The most advanced large-scale models are starting to be subject to export controls, similar to enriched uranium.

Last Friday, a letter took the world's two most powerful AIs offline simultaneously. An export control order from the US Department of Commerce prohibited any foreign citizens from accessing Anthropic's Fable 5 and Mythos 5 models. Since it was impossible to identify users' nationalities in real - time, Anthropic did the only thing it could: shut down these two models, which had been released just three days ago, for everyone globally.

For the first time, humanity has included an intelligent entity existing in a digital form in the same export control framework as enriched uranium.

The scenario of enriched uranium is repeating itself

Historically, export controls have only been applied to two types of things: hardware and formulas. As we know, these include uranium enrichment centrifuges, high - end lithography machines, military - grade encryption algorithms, and so on.

Their common feature is physical scarcity. Without the equipment and blueprints, you can't produce them. Export controls are effective because the controlled items have physical boundaries, can be intercepted at customs, and can be traced in the supply chain.

However, Fable 5 is a set of weight parameters. It can be copied infinitely, doesn't pass through customs, doesn't require containers, and there is no physical action of "smuggling". Theoretically, it can be activated on any server in the world. The US Department of Commerce found that all traditional tools were ineffective. It couldn't stop Fable 5 at the border and had to cut it off at the source.

The real object of control is the "capability density" condensed in this set of weight parameters: code generation ability, reasoning and planning ability, and cross - domain knowledge invocation ability. When these abilities were scattered in the minds of countless engineers, they never triggered any controls. But when they are compressed into a model, and when a person can invoke all these abilities with a single prompt, the compression itself poses a threat.

This is an exact mapping of the enriched uranium logic in the digital world.

Uranium ore is widespread in the earth's crust and has never been controlled. But when it is enriched to a certain level, it becomes the most closely monitored substance on earth. The same goes for models: when individual capabilities are scattered in open - source code libraries, technical Q&A, and academic papers, they are free. When all these capabilities are "concentrated" into a single - point callable interface, it crosses the threshold, and the price of crossing the threshold is to be shut down.

The history of enriched uranium provides a mirror for understanding this matter.

In 1938, Hahn and Strassmann discovered nuclear fission in Berlin. In the following decade, uranium changed from an obscure laboratory element to the most sensitive strategic material on earth. In 1946, the US passed the Atomic Energy Act, bringing all nuclear technologies, nuclear materials, and nuclear knowledge under government control. Private capital was excluded from the nuclear energy field, international exchanges among scientists were cut off, and even basic physical data was classified as confidential. A pure natural element was then shackled by politics.

Eighty years ago, that control was based on a cold - blooded logic: some forces are too powerful to be in the hands of any entity that doesn't take national interests as the ultimate consideration. Eighty years later, the same logic may be activated again, with the target shifting from nuclear fission to the forward propagation of neural networks.

Three things will happen in the next decade

The control of enriched uranium in the 1950s gave birth to a brand - new international governance structure: the International Atomic Energy Agency, the Nuclear Non - Proliferation Treaty, and the Supplier Group. Once technical control is established, it will irreversibly move towards institutionalization, multilateralization, and permanence.

AI will not be an exception. In the next decade, three things are likely to happen.

1. Capability review will become institutionalized.

Before the release of each new cutting - edge model, it will not only undergo security red - team testing but also third - party compliance review authorized by the government. The review standards will not come from within the company.

The evaluation of model capabilities will change from "benchmarking" to a "checklist". Each item on the checklist that may be misused will trigger additional control requirements. The "abundance" of the model - parameter volume, reasoning depth, and cross - domain generalization ability - will be accurately measured and thresholds will be set, just like the concentration of uranium - 235. Models that exceed a certain threshold will automatically trigger export control clauses.

2. Jurisdictional boundaries will become blurred.

The order for Fable 5 targets "foreign citizens" regardless of their geographical location. For the first time, the control tentacles of the US government have extended to every user in the world.

A developer in Singapore using the API of a US company is subject to US export control laws. A company in Berlin, the compliance obligation of its AI supplier depends not on German law but on an order from the US Department of Commerce.

This unilateral expansion of jurisdiction will force non - US companies to rethink their AI supply chains: the US suppliers you rely on may be required by the US government to cut off your access one day. The Fable 5 incident gives a clear answer: it can.

3. The technical path will split.

When closed - source cutting - edge models face the risk of repeated shutdowns, the global AI industry will be forced to adopt a dual - track system.

One track is the US closed - source cutting - edge models, which are subject to export controls and face the risk of being taken offline with each release. The other track consists of open - source models, locally deployed models, and models in non - US jurisdictions. They may not be as advanced as the former, but they are not threatened by the US government's shutdown.

The market share of open - source models will be driven not only by performance but also by the security attribute of "not being shut down". In the past three years, open - source models have been catching up with closed - source models in terms of capabilities. In the next decade, open - source models may gain a structural advantage over closed - source models in terms of "reliability".

The deepest rift lies in the property rights system

All the above inferences are based on a fundamental question that remains unanswered. The deepest crisis exposed by the Fable 5 incident is that digital civilization has not yet established a property rights system for "intelligence".

Legally, models are sold as a service. You pay, and I use my assets to do things for you. My assets always belong to me, and what you buy is only its output. This logic has worked in traditional service industries for thousands of years without any problems.

But AI is different. After your company has spent three months optimizing all internal tools based on the specific behavior patterns of Fable 5, training employees, and writing hundreds of automated scripts that rely on the specific output format of the model, you have, in fact, turned Fable 5 into your means of production.

But legally, it is still a service provided by Anthropic. It can be taken back at any time, and the compensation you can get does not exceed the subscription fee you paid in the past month.

You have made real input of means of production but only get service - level legal protection. The gap between the two is the loss suffered by all global corporate customers when Fable 5 was taken offline. This loss does not appear on any balance sheet, does not trigger any insurance claims, and is not covered by any legal provisions.

Humanity has spent three hundred years establishing a legal system for "property". A piece of land, a factory, a patent all have clear property rights, trading rules, and dispute arbitration mechanisms. But this system has a default premise: property is tangible or at least has a traceable carrier.

When a model is shut down, you can do nothing. It is neither stolen nor destroyed. It still exists, but you are not allowed to use it. This is a new form of deprivation: what is deprived is not the thing itself but the right to use, which has never belonged to you legally.

Enriched uranium has been controlled for eighty years and remains the most sensitive technology asset of humanity. The control of AI has just begun, and its end may be a permanently divided digital world. In this world, the smartest models may not be the ones that can be used. The models that can be used must be those with the clearest property rights. Not being taken away is much more important than being temporarily ahead at a certain historical juncture.

This article is written based on public information and is only for information exchange, not constituting any investment advice

This article is from the WeChat official account "Jinduan" (ID: jinduan006), written by Yuantai and published by 36Kr with authorization.