Anthropic Blocked Its Own Models After US Export Order Hit Foreign Users
The Trump administration ordered Anthropic to cut AI model access for all foreign nationals, including US-based users and staff. Here is what happened and why it matters.
LUMIEN4 min read
The Trump administration ordered Anthropic this week to cut off access to its newest AI models for all foreign nationals, including people physically located inside the United States and Anthropic's own employees. To comply, Anthropic blocked its Fable 5 and Mythos 5 models for everyone. The government cited national security authorities but has not publicly released the legal basis for the order. According to reporting by The Verge, one expert described it as the first time US export controls have been applied to restrict access to an AI model in this way.
What happened
Anthropic’s two newest AI models, Fable 5 and Mythos 5, went dark for users this week after the Trump administration issued an order requiring the company to terminate access for all foreign nationals. The directive covered foreign nationals anywhere, including those living or working inside the US, and it included Anthropic’s own employees who fall into that category.
To comply with the order, Anthropic had no clean surgical option. The company blocked both models for all users while it worked through the week to figure out how to restore access on compliant terms. Access was not restored quickly. Anthropic spent most of the week dealing with the fallout.
In a statement published on its own website, Anthropic said the government used “national security authorities” to justify what it described as “an export control” on its models. The Trump administration has not offered a public explanation of the specific legal authority behind the order.
Why it matters
Export controls are not new. The US government has used them for decades to restrict the sale or transfer of hardware, software, and technology with military or strategic value. Applying them to access to an AI model is a different thing entirely.
According to at least one expert cited in The Verge’s reporting, this appears to be the first time export controls have been used specifically to control who can access an AI model. If that precedent holds, it has significant consequences for any AI company operating in the US that serves international users, which is most of them.
Consider what this means in practice:
- A foreign national working at a US company could be cut off from AI tools their colleagues use.
- A company could be forced to shut down access to a product for all users in order to comply with an order targeting a subset of users.
- The legal basis for such orders may not be publicly disclosed, leaving companies little time to prepare or appeal.
The abrupt nature of the order is also worth noting. Anthropic was not given apparent advance warning that allowed it to build a targeted compliance system before the cutoff. The result was a blunt, total block.
Our take
From where we sit, this is a genuinely new kind of compliance risk for AI products, and most businesses using AI tools are not thinking about it yet.
If you run a product or workflow built on top of a third-party AI API, you now have a concrete example of that API going dark with little notice, for regulatory reasons that were not publicly explained in advance. That is not a hypothetical vendor risk. It happened this week to one of the largest AI labs in the world.
The fact that the legal basis has not been disclosed makes this harder, not easier, to plan around. Anthropic could not have published a compliance checklist ahead of time because the government did not publish one either. That is a problem that contracts and SLAs do not solve.
We are also skeptical that the “national security authorities” framing will stay narrow. Once a legal tool is used once, it tends to get used again. If AI model access can be shut down via export control order, that is a lever that exists now, regardless of which administration is in office next.
What to do about it
You probably cannot fully insulate yourself from this kind of regulatory action, but you can reduce exposure:
- Audit your AI dependencies. Know which models and APIs your core workflows rely on, and which have no fallback.
- Test a backup model now. If Fable 5 or Mythos 5 were your primary models, do you have a tested alternative ready? If not, add one to your stack and validate it against your actual use cases.
- Read vendor communications carefully. Anthropic published a statement on its website. If you are not subscribed to updates from your AI vendors, you may find out about outages from users, not the vendor.
- Talk to legal if you operate internationally. If your product serves users in multiple countries and relies on US-based AI APIs, the intersection of export controls and AI access is now a real compliance question.
The practical takeaway: treat your AI API dependencies the same way you treat any single-point-of-failure infrastructure. Plan for it going away, because this week proved it can.
