Particle.news

U.S. Restricts Anthropic's Top Models After Reported Jailbreaks

The directive treats cloud-hosted AI as an export-controlled national-security item, pushing allies toward sovereign or open-weight alternatives.

Overview

  • The Commerce Department issued an export-control directive on June 12–13 that ordered Anthropic to stop foreign nationals from accessing its newest Claude models, Fable 5 and Mythos 5.
  • Anthropic disabled both models for all users because it said it could not reliably screen users by nationality in real time and defaulted customers to an older model.
  • U.S. officials cited reported jailbreaks that let researchers coax Fable 5 to reveal or fix software flaws, with reporting that Amazon researchers demonstrated prompts that bypassed guardrails.
  • Anthropic has entered fast negotiations with White House, Commerce and cybersecurity officials and has held talks with EU representatives while offering controlled-access proposals to European cyber authorities.
  • The action is reshaping markets and policy: it raises regulatory risk for Anthropic’s planned IPO, accelerates interest in open-weight and sovereign-deployable models, and spotlights experts’ warnings that perfect jailbreak prevention may be technically unattainable.