Anthropic's Mythos AI model — designed to find operating system and browser vulnerabilities — was accessed by unauthorized hackers through a contractor's credentials

A small group of hackers gained access to Anthropic's Mythos AI model, a tool built to identify security flaws in major operating systems and web browsers, by exploiting credentials belonging to a third-party contractor working with Anthropic and using standard hacking techniques. The breach was disclosed to Bloomberg by one of the unauthorized users.

Mythos is designed as a general-purpose AI model with the explicit capability to discover and exploit vulnerabilities (weaknesses that could be used to break into systems) across Windows, macOS, Linux, Chrome, Firefox, Safari, and other major platforms — making it exactly the kind of tool that could cause widespread damage if misused.

For everyday users and business professionals, this breach means attackers may now have a powerful tool to find zero-day exploits (previously unknown security holes) in the software you use daily before those flaws are patched — potentially putting your devices, accounts, and work data at greater risk. For security teams, this incident signals that even AI safety-conscious companies can face credential theft from contractors with system access.