Security researchers at Calif have demonstrated a working kernel memory corruption exploit against Apple’s M5 silicon, bypassing the company’s five-year, billion-dollar Memory Integrity Enforcement (MIE) system in just five days with assistance from Anthropic’s Mythos Preview AI model.
Breaking Apple’s Hardware Memory Safety
Apple introduced MIE last year as a hardware-assisted memory safety system built on Arm’s Memory Tagging Extension (MTE) specification. MTE works by assigning a secret tag to every memory allocation; the hardware then verifies that any subsequent memory access request contains the correct tag, crashing the application if a mismatch occurs.
Apple determined that MTE was insufficiently robust under certain conditions and developed MIE as a proprietary enhancement, embedding it into the iPhone 17 series and, more recently, the M5-powered MacBook lineup. Apple’s own research claimed MIE disrupts every public exploit chain against modern iOS, including the leaked Coruna and Darksword exploit kits.
How Mythos Preview Accelerated the Attack
The Calif team disclosed that their macOS attack path was discovered accidentally. Bruce Dang identified the underlying bugs on April 25, Dion Blazakis joined the effort on April 27, and Josh Maine built the tooling, resulting in a working exploit by May 1. The exploit is a data-only kernel local privilege escalation chain targeting macOS 26.4.1, starting from an unprivileged local user and ending with a root shell.
According to the researchers, Mythos Preview played a critical role in both identifying the vulnerabilities and assisting throughout the collaborative development process. The model generalizes effectively across known bug classes, enabling rapid discovery, while human expertise remains essential for bypassing novel mitigations like MIE. The team has produced a 55-page technical report but will not release it until Apple ships a fix.
Implications for Enterprise Security
The Calif team met with Apple at Apple Park to share their vulnerability research directly. They noted that MIE was designed “in a world before Mythos Preview,” and that even small teams, when paired with advanced AI, can now defeat best-in-class mitigations within a week.
This development signals a fundamental shift in the threat landscape. As AI-assisted exploit development becomes more accessible, enterprise security teams must reconsider the durability of hardware-based memory protections. The question is no longer whether such defenses can be broken, but how quickly—and what comes next when the next AI-powered “bugmageddon” arrives.
— Originally reported by 9to5Mac. Adapted and republished with editorial context for MacThreat.
