Anthropic has spent much of this week struggling to get its latest AI models back online after the Trump administration suddenly ordered the company to cut off access for all foreign nationals, including US users and its own employees, forcing Anthropic to block access to Fable 5 and Mythos 5 for everyone.
“To my knowledge, this is the first time that US export controls have been used in this way to control access to AI models.”
The Trump administration has not publicly explained the legal basis of the order, but in a statement on its website, Anthropic said the government cited “national security authorities” to justify the “export control directive” on the models. (Anthropic also claimed that government concerns about “jailbreaks” potentially being used by China-linked groups to access its models did not allow users to bypass all of the company’s security measures.)
But why did the administration use export control rules to address this? Experts say the episode appears to be unprecedented, highlighting an uncertain and unstable phase in AI governance. And what exactly should Anthropic export? (The company did not respond) The VergeRequest for comment.)
Export controls traditionally apply to things that can be shipped across borders: weapons, hardware, equipment, that kind of thing. Over time, the framework has expanded to cover less tangible objects, such as software, source code, technical data, and even 3D-printed gun files. These are still separate things that can be copied, downloaded, published, or otherwise assigned and taken, not just used through a remote service like a chatbot. In the context of AI, President Joe Biden moved to regulate AI model weights – the original data that powers a model that can be copied and run elsewhere – in this manner; This idea was rapidly abandoned by the Trump administration in its second term.
The anthropological system does not fit properly into this framework. There’s no obvious transfer happening: Mythos and Fable are hosted on Anthropic’s servers, and users don’t receive a copy of the source code, model weights, or the models themselves, instead they get the chatbot’s responses to their questions. There may be some specific information produced by exported models, but it is not clear why there would be a need to disable access to the entire system rather than restricting only a portion of it. It could also be the access itself – although remote access to cloud services is a known gap in existing export control regimes, which Congress is already trying to close by pushing legislation through the Senate.
Hannah Dohmen, senior research analyst at Georgetown University’s Center for Security and Emerging Technology, said The Verge It is “an open question” whether the order supersedes existing rules without looking at the exact language behind it. “In any case, this regulation is quite notable because, to my knowledge, it is the first time that US export controls have been used to control access to AI models in this way.”
“To say this is a volatile area of export control rule-making would be an understatement,” said Andrew Reddy, a professor at UC Berkeley’s Goldman School of Public Policy. He said other regimes, such as export control rules and arms regulations, give governments “wide latitude” to restrict access to certain goods. But “square-stepping by successive administrations about what the responsibilities of model developers are” has made it hard for companies to understand what is expected of them, he said.
Due to this the industry is in crisis. If Anthropic was targeted because Mythos and Fable are uniquely capable, then this order raises obvious questions for OpenAI, Google, Meta, XAI, and any other Frontier Labs’ next generation of models. If they were targeted because of specific security issues, the government needs to outline what protections it considers adequate. And if Anthropic was singled out because of its poor relationship with the Trump administration, the order becomes even more difficult to understand.
“This episode makes clear the instability of the current regime.”
Either way, experts say this is not a sustainable way to manage frontier AI, especially if the US wants to maintain its lead globally. The incident has already fueled arguments that governments and companies outside the US should be wary of trusting US companies for access to strategically important systems.
Reddy also had similar concerns. “In some ways, I think this episode highlights the instability of the current regime,” he said. This is especially true if the government were more concerned about whether users could jailbreak models and bypass their security measures. “If building models that are impossible to jailbreak becomes the de facto standard for the United States, it will have no AI models.”
All of this points to the same problem: The Trump administration wants to have both ways on AI. It has repeatedly said it wants to take a pragmatic approach and champion American technology, yet has forced a domestic champion to unceremoniously remove its frontier models through an order it has yet to publicly clarify. If Washington wants to control who can access powerful AI systems, it needs to explain how, and give companies a real chance to comply before launch. Ad hoc interventions made without thought are not sustainable in the long run – and they are a good way to ensure that the US gets left behind in the AI race.
<a href