Claude Mythos uncovers macOS kernel exploit

The discovery marks a "qualitative leap" in AI's ability to autonomously weaponize zero-day vulnerabilities in hardened production systems. Mythos identified the underlying bug and generated the necessary scaffolding to link two kernel-level memory corruption flaws, successfully bypassing MIE protections on M5 systems—a feat previously considered beyond the scope of current AI models. This incident validates Anthropic’s "Project Glasswing" defensive strategy, which restricts access to Mythos to a vetted group of tech partners like Apple and Microsoft. For security teams, this signals the end of "security through obscurity" as high-reasoning models can now scan and exploit legacy patterns across OS-level codebases in seconds.