Anthropic Mythos helped Calif build a macOS exploit in five days

Key Points:

• Security researchers at Calif used Anthropic’s Mythos Preview AI model to develop the first public macOS kernel memory corruption exploit on Apple’s M5 silicon, bypassing Apple’s five-year Memory Integrity Enforcement (MIE) security system in just five days.
• MIE, introduced by Apple and built on Arm’s Memory Tagging Extension (MTE), is a hardware-assisted memory safety system designed to detect and block memory corruption exploits, initially focused on iOS and recently extended to MacBooks with the M5 chip.
• The exploit targets macOS 26.4.1, uses two vulnerabilities and several techniques to escalate privileges from an unprivileged user to root, and was developed through a combination of AI assistance from Mythos Preview and expert human collaboration.
• Calif plans to withhold their full 55-page technical report until Apple issues a fix but has already shared their findings directly with Apple at Apple Park, highlighting the evolving challenge of securing systems in an era where AI accelerates vulnerability discovery.
• The researchers emphasize that Apple’s current security mitigations were designed before AI tools like Mythos Preview, signaling a new phase of cybersecurity challenges as AI-powered bug discovery becomes more prevalent.