A security team called Calif used Anthropic's Mythos Preview to bypass Apple's Memory Integrity Enforcement on M5 Macs in just five days. The exploit chains two vulnerabilities and several techniques to corrupt kernel memory and gain root access — a feat that reportedly took Apple years and billions to build. Calif shared a 20-second video demonstrating the attack, which targets macOS 26.4.1 on bare-metal M5 hardware.

Apple developed MIE as a hardware-assisted memory safety system built on Arm's MTE specification. The company claims MIE disrupts every public exploit chain against modern iOS, including leaked Coruna and Darksword kits. Calif's team discovered the bugs on April 25th, built tooling within days, and had a working exploit by May 1st — all with Mythos Preview assisting in bug identification and exploit development.

The team credited the AI model's ability to generalize across known bug classes, though autonomously bypassing MIE still required human expertise. A 55-page technical report exists but won't be released until Apple ships a fix. The discovery earned Calif a visit to Apple Park, where they shared their findings directly with the company.

Apple's MIE was built before Mythos Preview existed, and Calif warned this kind of AI-assisted exploit development represents a new reality for security mitigations. Even with the best protections on Earth, small teams pairing top AI models with human expertise can break through in days.