The US government has ordered Anthropic to cut off access to its most capable models.

On June 12, 2026, the government issued an export-control directive on national security grounds, requiring Anthropic to suspend access to Claude Fable 5 and Claude Mythos 5. Export controls have long applied to advanced semiconductors and military-related technology. This is the first time that framework has reached a commercial large language model already used by hundreds of millions of people.

For anyone responsible for adoption decisions, the directive carries a plain message: dependence on a specific model can be severed for reasons that have nothing to do with how the technology performs. What follows sets out what happened, why it stands as a precedent, and what it implies for model continuity.

What Happened

Anthropic received the directive at 5:21 pm Eastern Time on June 12.

According to the company's public statement, the government invoked national security authorities to issue an export-control directive suspending all access to Fable 5 and Mythos 5 by any foreign national. The initial target was foreign nationals — people of foreign citizenship inside or outside the United States, including foreign-national Anthropic employees.

Anthropic cannot screen users by nationality in real time and block only foreign nationals. To ensure compliance with the directive, it disabled both models for every customer. The scope of the order (foreign nationals) and the scope of the actual impact (all customers) are therefore different, and that gap sits at the center of the story.

The suspension is limited to Fable 5 and Mythos 5. Access to all other Anthropic models continues unaffected.

What Triggered It

The trigger was a claim about a jailbreak — a bypass of the model's safety controls.

By Anthropic's account, another company claimed it had bypassed Mythos's safety controls, and the administration grew concerned in response. The technique Anthropic actually reviewed was a narrow one that exploits a known, minor weakness, and is not universal. The company says comparable capability is widely available from competitors' models as well.

The proper names come from reporting rather than from the official statement. According to Axios, Commerce Secretary Howard Lutnick sent a letter to Anthropic CEO Dario Amodei stating that the two models would be subject to export controls. Those names do not appear in Anthropic's own statement. The directive itself, the time of receipt, and its stated scope come from Anthropic.

Anthropic's Position

Anthropic is complying with the directive while publicly disagreeing with the decision.

In its statement, the company says that the discovery of a narrow potential jailbreak should not be cause for recalling a commercial model already in wide deployment. It warns that applying this standard across the industry could effectively halt the release of every new frontier model.

Anthropic frames its objection not as a denial of the government's authority to act, but as a question of process. It is calling for a transparent, fair, statutory process grounded in technical facts, and says it is working to restore access as soon as possible.

Why It Reads as a Precedent

The weight of this case lies in the novelty of the framework applied.

Export controls have historically targeted physical goods and foundational technologies — advanced chips, hardware, military-related systems. Here the same instrument has been used against software, and against a commercial LLM already offered broadly to the public. It is a concrete demonstration that a government can intervene directly in who is allowed to use a particular AI model.

The precedent reaches beyond the labs that build frontier models. Conditions such as a customer's nationality, an end user's citizenship, and where an engineer is based could all become factors that shape how a product can be shipped.

What It Implies for Adoption

A final word for those who embed these models in their operations.

A specific model was switched off all at once, for reasons unrelated to its technical evaluation, and only shortly after launch. The stronger a model performs, the more tempting it is to place at the core of a workflow — yet the chance that such dependence is severed by a regulatory or policy decision belongs on the ledger as an availability risk.

How is model continuity to be guaranteed? How much reliance on a single model is acceptable, and what fallback paths are prepared in advance? Alongside performance and cost, there is now one more reason to weigh continuity of supply as a selection criterion.

Should AI procurement be decided on performance alone?

Source: Selected by the editorial team from the AI news inbox (collected June 13, 2026 — special report). Primary information is Anthropic's official statement; reporting draws on Axios, CNBC, and NBC.