AI-Driven Exploit Bypasses Apple's M5 Security in Days

Researchers from Calif, a Palo Alto-based cybersecurity startup, successfully exploited Apple's Memory Integrity Enforcement (MIE) on the M5 chip within a week using Anthropic's AI model, Claude Mythos. The exploit involved linking two bugs and several techniques to achieve kernel memory corruption, allowing unauthorized access to restricted system areas. The exploit is notable as it is the first public demonstration of a macOS kernel memory corruption bypassing MIE protections. Calif's team discovered the initial vulnerabilities on April 25, 2026, and had a working exploit by May 1, 2026. They reported their findings directly to Apple, emphasizing the role of human expertise alongside AI in developing the exploit. The incident raises concerns about the accelerating capabilities of AI in offensive cybersecurity. Apple is currently investigating the reported vulnerabilities and is expected to release a patch soon.

Key Points: • Calif exploited Apple's M5 chip security in under a week using AI assistance. • The exploit is the first public demonstration of bypassing Apple's Memory Integrity Enforcement. • Human expertise was crucial in conjunction with AI for developing the exploit.