Researchers successfully bypassed Apple's Memory Integrity Enforcement (MIE) security feature on M5 silicon using an AI-assisted exploit that chains two unknown bugs via standard system calls, achieving a root shell from an unprivileged local user. The exploit targets macOS 26.4.1, and while Apple is aware and working on a fix, no specific patch version or workaround is provided in the article.
AI/ML Researchers bypass Apple’s M5 security with AI-powered macOS exploit May 15, 2026 Share By SC Staff (Credit: ink drop – stock.adobe.com) As outlined in Tech Radar, cybersecurity researchers have successfully created a macOS kernel memory corruption exploit targeting Apple's M5 silicon, bypassing a recently implemented security feature. The exploit was developed using Anthropic's Mythos Preview AI tool, highlighting the growing capabilities of AI in discovering complex software vulnerabilities. Researchers from Calif utilized Anthropic's Mythos Preview AI to chain two previously unknown bugs and several techniques, ultimately creating a functional exploit for Apple's M5 chip. This exploit successfully bypassed the new Memory Integrity Enforcement (MIE) feature, which Apple introduced in September 2025 to prevent memory corruption attacks. Despite Apple's significant investment in MIE, the Calif team achieved a root shell within five days of using Mythos. The exploit targets macOS 26.4.1 and operates from an unprivileged local user, leveraging only standard system calls. This development underscores the potential of AI tools like Mythos to uncover sophisticated flaws in hardware and software, even those protected by advanced security measures. Apple is reportedly aware of the vulnerability and is working on a fix. Source: Tech Radar An In-Depth Guide to AI Get essential knowledge and practical strategies to use AI to better your security program. Learn More SC Staff Related AI/ML Permiso launches AI agent runtime security SC Staff May 15, 2026 The new capabilities extend Permiso’s unified identity platform to address the growing challenge of securing AI agents, which are increasingly making autonomous decisions and interacting with various systems without direct human oversight. AI/ML The blind spot in AI security: Resilience for model and training data Chris Mierzwa May 15, 2026 AI resilience, not speed alone, will decide which enterprises survive AI-era breaches. AI/ML SecurityScorecard acquires Driftnet to enhance AI-driven risk management SC Staff May 14, 2026 The acquisition of Driftnet will see its internet scanning engine integrated into SecurityScorecard's TITAN AI platform. Get daily email updates SC Media's daily must-read of the most current and pressing daily news Business Email By clicking the Subscribe button below, you agree to SC Media Terms of Use and Privacy Policy . Subscribe You can skip this ad in 5 seconds