Particle.news
Download on the App Store

U.S. Orders Anthropic to Secure Licenses for Top AI Models and Company Disables Access

Commerce said the models pose an unacceptable military‑intelligence risk, with talks under way over licensing and allied access.

Overview

  • Citing a June 12–13 directive, the Commerce Department’s Bureau of Industry and Security told Anthropic it must obtain individually validated export licenses for Claude Mythos 5 and Claude Fable 5 for all foreign persons worldwide.
  • Anthropic shut off global access to both models because it could not reliably block foreign‑person users in real time and chose full suspension to ensure compliance with the licensing demand.
  • Reporting traced the immediate trigger to Amazon researchers who reportedly used prompts that bypassed Fable 5’s guardrails, and BIS said that ability created an unacceptable risk of diversion to military‑intelligence use.
  • G7 leaders pushed for mechanisms to restore allied access under shared safeguards and a coalition of more than 150 cybersecurity experts publicly urged the administration to lift or clarify the restriction to avoid weakening defenses.
  • Legal and policy specialists say using the Export Administration Regulations to control cloud‑hosted API access is untested and likely to face legal challenge, a development that could reshape how frontier AI is governed and how defenders access security tools.