AzarxAI | Building with AI

Anthropic has announced Claude Mythos Preview, its most powerful AI model to date, but is withholding it from public release due to its extraordinary cybersecurity capabilities.

CEO Dario Amodel says the model was never explicitly trained for cybersecurity — rather, its cyber abilities emerged as a side effect of its coding excellence: "We haven't trained it specifically to be good at cyber, we trained it to be good at code. But as a side effect of being good at code it's good at cyber too."

The result has been striking. Mythos Preview autonomously discovered thousands of high- and critical-severity zero-day vulnerabilities across every major operating system and web browser including a 27-year-old flaw in OpenBSD and a 16-year-old bug in FFmpeg, without being explicitly designed to do so.

Given the dual-use risks, Anthropic launched "Project Glasswing", restricting access to roughly 40 select organizations including Google, Microsoft, Apple, Amazon, and CrowdStrike. The goal is to use Mythos defensively — finding and fixing critical vulnerabilities before bad actors can exploit them. Anthropic has committed up to $100 million in usage credits to support the effort.