Trump Admin Objects to Anthropic Mythos Expansion

The Trump administration has objected to Anthropic's plan to expand access to its advanced cybersecurity AI model, Mythos, from about 50 organizations to roughly 120, citing national security risks and insufficient computing resources that could impair government use. At the same time, the White House is preparing an executive action to enable federal agencies to resume using Anthropic's models, bypassing a Pentagon supply chain risk designation imposed earlier this year.

Anthropic announced Mythos in early April as part of its Project Glasswing initiative, restricting access to a select group for testing on their own systems. The model demonstrated exceptional capabilities, autonomously identifying thousands of zero-day vulnerabilities across major operating systems and web browsers, succeeding on 73 expert-level capture-the-flag tasks, and completing a 32-step simulated corporate network attack. Organizations like Mozilla reported using it to patch 271 vulnerabilities in the Firefox browser during testing.

Administration officials expressed concerns over potential misuse of Mythos for cyberattacks, as well as operational issues from limited compute capacity. The National...