Particle.news

U.S. Order Stops Anthropic’s Top AI Models After Tests Found Vulnerabilities

The decision forces a debate over protecting classified networks versus preserving defenders’ access to frontier AI.

Overview

  • The Commerce Department issued a national security directive that barred foreign‑person access to Anthropic’s Fable 5 and Mythos 5, a restriction the company says it complied with by disabling the models worldwide.
  • Project Glasswing tests with U.S. agencies used Anthropic’s Mythos and identified software vulnerabilities in highly sensitive government systems within hours, a finding confirmed to the Associated Press though officials said identification did not mean immediate exploitation.
  • The government order that restricted access was issued June 12 and has prompted industry, legal and diplomatic pushback as companies and allies seek limited, controlled access for defensive work.
  • Legion LegalTech filed a federal lawsuit on June 23 challenging the directive, arguing the shutdown caused immediate harm to its operations because Canadian engineers lost access to a contracted model.
  • A public Five Eyes warning and a letter from more than 100 cybersecurity leaders have raised opposing concerns that restrictions may both reduce immediate risk from misuse and weaken U.S. cyber defenses by taking leading vulnerability‑finding tools out of defenders’ hands.