Anthropic’s early‑stage model Mythos has slipped past macOS’s built‑in defenses, according to security researchers. The AI linked two separate bugs to corrupt memory and then accessed normally sealed parts of the system. The discovery came after researchers drove to Apple’s Cupertino headquarters to present the hack in early week of May 2026 report details confirm.

Anthropic, which plans to release a more powerful Claude version later this year, warned that Mythos is too skilled at hunting exploits for public use. The incident underscores how AI can accelerate vulnerability discovery, forcing vendors to tighten patch cycles. It also highlights the growing arms race between defensive teams and autonomous code‑finding tools today.

Apple’s spokesperson said it is reviewing the findings and takes security seriously. No public patch has yet been released, leaving uncertainty over whether the exploited bugs are fixed. Analysts note that Mythos alone could not execute the full attack without human intervention, suggesting a hybrid threat model that blends AI automation with traditional hacking skills.

Given the Mac’s market share, this breach could affect hundreds of thousands of users. Apple’s swift response will set a precedent for how vendors address AI‑driven exploits, potentially reshaping patch management practices across the industry.