Humanity has been affected by export rules that no one understands

Anthropic has spent much of this week fighting to get its latest AI models back online after the Trump administration He ordered suddenly The company has cut off access to all foreign nationals, including users within the United States and its employees, forcing Anthropic to block access to Fable 5 and Mythos 5 to everyone.

The Trump administration has not publicly clarified the legal basis for this, but in a statement on its website, Anthropic He said The government cited “national security authorities” to justify the “Export Control Directive” on the forms. (Anthropic also claimed that government concerns about “Escape from prison“It is likely to be used by groups linked to China Access to their models It does not allow users to circumvent all company warranties.)

But why did the administration use export control rules to address this matter? Experts say that this incident seems unprecedented, revealing an uncertain and unstable stage in the management of artificial intelligence. And what exactly is Anthropic supposed to produce? (The company did not respond to EdgeComment request.)

Export controls have traditionally been applied to things that can be shipped across borders: weapons, devices, tools, etc. Over time, the framework expanded to include less tangible goods, such as software, source code, technical data, and even… 3D printed weapons files. These are still separate things that can be copied, downloaded, posted, delivered and taken, and not simply used by a remote service like a chatbot. In the context of artificial intelligence, President Joe Biden Move To control the weights of AI models – the underlying data that makes the model work and which can be copied and run elsewhere – in this way; That was the idea It was quickly abandoned By the Trump administration in the second term.

The humanitarian system does not fit neatly into this framework. There’s no straightforward transfer process: Mythos and Fable remain hosted on Anthropic’s servers, and users don’t receive source code, model weights, or a copy of the model themselves, instead getting chatbot responses to their queries. The export could be some specific information produced by the models, but it is not clear why this would require disabling access to the entire system rather than restricting only part of it. It could also be the access itself — although remote access to cloud services is a known gap in existing export control systems, one that Congress is already trying to solve. Closes Through legislation now moving through the Senate.

said Hannah Dohmen, senior research analyst at Georgetown University’s Center for Security and Emerging Technology Edge It is an “open question” whether the system is pushing existing rules without seeing the precise language behind it. “In any case, this regulation is very notable because, to my knowledge, this is the first time that US export controls have been used to control access to an AI model in this way.”

“To say this is an unstable area in terms of export control rulemaking would be an understatement,” said Andrew Reidy, a professor at the Goldman School of Public Policy at the University of California, Berkeley. He said export control rules and other regulations such as arms regulations give the government “broad latitude” to restrict access to certain goods. But he said “confusion by successive administrations regarding the responsibilities of model developers” had made it difficult for companies to understand what was expected of them.

This leaves the industry in a dilemma. If Anthropic is targeted because Mythos and Fable are uniquely capable, it raises obvious questions for the next generation of models from OpenAI, Google, Meta, xAI, and any other frontier lab. If they are targeted because of specific prevention issues, the government must specify what type of protection it considers adequate. And if the anthropy is assigned because of it The tense relationship with the Trump administrationit becomes more difficult to understand.

Either way, experts say this is not a sustainable way to manage frontier AI, especially if the United States wants to maintain its global leadership. the The incident has already added fuel to the arguments Governments and companies outside the United States should be careful about relying on American companies to access strategically important systems.

Reddy had similar concerns. “In some ways, I think this event demonstrates the unsustainability of the current system of governance,” he said. This is especially true if the government is more concerned about whether users can jailbreak models and bypass their safeguards. “If creating impossible-to-jailbreak models becomes the de facto standard for the United States, it will not have AI models.”

All of this points to the same problem: The Trump administration wants to do both when it comes to AI. I have He said over and over again It wants to adopt a hands-off approach and support American technology, but it has forced a local hero to unceremoniously snatch up its border models through something it has yet to explain publicly. If Washington wants to control who can access powerful AI systems, it must determine how, and give companies a real chance to comply before launch. Ad hoc interventions implemented on a whim appear to be unsustainable in the long term — a good way to ensure the United States falls behind in the AI ​​race.