Security Researchers Claim Breakthrough Exploit Against Apple’s M5 Chip

In a striking development for cybersecurity, a Vietnam-based security firm, Calif, claims to have successfully developed a macOS kernel exploit that bypasses Apple’s new Memory Integrity Enforcement (MIE) protections on M5 hardware. The breakthrough, achieved in less than a week, reportedly benefited from a preview version of Anthropic's Claude Mythos AI, which identified vulnerabilities during the exploit's development.

This announcement arrives as Apple devices are considered among the most secure consumer systems, thanks to the company's rigorous integration of hardware and software security measures. However, Calif's findings indicate that even the latest security advancements may not be immune to sophisticated attacks.

Details of the Exploit

Calif describes its creation as the first public macOS kernel memory corruption exploit capable of surviving MIE protections on Apple’s M5 chip. The company shared its findings directly with Apple during a meeting at the tech giant's California headquarters, a strategic move to ensure their work did not get overlooked amid the typical influx of submissions from researchers and hackers.

According to Calif, the exploit was developed after the accidental discovery of bugs on April 25, with the exploit chain becoming operational by May 1. This exploit targets macOS 26 on Apple M5 systems, starting from an unprivileged local user account and escalating to root access through standard system calls. Notably, the exploit combines two vulnerabilities with additional techniques aimed at the bare-metal M5 hardware while MIE is enabled.

Despite the significant role of Mythos Preview in identifying these vulnerabilities, Calif stressed the importance of human expertise to maneuver around Apple’s stringent MIE protections. The firm stated, "Part of our motivation was to test what’s possible when the best models are paired with experts." This collaboration between AI capabilities and human insight led to a noteworthy achievement: executing a kernel memory corruption exploit against advanced protections in just a week.

Implications for Cybersecurity

Memory corruption vulnerabilities have long been a primary method for attackers seeking to breach operating systems and applications. These types of bugs can enable attackers to crash programs, steal sensitive data, or gain complete control over a system. Apple's MIE feature employs memory-tagging technology designed to complicate these attacks, highlighting the challenges faced by cybersecurity professionals.

The implications of Calif’s exploit are significant, especially considering that Anthropic’s Mythos AI was released under restricted access through its Project Glasswing initiative. This initiative aims to provide select technology companies, banks, and researchers with access to advanced AI capabilities. Interestingly, the U.S. National Security Agency has also reportedly utilized Mythos, despite a contentious relationship with the Trump administration.

Calif's announcement has fueled speculation about the future of AI in cybersecurity. The firm characterized the exploit as a “glimpse of what is coming,” suggesting that traditional security measures may need to adapt in response to the capabilities demonstrated by Mythos. As Calif noted, “Apple built MIE in a world before Mythos Preview,” hinting at a looming shift in cybersecurity where AI models can autonomously identify and exploit software vulnerabilities.

Looking Ahead

Market sentiment regarding the public release of Claude Mythos remains cautious. Users on Myriad, a prediction market platform, currently assign only a 10.5% chance for a full public launch of Mythos by June 30. This hesitance reflects broader concerns about the ethical and security implications of releasing such powerful AI tools.

With the emergence of exploits like Calif’s, the discussion around AI’s role in both enhancing and undermining cybersecurity is becoming increasingly urgent. As organizations navigate these developments, it is clear that integrating advanced AI models into the cybersecurity framework will significantly shape the future of digital security. The ongoing evolution of both AI technology and cybersecurity measures is poised to usher in a new era of challenges and innovations in the tech landscape.