Calif Team Utilizes Anthropic Mythos to Develop macOS Exploit in Five Days
A team of security researchers from Calif has successfully developed a macOS kernel memory corruption exploit on M5 silicon using Anthropic's Mythos Preview model. This exploit was achieved in just five days, bypassing Apple's Memory Integrity Enforcement (MIE), a system designed to prevent such vulnerabilities. MIE, based on Arm's Memory Tagging Extension, was introduced by Apple to enhance memory safety by tagging memory allocations with a secret to prevent unauthorized access. The Calif team discovered two bugs and employed several techniques to gain unauthorized access to parts of the Mac's memory, which should have been protected by MIE. The exploit starts from an unprivileged local user and ends with a root shell, targeting macOS 26.4.1 on M5 hardware. 
