Anthropic's Mythos Model Breaks Apple M5 Chip's MIE Defense in Five Days

According to security research team Calif, they used Anthropic's unreleased Mythos Preview model to successfully build the first public exploitation chain for a macOS kernel memory corruption vulnerability on Mac devices with M5 chips, in just five days from vulnerability discovery to successful exploit. Apple spent five years and billions of dollars developing MIE (Memory Integrity Enforcement), a flagship hardware security mechanism for M5 and A19 chips, designed to significantly raise the cost of exploiting memory corruption flaws rather than provide absolute immunity.

The attack chain consists of two vulnerabilities and multiple techniques, escalating from an unprivileged local user to root access through standard system calls on a naked M5 hardware running macOS 26.4.1 with kernel MIE enabled. Mythos excels at generalizing attack classes after learning them, helping the team rapidly identify defects within known vulnerability categories, while human experts then bypassed the new hardware defenses. The breakthrough demonstrates how AI-assisted vulnerability discovery combined with expert exploitation can overcome significant technological barriers built by large organizations.