Particle.news

U.S. Orders Anthropic to Block Foreign Access to Two Claude Models

Citing a reported way to “unlock” the models, the U.S. directive forced a global suspension and signals tighter export-style controls on frontier AI.

Overview

  • The U.S. government issued a directive on Friday that bars any foreign citizen, including foreign Anthropic employees, from accessing Claude Fable 5 and Mythos 5.
  • Anthropic responded by deactivating both models for all customers to comply with the order and said the directive lacked detailed technical justification.
  • U.S. officials flagged a reported technique that could bypass Fable 5’s protections and surface software vulnerabilities, while Anthropic says tests found only a limited set of known flaws.
  • Investors warned the sudden restriction has raised regulatory risk for Anthropic’s planned IPO and leaders in allied countries called for more diversity in AI supply chains.
  • The move deepens a policy clash between Anthropic and U.S. defense agencies over military and surveillance use and could set a precedent for export-like limits on advanced AI models, so talks and technical disclosures in Washington will be closely watched.