Officials say Anthropic refused to fix Fable 5. Anthropic disputes the account.

When a US export-control directive forced Anthropic to pull its two most capable models offline on Friday, the company's account was the only one on the record. Over the weekend, that changed, and a one-sided story became an open dispute.

The June 12 order forced Anthropic to suspend Claude Fable 5 and Mythos 5 for all users1; we covered it that night. The government had said nothing then. Now administration officials have gone public with a version Anthropic sharply contests.

The government's case: David Sacks, the White House AI czar, laid out the administration's account on X, embedded below. By his telling, a trusted partner discovered a working jailbreak of Fable's guardrails; the administration asked Anthropic to fix it or pull the model; and CEO Dario Amodei refused.2

In the same thread, Sacks questioned how Anthropic could deem a jailbreak that made a cyber weapon operable anything but serious, and accused the company of putting its consumer product ahead of safety. His full post runs below.

Anthropic's rebuttal: the company has not backed down. In its public statement, it said the disclosed jailbreaks were "either entirely benign responses or are minor findings that provide no Mythos-specific uplift," and that the government had given it only "verbal evidence" of a narrow jailbreak before the directive landed. It calls the episode a misunderstanding and says it is working to restore access.3

The feud is older than the directive. Hours after the suspension, Defense Secretary Pete Hegseth posted that his department had barred Anthropic from its building three months ago, "forever," and that "every passing day proves why that was the right move."4 It reframes the export action as the latest round in a running clash between the administration and Anthropic, not a bolt from the blue.

Why it matters: the dispute is no longer about whether a model misbehaved. It is about who decides what counts as "serious," on what evidence, and whether a company that lobbied for the government's power to block dangerous models can object when that power is turned on it. Both sides are now using the same word — safety — to mean opposite things.

What we can confirm: the Sacks and Hegseth statements are on their verified accounts and embedded above, and Anthropic's rebuttal is quoted from its published statement; those are first-hand. The Amazon and China threads below come secondhand, from other outlets, and we flag them as unverified by us. The directive itself remains unpublished.

What we still don't know: which agency issued the order and under what authority; what the government believes the jailbreak enables; and when, or whether, the models come back. Outlets including TechCrunch and Axios have reported the "trusted partner" was Amazon;5 Semafor has tied the urgency to concern over Chinese access to Mythos.6 Neither is confirmed on the record, and we have not verified either. As of the weekend, the models had not been restored.

Day one belonged to Anthropic's telling. By the weekend the government had a telling of its own: Sacks's accusation and Hegseth's older grievance, with Anthropic disputing all of it. The facts that would settle it are still inside a letter no one has published.