Anthropic’s Mythos Preview Used to Create macOS M5 Kernel Exploit in Five Days

A security team known as Calif has developed the first public macOS kernel memory corruption exploit targeting M5 silicon. The team utilized Mythos Preview, a tool from Anthropic, to bypass a security effort by Apple that spanned five years, completing the exploit in five days.

The exploit targets Memory Integrity Enforcement, or MIE, which is a hardware-assisted memory safety system. Apple designed MIE to make the execution of memory corruption exploits significantly more difficult.

Technical Foundation of MIE and MTE

MIE is built upon the Memory Tagging Extension, or MTE, a specification introduced by Arm in 2019. Apple described MTE as as a tool for hardware to help find memory corruption bugs.

At its core, MTE operates as a system for memory tagging and tag-checking. Within this system, every memory allocation is assigned a secret tag.

The hardware is designed to guarantee that requests to access memory are granted only if the request contains the correct secret. If the secrets do not match, the application crashes and the event is logged, allowing developers to identify memory corruption bugs as they happen.

Apple developed MIE because the company found that the original MTE specification was not robust enough under certain circumstances.

Hardware Integration and Impact

Apple integrated MIE into both the hardware and software of all iPhone Air and iPhone 17 models. The system was intended to detect and block specific memory corruption attacks before they could be successfully exploited.

The ability of the Calif team to bypass these protections on M5 silicon using Mythos Preview highlights a shift in the speed at which complex security measures can be analyzed, and overcome.