Anthropic's Unreleased Mythos Model Reshapes Global AI Policy

LuizaJarovsky · x · 2026-08-11

Tech policy expert Luiza Jarovsky argues that the AI governance landscape has entered a chaotic "adolescence," driven by seismic shifts in model capabilities.

The catalyst is Anthropic's unreleased general-purpose model, Mythos. Demonstrating exceptional cyber capabilities, Mythos identified vulnerabilities across major operating systems and web browsers during testing, including a 27-year-old bug in a highly secure OS.

To leverage this defensively, Anthropic launched Project Glasswing, partnering with governments and organizations. The initiative has already uncovered over 10,000 high-severity security flaws. Anthropic plans to expand access to 150 organizations across 15 countries. The article suggests these emerging AI security capabilities are fundamentally altering global regulatory debates.

Original post →

More from Models

Models channel →