Anthropic’s safety warnings may have just backfired — the government has pulled the plug on its most powerful AI
TL;DR
The US government ordered Anthropic on June 12 to disable Claude Fable 5 and Mythos 5. The order is framed around export controls and national security, but the practical effect was a worldwide shutdown for both models. Mythos 5 was Anthropic’s most capable, tightly restricted security model. Fable 5 was the guarded commercial version and had launched only three days before the shutdown.
Nauti's Take
Safety transparency is no longer just good PR — it can become the kill switch. If you build models, you need to turn jailbreak findings into clear controls, timelines, and limits regulators can understand.
Otherwise the safest model does not win; the least attackable narrative does.
Briefingshow
This shows how blurry the line between safety testing, export controls, and product shutdowns is becoming for frontier models. If a narrow jailbreak can trigger a recall-style action, providers cannot confidently plan stable commercial rollouts. For companies, model access is now a regulatory continuity risk, not just a technical or pricing decision.
Tweets
Sources