Anthropic’s Mythos AI Uncovers Critical macOS Security Flaws

Security researchers have successfully used Anthropic’s powerful new AI model, known as Mythos, to uncover critical vulnerabilities within Apple’s macOS, according to a report by The Wall Street Journal.

The research was conducted by Calif, a security firm based in Palo Alto. By leveraging an early version of the Claude Mythos Preview model, the team was able to identify a sophisticated exploit chain that targets the memory of Apple’s desktop software.

Specifically, the researchers developed software that links two distinct bugs together to bypass standard security protocols. This technique allowed the team to achieve what is known as a “privilege escalation exploit.” Essentially, it gives a user or a program access to restricted parts of the operating system that should be completely off-limits.

Anthropic has been famously cautious with the release of Mythos. The company has previously stated that the model is so effective at finding software flaws that it poses a potential risk to global digital infrastructure if released publicly. To manage this risk, Anthropic launched Project Glasswing, a controlled initiative that gives select partners, including Apple, Microsoft, and Google, access to the tool for defensive purposes.

The team delivered a 55-page report to Apple’s headquarters in Cupertino to ensure the company could address the issues. Apple, which prides itself on the robust security of its ecosystem, is currently reviewing the findings. An Apple spokesperson reiterated that security is their top priority and that they take such reports seriously.

The discovery in macOS is a testament to the model’s capabilities. Mythos has already reportedly uncovered thousands of high-severity vulnerabilities across every major operating system and web browser. In some cases, it has found bugs that have survived decades of human audits and traditional automated testing.