On June 13, 2026, the U.S. Commerce Department ordered Anthropic to shut down access to two of its most powerful AI models—Fable 5 and Mythos 5—everywhere in the world. Anthropic complied by Friday. AP News reported this is the strongest government restriction on AI model access to date.

The formal order came in a letter from the Trump administration. It specifically required Anthropic to prevent foreign nationals from accessing the models. The Washington Post later reported that the relationship between Anthropic and the White House had broken down before this order arrived.

Why This Is Hard to Enforce

Stoppingpeople from using software over the internet is different from stopping the spread of physical goods. When the U.S. restricts computer chips, it can intercept them at factories or ports. With an AI model delivered online, the company providing it has to do the blocking itself. Anthropic chose to disable the models entirely rather than try to figure out who is or isn't allowed to use them—a simpler and more reliable approach.

The Policy Background

This did not happen by surprise. President Trump signed an executive order in July 2025 aimed at controlling AI exports as a national security matter. That order gave the government legal grounds to restrict access to AI models the way it restricts semiconductors or weapons technology.

Anthropic itself had asked for these kinds of restrictions before the government imposed them. In April 2025, the company urged the government to limit sales of advanced AI chips and model weights worldwide, saying this was important to American strength. The company also updated its sales terms in September 2025 to prevent companies controlled by restricted countries from buying its products.

There is something worth noticing: Anthropic pushed for the exact type of restriction that the government then applied to Anthropic itself. Whether the company expected this to happen is not clear.

Who Gets Hurt

Companies and researchers outside the U.S. lost access immediately. The European Commission said on June 14 it is looking into what this means for European users. European businesses using these models in their work were suddenly cut off, which is likely to make European governments more concerned about relying on American AI companies.

This situation is similar to what happened with semiconductor restrictions, which got stricter step by step starting in 2022. The difference is that AI models are harder to work around because the company itself has to enforce the ban—there is no way to buy them on the black market or reroute them through other countries.

One potential problem: the government named specific model versions instead of setting rules based on how powerful the AI is. This means Anthropic will probably need a new directive for each new version of Fable or Mythos that comes out, unless the government creates clearer rules.

Anythingorganization worldwide that was relying on these models now has to find a replacement quickly. More broadly, it raises a question: will the U.S. government routinely use export controls on AI models, or was this a one-time action. The answer will affect how AI companies build and sell their products globally.