Particle.news
Download on the App Store
3 ARTICLES
·
6d ago

U.S. Order Forced Anthropic to Pull Two Top AI Models Then Approve a Tied Safety Fix

The Commerce-led review established U.S. authority to vet cloud AI by requiring a targeted safety classifier and operational limits on who may use the most powerful model.

The return of Anthropic's Claude Fable 5 to global availability on July 1 is the result of two weeks of Washington negotiations.
Image
Image

Overview

  • On June 12 the Commerce Department issued an export-control directive after Amazon reported a prompt-based jailbreak, prompting Anthropic to take Mythos 5 and Fable 5 offline worldwide until government concerns were resolved.
  • Anthropic developed and deployed a targeted safety classifier that blocked the reported jailbreak in more than 99% of test cases and agreed to faster reporting of vulnerabilities and pre-release government review.
  • The government’s Center for AI Standards and Innovation and other agencies independently tested and approved the new safeguards before lifting controls, restoring global access to Claude Fable 5 on Wednesday, July 1.
  • Mythos 5 returned with tighter limits on Friday, June 26 when access was reopened only to vetted U.S. agencies and companies, leaving allied access and long-term distribution rules unresolved.
  • The episode disrupted enterprise workflows, raised diplomatic and investor concerns about U.S. control over cloud AI, and accelerated industry work on shared jailbreak scoring, researcher reporting channels, and alternatives to centralized models.