Anthropic has spent much of this week fighting to restore access to its newest AI models after the Trump administration abruptly ordered the company to cut access for all foreign nationals, a directive that forced Anthropic to block Fable 5 and Mythos 5 for everyone, including users inside the United States and the company’s own employees. The order, which the government did not publicly explain, cites national security authorities as the basis for what it calls an export control directive on the models. But the legal framework being invoked was never designed for this scenario, and the episode exposes a widening gap between existing export control rules and the reality of how frontier AI systems are actually delivered and used.
An Unprecedented Use of Export Controls on AI Models
To my knowledge, this is the first time US export controls have been used to control access to an AI model in this way, said Hanna Dohmen, a senior research analyst at Georgetown University’s Center for Security and Emerging Technology. The statement, delivered without the precise language of the order in hand, captures the core uncertainty that now confronts the entire frontier AI industry. Export controls have traditionally applied to things that can be shipped across borders: weapons, hardware, tools, and, over time, less tangible goods such as software, source code, technical data, and even 3D-printed gun files. These are discrete items that can be copied, downloaded, published, or otherwise handed over and taken. A chatbot hosted on remote servers does not fit that description.
The Biden administration had moved to control AI model weights in this manner, a policy that was swiftly abandoned by the Trump administration in its second term. The Anthropic order does not fit neatly into that framework either. Mythos and Fable remain hosted on Anthropic’s servers. Users do not receive source code, model weights, or a copy of the model itself. They receive the chatbot’s responses to their queries. The export, if there is one, could be specific information produced by the models, though it is not clear why that would require disabling access to the entire system rather than restricting part of it. It could also be access itself, though remote access to cloud services is a known gap in current export control regimes, one that Congress is already trying to close through legislation now moving through the Senate.
What the Government Hasn’t Said
The Trump administration has not publicly explained the legal basis for the order. In a statement on its website, Anthropic said the government cited national security authorities to justify an export control directive on the models. Anthropic also claimed that the government’s concerns about a jailbreak potentially used by groups linked to China to access its models did not allow users to circumvent all of the company’s safeguards. The combination of an opaque legal rationale and a lack of publicly available detail leaves the industry guessing about what precisely triggered the action and what it signals for the future.
To say that this is an unsettled area of export control rule-making would be an understatement, said Andrew Reddie, a professor at UC Berkeley’s Goldman School of Public Policy. Export control rules and other regimes like arms regulations give the government wide latitude to restrict access to certain goods. But the equivocation by successive administrations regarding what the responsibilities of model developers are has made it hard for firms to understand what is expected of them. This uncertainty is not merely academic: it directly affects how companies design, train, and deploy their most advanced systems.
The Governance Vacuum at the Heart of Frontier AI
If Anthropic was targeted because Mythos and Fable are uniquely capable, the order raises obvious questions for the next generation of models from OpenAI, Google, Meta, xAI, and any other frontier lab. If they were targeted because of specific safeguard issues, the government needs to outline what protection it considers sufficient. And if Anthropic was singled out because of a testy relationship with the Trump administration, the order becomes even harder to make sense of as a policy matter.
Either way, the current approach is not a sustainable way to manage frontier AI, especially if the United States wants to maintain its global lead. The incident has already added fuel to arguments that governments and companies outside the US should be wary about relying on American firms for access to strategically important systems. Reddie had similar concerns. In some ways, this episode makes clear the unsustainability of the existing governance regime, he said. That is especially true if the government was more concerned about whether users could jailbreak models and bypass their safeguards. If creating models that are impossible to jailbreak becomes the de facto standard for the United States, then it will have no AI models.
What Are Export Controls and How Do They Apply to AI Models?
Export controls are legal frameworks that restrict the transfer of certain goods, technology, or information across national borders. They have traditionally applied to physical items such as weapons and hardware, but have expanded over time to cover software, source code, technical data, and other intangible goods. In the context of AI, the question is whether providing remote access to a hosted model constitutes an export the same way shipping model weights or source code would. This remains an open legal and regulatory question, and the Anthropic order is the first time export controls have been used to restrict access to an AI model delivered as a cloud service.
An Industry in a Bind
The implications extend well beyond Anthropic. If the government can order a company to disable global access to a frontier model without a publicly articulated standard, every lab operating at the cutting edge faces the same risk. The Trump administration has repeatedly said it wants to take a hands-off approach and champion American technology, yet it forced a domestic champion to unceremoniously yank its frontier models through an order it has still not publicly explained. If Washington wants to control who can access powerful AI systems, it needs to say how and give companies an actual chance of complying before launch. Ad hoc interventions seemingly delivered on a whim are not sustainable in the long run, and they are a good way to make sure the United States falls behind in the AI race.
The absence of clear, predictable regulation creates a chilling effect on investment and innovation. Companies cannot plan their product roadmaps or their safety deployments if the rules of the road can change overnight. International partners and customers cannot rely on access to American AI systems if that access can be revoked without warning or explanation. The current uncertainty benefits no one except those who argue that the US is not a trustworthy supplier of critical AI capabilities.
What This Means for Developers and Decision-Makers
The immediate takeaway for anyone building on or relying on frontier AI systems is that the governance landscape has entered a period of acute instability. The Anthropic order is not an isolated incident: it is a signal that the existing regulatory framework is being stretched beyond its original design, and that the government is willing to act unilaterally and opaquely. Anyone evaluating AI suppliers should consider not just model capability and price, but the regulatory and geopolitical risks that come with depending on a system that can be turned off by government order. For developers inside frontier labs, the message is equally stark: jailbreak resistance and safeguard robustness are no longer just safety engineering challenges. They have become compliance requirements with potentially existential business consequences. The episode makes clear that the industry needs a new regulatory framework, one that is transparent, predictable, and designed for the realities of cloud-delivered AI. Until that framework exists, every frontier model launch carries the risk of being the next target.
