AI-Assisted Research Uncovers Apple M5 Root Exploit and 16 Windows Flaws

Researchers from Calif disclosed a local privilege-escalation exploit against Apple systems using M5 silicon that reportedly bypasses the chip’s Memory Integrity Enforcement (MIE) protections and grants a root shell on macOS 26.4.1 from an unprivileged user context. The team said Anthropic’s Mythos Preview helped chain two previously unknown bugs and other techniques into a working exploit in five days, using only standard system calls. Apple was reportedly notified in advance and is working on a fix. The finding is significant because MIE was introduced to harden M5 and A19 devices against memory-corruption attacks such as buffer overflows and use-after-free flaws.

Microsoft separately detailed MDASH, an internal AI-driven vulnerability discovery platform that coordinates more than 100 specialized agents and has already found 16 previously unknown vulnerabilities in Windows networking and authentication components. Four of those flaws, including critical remote code execution bugs in tcpip.sys and the IKEv2 service, were patched in the latest Patch Tuesday release. Microsoft said MDASH also identified issues in IPsec and Netlogon, including vulnerabilities reachable over the network without credentials, and described the system as a production-grade defensive capability now in internal use and limited private preview.