Milei vs. Harari: The Argentine Move That Could Change the World (and the Historian's Fear)

In just over a month, artificial intelligence brought together voices that rarely share a stage: Alex Karp, CEO of Palantir; the Pope; Chris Olah, co-founder of Anthropic; and President Donald Trump. These are interventions of very different scale and register, but all orbit the same axis. The debate on artificial intelligence moves between two poles that never quite meet: responsibility or innovation, prudence or creation, fear or possibility. Here we want to focus on the most concrete and closest clash of recent weeks: the one between Javier Milei and Yuval Harari in the pages of the Financial Times.

Milei opened the exchange with a column in which he announced that Argentina will create a new legal figure: the non-human corporation, a company operated by AI agents. His argument is historical and rests on an example: just as the limited liability company — created in 1602 with the founding of the Dutch East India Company — allowed capital to be pooled, risk to be financed, and Amsterdam to become the financial and commercial capital of the world, the non-human corporation would today unleash the productivity of AI. The innovation was legal, not technical: limited liability ensured that each investor lost, at most, what they invested. Suddenly it became rational to finance huge and risky companies — a fleet that sailed to Asia and could sink entirely — because exposure was capped in advance.

Days later, in the same outlet, Harari responded. He acknowledged Milei’s audacity and conceded a central point: the limited liability company was indeed one of the most momentous inventions of modern history, and the non-human corporation could be a step of equal magnitude. But he shifted the focus. The consequences of that innovation, he says, were not felt in Amsterdam but in the port of Jayakarta, in present-day Indonesia, which the Company razed in 1619 to build Batavia, the capital of its Asian commercial empire. Days later, Milei would respond again in an open letter, with arguments that largely coincide with those we have come to develop; we wanted, nonetheless, to add a few more elements.

Harari’s argument is elegant, but the historical example that supports it is, to say the least, anachronistic and, one might say, biased. Harari calls the Dutch Company a «state-enterprise»: a political entity governed by a private firm. And from there he draws the warning: what the Company did in Jayakarta, an AI corporation would do in Buenos Aires. But the anachronism lies precisely there, in bringing violence as an example. The Company did not burn Jayakarta because it was a limited liability company, but because it operated in an imperial world, prior to the modern state, where conquest was a legitimate business and companies went out to explore and subdue under the auspices of the crown. The boundary between state and market that we take for granted today — and within which we discuss with complete naturalness what belongs to each — was yet to be drawn. Drawing a parallel between the violence of those companies that owned armies and a modern state that opens up to AI is projecting onto the present a world that no longer exists: the order capable of containing those excesses was built later, and it is precisely the one in which Argentina legislates today.

In that same book where he tells the story of the Dutch Company — chapter XVI of Sapiens — Harari brings an example that weakens his own argument: the English East India Company. Founded in 1600, two years before the Dutch one, its strength came from a second component of the same innovation: capital divided into shares, which allowed risk to be spread and sums to be raised that no individual could gather alone. With that money, fleets, forts, and armies were financed — the Company eventually maintained an army of three hundred fifty thousand soldiers. Yes: companies had armies. Within the framework of imperial conquests, a company was something else, and its boundary with the public was something else: it waged wars, administered territories. The difference lies in what came later, and that is what Harari omits. In 1858, the crown nationalized those territories and absorbed that private army. By then we were already facing a modern form of the state, which had learned enough to know where to draw the line: the monopoly of force, for the crown; commerce, for the company. It was experimentation with legal innovations — not outright prohibition — that was the condition for understanding the phenomenon and defining its limits.

There is a thread connecting those legal innovations. It goes from business — the limited liability company, which invented a legal subject capable of owning, owing, and being sued without being anyone in particular — to the modern state, with the assumption of the legitimate and legal monopoly of violence. The fictions that modernity used to build both the company and the state were never an anomaly: they were its basic mechanism to safely enable new activity. And the state’s capacity to identify risks was not prior to innovation but subsequent: it arose from the interaction between a private sector that innovated and a state that, by dealing with that innovation, modernized itself. That is the logic that Japan is already testing with its AI Promotion Act: identify opportunities, encourage business and innovation, and, based on that learning, modernize the state.

The constitutional republic itself belongs to that lineage of bets. When the United States was founded in 1787, republican precedents were distant and on a different scale; turning principles of political philosophy into a concrete architecture — a constitution, the separation of powers, federalism — was an experiment with no guarantee of success. It worked, and today it seems obvious; in its origin it was a bold legal innovation that only showed its fruits when put into practice. Milei’s non-human corporation would be, in this reading, another link in that tradition, and what links it to the experience of 1787 is not the subject matter — there sovereign power, here AI agents — but the gesture: risking a new legal form as a condition for discovering new opportunities and progressing.

So: what problem does the non-human corporation seek to solve? That of responsibility when the actor is not a person. It is not a made-up dilemma or a laboratory hypothesis: AI agents are already operating, and the law lags behind. Argentina can offer an answer. If an AI agent decides autonomously — buys, sells, contracts, invests, signs — the law does not know whom to hold accountable: the programmer, the owner, the user? That uncertainty hinders deployment, because no one wants to operate an autonomous system if they are personally liable for its actions without limit. Milei’s argument is that limited liability would do for AI what it did for 17th-century maritime trade: if the agent acts within an entity that can own assets, be liable for its debts up to a certain limit, and be sued, risk becomes calculable and activity becomes financeable. And here it is worth clarifying what we are talking about, because this is where Harari argues biased by the extreme fear of technology he has been expressing in recent years: the non-human corporation, like an LLC, would be a legal person under private law and of a commercial nature, not a subject with political rights. The distinction is clear.