When the United States Switches Off the AI

A few days ago Anthropic, the company that owns the Claude artificial intelligence, released its Fable model. This model is supposedly designed to improve the capabilities of generative AI, allowing users to perform far more sophisticated tasks in less time—that is, it increases productivity in certain tasks. The United States government has responded by prohibiting foreign citizens from using the model, citing national security reasons. In what is so far the latest development, Anthropic has withdrawn the model from everyone.

This move by the U.S. government has been placed within what some have called a “digital cold war,” and it can only be understood in the context of the decline of the American power’s hegemonic position. Just as Trump’s tariff policies seek to protect American industry from the effects of free trade—which is now assumed to benefit adversary countries, especially China—restricting the use of the most advanced AI to domestic users is an attempt to seize any opportunity to gain an economic advantage. Both measures are part of a protectionist package that some of us call neomercantilism, and for which there are historical precedents that can help us understand the current situation.

Perhaps the paradigmatic example was eighteenth-century England, which was beginning to surge ahead thanks to its mechanics’ ability to build more effective and efficient machines for the textile industry. So important was this knowledge that, from the start of the century, the English government banned the emigration of skilled workers. From the second half of the century onward, the government extended the ban to the machines themselves, already built, so that rivals could not replicate them and use them for their own economic development. The mercantilist policy of the era held that this advantage in knowledge and technology should be kept “private” in order to provide an edge in the economic struggle; there was as yet no such thing as free trade, which the United Kingdom adopted only a century later, when it was already the undisputed economic power.

This was not a strategy of England alone, but also of the Netherlands and some Italian republics such as Venice—that is, of those more advanced regions where industrialisation was beginning to take firmer hold. Nevertheless, there is consensus that most of those measures were not very effective. The reason is that the lagging countries, such as France, Prussia, or the United States, responded by combining (illegally) policies to lure away workers with industrial espionage. For example, the American Samuel Slater, known in England as “Slater the traitor,” emigrated to the British Isles disguised as a farmhand in order to memorise Arkwright’s spinning machinery; thanks to that, the United States acquired the technology that for decades had given the United Kingdom its key economic advantage.

The current pattern is similar. Although there are academic debates on the matter, it is obvious to anyone that AI does raise productivity in certain tasks (for example, in processing and manipulating large amounts of data or drafting documents). If such a function were available only to the United States, it would grant a notable advantage to its companies compared with their competitors. Although the U.S. government cites “National Security” as the reason for its decision, this is a typical way of disguising mercantilist policies that ultimately seek an economic advantage for its companies.

We should add that the United States’ generative artificial intelligence programs are not entirely its own. As Albert Einstein recalled in “Why Socialism?”, no individual achievement arises from nothing: each person’s life is made possible thanks to the work and achievements of the many millions, past and present, and it is society that provides language, the forms of thought, and most of their content. In this sense, all knowledge is a collective and intergenerational inheritance. Just as the machines of the industrial revolution embodied the resources that were expropriated and plundered from the colonies and the world’s periphery—for without them it was impossible to conceive of building the artefacts in question—generative AI embodies knowledge that is, by default, common property. Even all our data, including my own articles, is appropriated by generative AI both to train it and to construct the results offered to the user. Thus, one of the questions of our time might be: who controls and appropriates knowledge of common origin?

At the same time, the whole process depends on a network of infrastructure that is built asymmetrically according to the logic of capital—that is, heading toward those places where nature and human beings are most accessible, most pliable, and, above all, cheapest. This is what Cecilia Rikap, in her superb book “A Theory of Digital Dependency,” has called “twin extractivism.” The result is a process that concentrates profits in a few actors (the large AI corporations) and distributes the economic, social, and ecological costs across peripheral regions. And the on/off switch for the AI’s essential functions still resides in the center, whether the United States or China.

The case described here highlights a critical vulnerability for the rest of the world’s countries, including European ones, which already depend on the artificial intelligence services of American companies. Although many countries have plunged into a struggle to attract AI data centers, the truth is that this infrastructure—which is, moreover, highly intensive in energy and water consumption—is only the part of the chain where data is processed, and it generates neither many skilled jobs nor any capacity for control over the software. They are merely extraction enclaves over whose crumbs certain regions of the world fight—regions that, like Argentina, officially offer precisely abundant energy and cheap labor as an “advantage.” In any case, the owners of the AI can always selectively disconnect its use, as they have just done with the Fable model, at the behest of the U.S. government or for even more mundane reasons.

The only solution available to other countries is to deploy their own artificial intelligence models, which is no simple task. On the one hand, the effectiveness of generative AI is the result of accumulating vast amounts of knowledge and of using cheap labor that, through arduous and repetitive tasks, helps improve the model’s accuracy. The implications of that “gestation” process are not minor for building an alternative to American software. On the other hand—and perhaps the most important factor—no European or Latin American country will, on its own, be able to build a competitive alternative to the American or Chinese models. The prerequisite for achieving this is necessarily a change in scale, that is, working cooperatively among countries. But scale is not enough, and if knowledge is a common inheritance, the alternative cannot be limited to reproducing the same extractivism under a European flag. It should aim at public and open models, governed as what that knowledge is at its origin: a collective good. It is precisely this hypothesis that the far right and the MAGA movement are trying to sabotage at the root by eroding Europe’s fragile institutions.