Apple Mac M5 System Exploited With Anthropic's Claude Mythos AI, Researchers Claim
A security startup, Calif, used Anthropic’s Claude Mythos AI to develop a working macOS kernel memory corruption exploit against Apple’s new M5 chip protections in under a week. This exploit bypasses Apple’s Memory Integrity Enforcement (MIE), combining two vulnerabilities and additional techniques to escalate privileges from a local user account to root on macOS 26 with MIE enabled. Calif accidentally discovered the attack path in late April and subsequently met with Apple to share the findings. The company credits Mythos Preview for aiding vulnerability identification and exploit development, but notes that human expertise was necessary to bypass MIE. Apple’s MIE uses memory-tagging to defend against memory corruption attacks, which remain a common threat. Anthropic’s Mythos, released only to select partners and not the public, reportedly possesses autonomous vulnerability identification capabilities surpassing previous AI models. Other tests show Mythos uncovers numerous vulnerabilities and can simulate multi-stage cyberattacks. Calif warns that this Apple M5 exploit previews the heightened challenge traditional software mitigations face in the era of advanced AI-assisted hacking.
