Particle.news
Download on the App Store

US Lifts Export Controls on Anthropic’s Fable 5 and Mythos 5

The move follows Anthropic’s addition of a targeted safety classifier and a pact to let U.S. agencies test models before release, a step that could reshape how frontier AI is vetted.

Overview

  • The Department of Commerce withdrew export controls on Tuesday, June 30, 2026, and Anthropic said it will begin restoring global access to Claude Fable 5 and Mythos 5 starting July 1.
  • The controls were introduced after a reported prompt‑based jailbreak showed a way to get Fable 5 to reveal software vulnerabilities, prompting the company to suspend both models on June 12.
  • Anthropic told officials it trained a new safety classifier to detect and block the reported bypass method and to route flagged requests to a less capable model, and U.S. reviewers tested those mitigations.
  • Before the full lifting, Commerce allowed Mythos 5 to be used by about 100 vetted U.S. organizations for defensive testing, and Anthropic agreed to notify agencies about misuse and cooperate on standards.
  • The episode establishes a new, government‑involved pathway for pre‑release review of powerful models, raising questions about allied access, industry rollout practices, and how future jailbreaks will be classified and handled.