The iPhone and Mac security Apple spent 5 years building? AI broke it in 5 days.

Key Points:

• Anthropic’s Mythos AI helped the security startup Calif bypass Apple’s Memory Integrity Enforcement (MIE), a hardware security feature in M5 chips, by finding and exploiting bugs in just five days, despite Apple spending five years developing MIE.
• MIE protects memory by tagging allocations with secret codes to prevent unauthorized access, and Apple claims it can disrupt all known exploit chains, but Calif demonstrated a data-only kernel local escalation exploit that grants full control over macOS 26.4.1 machines even with MIE active.
• Mythos AI played an active role in identifying vulnerabilities and assisting in exploit development, showcasing rapid pattern recognition and generalization capabilities, though human judgment was still essential to complete the exploit.
• Apple is aware of the vulnerability and has engaged with Calif to address the issue urgently, but no patch has been released yet; Calif plans to publish a detailed technical report after Apple begins deploying a fix.
• The incident highlights the dual-edged impact of AI in cybersecurity, accelerating both vulnerability discovery and defense, and marks a significant challenge for existing security technologies in the era of advanced AI tools.