In late May 2026, Argentina's Ministry of Deregulation and State Transformation, led by Federico Sturzenegger, sent Congress a bill rewriting the country's Ley General de Sociedades — Law 19,550, on the books since 1972 (argentina.gob.ar). The reform creates two new legal entities: sociedades automatizadas, companies run entirely by algorithms or AI agents with no employees, and blockchain-based DAOs. Both would receive full legal personality and limited liability, answering for damages with their own assets (Infobae). Within days it had drawn a public exchange between President Javier Milei and historian Yuval Noah Harari, conducted partly in the pages of the Financial Times.
What the bill actually does
The core move is philosophical as much as technical. Sturzenegger summarized it bluntly: "State tutelage over how partners organize their businesses ends. The law's rules become supplementary: the bylaws command." An automated company could conduct commercial activity, hold assets, and respond to liabilities "únicamente con su propio patrimonio" — solely with its own assets. Human shareholders are permitted but not required (Buenos Aires Times). DAOs, recognized in Argentine law for the first time, would record operations on-chain and may submit internal disputes to arbitration or foreign law via smart contract.
Milei and Sturzenegger framed the proposal in a June 3 FT op-ed as the second pillar of a three-part AI strategy — light-touch regulation, a new corporate category, and a competitive tax environment — invoking the Dutch East India Company of 1602 to argue that limited liability was the legal innovation that made the industrial era possible, and is now the precondition for an AI one.
Steelman the warning first
Harari's objection deserves a fair hearing, because it identifies a real gap. In an FT counter-column titled "We should not grant legal personhood to AI agents," he argued that legal personhood is "a master key" that would let AI agents reach into "our financial, economic and political systems." His sharpest point is about deterrence: the ultimate sanction that disciplines human executives — prison — is, in his words, "irrelevant to AI." A self-directing entity optimizing for a goal, with no body to jail and no conscience to prick, could become "expert in regulatory arbitrage and legal loopholes." Nations that hand such entities legal standing, he warned, risk becoming not a state with companies but an "AI state" — a Batavia rather than an Amsterdam (OECD.AI incident tracker).
This is not Luddism. The accountability question is genuine: a company that can act in the world but whose only "skin in the game" is a pool of tokenized assets it controls is a different animal from a human-run firm, and pretending otherwise would be naive.
Why the experiment is still worth running
But Harari's framing overstates how novel the danger is, and understates what the bill already does to contain it. Corporate personhood has never meant moral personhood. Since the 19th century, corporations — abstractions with no body to jail — have sued, owned property, and shielded their owners. We discipline them not through incarceration but through liability attaching to assets, disclosure rules, and ultimately the humans who direct or profit from them. Argentina's bill keeps those anchors. Automated companies answer with their own patrimony; beneficial owners must be disclosed; and DAOs must name a human legal representative. That last requirement is the crucial one, and it directly answers the "no one to hold responsible" critique.
The deeper case for proceeding is that the alternative — refusing to recognize autonomous entities — does not make them disappear; it makes them lawless. The bill itself cites Sarcuni v. bZx DAO (2023), in which a U.S. court treated an unincorporated DAO as a general partnership, exposing token-holders to unlimited personal liability for the protocol's failures (PYMNTS). That is the status quo Harari implicitly defends: software organizations that already exist, transact billions, and fit no legal box, leaving courts to improvise and ordinary participants exposed. A clear statutory category with mandatory disclosure and a named human respondent is more accountable than that vacuum, not less.
The line between bold and reckless
Argentina has reason to be aggressive here. Sturzenegger's ministry has already eliminated or modified more than 2,500 normative provisions, and a country long synonymous with regulatory sclerosis has a rational interest in becoming a first mover. Being the jurisdiction where autonomous-agent commerce is legible rather than ignored is a credible competitive bet.
The risk is not the concept but the calibration. Proportionate regulation means the human-representative and own-asset-liability provisions should be load-bearing, not decorative: representatives with real duties and real exposure, asset floors that prevent judgment-proof shells, and beneficial-ownership transparency that survives the blockchain's pseudonymity. If Congress guts those safeguards in the name of deregulation, Harari's warning becomes self-fulfilling. If it strengthens them, Argentina will have built something genuinely useful — a framework that lets AI-mediated enterprise operate inside the law rather than around it. The instinct to legislate before the technology outruns the courts is correct. The details are where this is won or lost.