Particle.news

U.S. Export Controls Force Anthropic to Disable Its Two Most Advanced AI Models

Commerce officials said a reported jailbreak could expose powerful cybersecurity functions and the decision signals tighter U.S. controls over cloud‑hosted frontier AI.

Overview

  • Anthropic released Fable 5 and Mythos 5 and on June 12 the Commerce Department issued a directive barring access to the models by any foreign national, a move the company complied with by taking both models offline worldwide.
  • U.S. officials say Amazon researchers flagged a method to bypass Fable 5’s guardrails that the NSA judged could expose Mythos‑class cybersecurity capabilities, prompting the emergency export restriction signed by Commerce Secretary Howard Lutnick.
  • Anthropic disputes the government’s assessment, calling the reported vulnerability narrow, while some security researchers argue removing the models hurts defenders who use them to find and fix software flaws.
  • Allied governments have asked for exemptions and diplomats raised the issue at the G7, highlighting wider concerns about technological sovereignty and reliance on U.S. cloud‑hosted models.
  • Talks between Anthropic and U.S. agencies have continued without a deal to restore access and the standoff could reshape how frontier models are governed, shared with partners, and deployed by defenders and industry.