Particle.news

U.S. Export Control Forces Anthropic to Disable Its Top AI Models

Officials cited national-security risks after a reported demonstration suggested the models' safeguards could be bypassed, which may drive allies to seek more regional AI options.

Overview

  • Anthropic disabled its highest-performance models, Mythos 5 and Fable 5, for all users in mid-June after receiving a U.S. export-control directive and opened a limited refund process with a reported June 20 deadline.
  • U.S. officials privately urged Anthropic not to deploy the new models before the public action and cited a reported demonstration to the Commerce Department that showed a way to break Fable 5; some reports identified the demonstrator as Amazon while Anthropic disputes the scope of the finding.
  • Anthropic has begun issuing refunds but users report partial reimbursements, loss of upgraded access and confusing rules for App Store purchases that require refunds through Apple rather than Anthropic.
  • The European Commission said it is assessing effects and warned emergency measures should not discriminate against partners, and Canada's prime minister publicly urged diversification away from dependence on a few dominant models.
  • The episode raises fresh questions about applying export controls to commercial AI, risks to Anthropic's planned IPO and whether the models can be fixed and restored, with key technical details and final regulatory decisions still unresolved.