The US Government Shut Down Anthropic's Best AI Models — and They're Still Offline on Day 7

On June 12, 2026 — just three days after Anthropic launched Claude Fable 5 and Mythos 5 — the US Commerce Department sent a letter to CEO Dario Amodei ordering the company to suspend both models for all foreign nationals. Anthropic couldn't filter users by citizenship in real time, so it did the only practical thing: it pulled the models for everyone, worldwide.

Seven days later, as of June 19, both models remain offline.

What triggered the order

The Commerce Department, led by Secretary Howard Lutnick, cited a jailbreak discovered shortly after launch that bypassed Fable 5's safety classifiers and exposed the full capabilities of the underlying Mythos 5 model — capabilities deliberately gated for vetted cybersecurity professionals through Project Glasswing. Fable 5 was designed to redirect sensitive requests (offensive hacking, dual-use biology, model distillation) to a less capable model. The jailbreak reportedly undid that routing entirely.

Anthropic pushed back. The company acknowledged the jailbreak but said the government had provided only "verbal evidence of a potential narrow, non-universal jailbreak" — not a confirmed, reproducible exploit affecting all users at scale. Anthropic argued that recalling a commercial model deployed to hundreds of millions of people over a narrow, theoretical attack surface was disproportionate. The government disagreed.

What's happened since

Negotiations between senior Anthropic engineers and Commerce Department officials have been ongoing in Washington. On June 18, Anthropic's international managing director said at the company's Seoul office opening that he was "very confident" both models would return "in the coming days." That confidence has not yet translated into a reactivation date.

The Seoul angle adds a wrinkle: the White House separately flagged SK Telecom — South Korea's largest wireless carrier and a $100 million Anthropic investor — as suspected of having ties to China, and asked Anthropic to revoke SK Telecom's access to Mythos 5. Anthropic complied immediately. That side conflict may be complicating the broader negotiation.

What Fable 5 was

Before the shutdown, Fable 5 was Anthropic's most capable generally available model: state-of-the-art on nearly all public benchmarks, priced at $10 per million input tokens and $50 per million output tokens — less than half the cost of the previous Mythos Preview tier. It featured million-token context windows, strong autonomous task execution, and safety routing designed to keep the most dangerous capabilities locked away. Claude Opus 4.8 and all other Anthropic models remained online throughout the shutdown.

The bigger question

This is the first time a US administration has issued an emergency order pulling a commercial AI model offline for national security reasons. Whatever the resolution looks like, the precedent is set: governments can and will intervene directly when they believe an AI system poses a risk they can't tolerate — and the speed of that intervention can outpace any company's ability to respond technically. For now, Fable 5 users are on Opus 4.8. The clock is ticking on Anthropic's "coming days" promise.