Anthropic's Mythos helps discover macOS vulnerabilities

Researchers at Calif used Anthropic's unreleased Mythos model to find two previously unknown security holes in Apple's macOS. These bugs allowed for privilege escalation and memory protection bypass. The Calif team chained them into an exploit sequence capable of giving full device control.

A 55-page report has been sent to Apple, which confirmed receipt and is reviewing the findings. This is the first time a major OS vendor has received a vulnerability report generated with Mythos' help. Previously, the model found critical errors in OpenBSD and Linux.

Calif's CEO, Tai Duong, emphasized that AI alone was not enough. Mythos accelerates the search, but human expertise is still required to craft a working exploit. Anthropic keeps the model under wraps through Project Glasswing, citing risks to global infrastructure. We previously reported on Mythos helping the US banking sector uncover hidden threats in their systems.