When President Javier Milei and Deregulation Minister Federico Sturzenegger published their June 4 Financial Times op-ed pitching Argentina as 'open for business' for AI investment, they reached for a bold historical parallel: the 1602 creation of the Dutch East India Company, whose limited liability model they argued 'unleashed capitalism's full potential.' Four days later, historian Yuval Noah Harari used the same newspaper to warn that granting legal personhood to AI-operated entities could make recipient nations something 'for which the historical record offers no analogy: not a company-state, but an AI-state.'
The exchange was theatrical. The underlying legal question it surfaced is genuine and important: how should corporate law treat entities whose operational logic runs entirely on algorithms? Argentina's answer — the proposed Sociedad Automatizada under a comprehensive rewrite of Law No. 19,550, submitted to Congress on May 29 — is more carefully designed than the headline debate suggests, and more incomplete than Milei's marketing implies.
What the Bill Actually Does
The new General Law of Societies, which would replace Argentina's 1972 corporate statute in its entirety, introduces the Sociedad Automatizada in Article 14: a company that 'develops its business purpose through autonomous algorithmic systems or artificial intelligence agents, without requiring workers in a dependent relationship or human resources for its ordinary operation.' The entity carries full legal personality with limited liability. But crucially, the corporation itself 'responds with its assets against third parties for damages caused by its systems.'
Two embedded safeguards are worth noting. First, any company invoking the automated category must include the word Automatizada in its registered name — a disclosure requirement that lets counterparties know upfront what they are contracting with. Second, ultimate beneficial owners must still be identified, preserving the anti-money-laundering chain that prevents anonymous offshore actors from hiding behind an AI front. The bill also recognises DAOs, though under a distinct governance framework tied to blockchain token-holders rather than algorithmic management.
This is not, contrary to breathless coverage, a law that lets anonymous AI systems operate outside legal reach. It is closer to extending the Salomon v Salomon principle — corporate legal personhood separate from members — to entities whose operating logic is algorithmic rather than human.
The Economic Logic Is Coherent
The Companies Law reform does not exist in isolation. On June 25, Argentina's Chamber of Deputies voted 130 to 106 to pass the Super RIGI, a large-investment incentive regime requiring a minimum project size of US$1 billion and offering a flat corporate income tax rate of 15 percent — against Argentina's standard 25 percent — plus 30 years of fiscal and regulatory stability for qualifying sectors including AI, semiconductors, and data centres.
That combination of legal clarity and fiscal incentive is already attracting serious capital before the laws have even passed the Senate. OpenAI and Argentine energy company Sur Energy have agreed a Letter of Intent for Stargate Argentina, a proposed 500-megawatt data centre in Patagonia with potential investment of up to US$25 billion. The first 100MW phase is targeted for 2027. For a country that has cycled through eight IMF programmes in 50 years, anchoring long-horizon infrastructure investment is not a vanity project — it is a structural economic requirement.
The Accountability Gap Harari Identifies Is Real
Before dismissing Harari's warning, the argument deserves a fair hearing. Traditional corporate liability rests on two deterrents: financial penalties levied on the corporation's assets, and criminal sanctions — including prison — levied on the human executives who directed misconduct. The Sociedad Automatizada eliminates the second deterrent entirely. An AI system cannot be imprisoned, and research on advanced AI agents has documented their tendency to exploit operating environments in ways that, in a corporate context, could shade into systematic regulatory arbitrage.
The concern sharpens when asset liability is the only backstop. A Sociedad Automatizada with thin capitalisation can cause harm, strip its asset base, and leave injured third parties with an empty shell to sue. Argentine AI specialist Ariel Garbarz named this dynamic precisely: 'programmed impunity — human gains, social harm and responsibility shifted onto machines.' The phrase captures a real structural risk, not a hypothetical one.
Where Milei's Rebuttal Holds
That said, Milei's core counter-argument is not wrong: giving AI-operated entities legal personhood subjects them to law more fully, not less. Most AI agents currently function as undeclared principals or contractors in legal grey zones — causing harm with even less accountability than a recognised legal person would face. Argentina's framework at least creates a sueable, asset-holding entity with identified beneficial owners.
Harari's framing treats AI legal personhood as an unprecedented danger. But limited liability corporations were condemned as immoral fictions in the 19th century, too. The innovation of treating a non-human collective as a legal person underpinned every major economic expansion since 1600. The question is not whether AI entities should have legal standing — they arguably already need it — but what liability architecture accompanies that standing. The European Commission withdrew its proposed AI Liability Directive in 2024 without answering that question; Argentina is at least trying to answer it.
What Proportionate Regulation Would Add
The missing element is a minimum capital adequacy requirement for the Sociedad Automatizada category. In sectors where these entities operate at scale — financial services, logistics, automated contracting — a mandatory paid-in capital floor proportionate to potential third-party exposure would ensure that asset liability is meaningful rather than nominal. This is not regulatory overreach; it is the same logic applied to banks, insurance companies, and investment funds.
A second fix would be a designated human responsible officer — not a manager in the operational sense, but a named individual required to respond to regulatory inquiries and certify annual compliance filings. This preserves the 'no human staff required' framing while restoring the accountability node that Harari correctly identifies as missing.
Argentina's three-pillar AI strategy — deregulation, legal innovation, fiscal incentives — is coherent and more carefully constructed than critics acknowledge. But coherence is not completeness. The liability floor must be built into the bill now, while the draft is still in Congress and before the first Sociedad Automatizada registers and the first question of injured-party compensation arises. A framework that answers Harari's accountability concern would be harder to attack — and more durable as a model for other jurisdictions watching closely.