fast track to break high‑cost mitigations, reshaping the defensive‑offensive balance and compelling vendors to rethink protection strategies as AI‑driven exploits become feasible across platforms in days for attackers.
