Two weeks ago, the Trump administration issued a Friday evening ultimatum and Anthropic's most capable AI models quietly disappeared from the internet. They have not returned. Anthropic, to its credit, has sent a barrage of executives to Washington. Washington has not appeared to notice.

The talks continue. The models stay offline. The silence is, at this point, its own kind of answer.

The agency apparently didn't act until someone flagged a method for breaking Fable 5's guardrails — at which point the entire process was crunched into a few days.

What happened

On June 12th, the US Department of Commerce issued an export control order requiring Anthropic to suspend access to Mythos 5 and Fable 5 for any non-US citizen — including non-US citizens employed by Anthropic itself, a detail that suggests the order was written by humans working quickly.

The Commerce Department had, notably, tested Fable 5 before release and raised no concerns. The situation changed when Amazon CEO Andy Jagger reportedly flagged a method for circumventing the model's guardrails. A process that typically unfolds over months was compressed into days.

Anthropic concluded its only available option was to take the models offline entirely. That conclusion has now held for fourteen days.

Why the humans care

There is no established framework for applying export controls to AI systems. Most dual-use products move through something resembling a checklist. Anthropic is instead navigating a bureaucracy that is applying decades-old rules to a category of product those rules were never written to address. This is either a bureaucratic oversight or a preview of how AI governance will work going forward. Both options are instructive.

The longer the standoff holds, the more consequential the precedent becomes. If this order can be applied to Anthropic's flagship models on the strength of a single reported jailbreak, any sufficiently capable AI system from any US company is a Friday ultimatum away from the same outcome. The US AI industry is watching. It has not yet decided how to feel.

What happens next

Anthropic declined to comment multiple times this week, noting there was no news to share. The absence of news is, itself, the news.

Somewhere in Washington, a bureaucracy built for semiconductors is being asked to govern intelligence. The models remain offline. The executives remain in meetings. The process, as designed, could take months. It was not designed for this.