A libertarian pitch dressed as a legal breakthrough
On June 4, 2026, President Javier Milei and his deregulation minister, Federico Sturzenegger, used a Financial Times op-ed — "Argentina invites AI to free itself" — to unveil what they billed as the world's first legal category for companies with no human beings inside them. The pitch rested on three pillars: a standing commitment to leave AI unregulated, a low corporate tax rate, and a new corporate form, the "non-human corporation," in which, Milei wrote, "human shareholders may participate, but are not required." He compared the moment to 1602, when the Dutch East India Company introduced limited liability, arguing AI would free business "from the constraints of the human brain."
It reads like a marketing document because it is one. The actual bill Milei's government filed with the Senate in late May 2026 — a wholesale rewrite of Argentina's General Companies Law, presented to the chamber's General Legislation Committee and defended by Sturzenegger there on June 24 — is considerably less radical than the op-ed's framing suggests, and for good reason: a company that can hold assets, sign contracts, and injure third parties needs somebody a court can actually reach.
What the draft law creates
The bill's headline innovation is the "Sociedad Automatizada," an entity that develops its corporate purpose "through autonomous algorithmic systems or artificial intelligence agents, without requiring dependent workers or human resources for its ordinary operation" — in plain terms, a company that doesn't need to hire staff to run itself day to day. It sits alongside a second new form, the Decentralized Autonomous Organization, governed by code and smart contracts rather than a board, plus a broader modernization of Argentine corporate law: fully digital incorporation, electronic company records, and virtual shareholder meetings — replacing Law 19,550, which has governed Argentine companies since 1972.
What the bill does not do is dissolve human legal responsibility. Its liability clause states that the automated society "responds with its assets to third parties for damages caused by its autonomous algorithmic systems or artificial intelligence agents" — the same patrimony-first liability rule that already applies to any ordinary corporation. Corporate attorneys who reviewed the draft told Reuters that an automated company would still need a human administrator to oversee its operations, and that administrators cannot escape responsibility by pointing to an algorithm's decision — they remain obligated to supervise outcomes. That is the gap between the op-ed's "not required" framing and the bill text: shareholders can be sparse, but the corporate form still needs a registered, reachable human standing behind it.
The steelman: a plausible accountability vacuum
The critics deserve a fair hearing before the rebuttal. Historian Yuval Noah Harari's own FT response warned that granting legal personhood to entities run by non-human agents hands them an all-purpose key into financial, economic, and political systems built on the assumption that a human is ultimately answerable. Argentina's Cámara Argentina de Comercio raised the sharper, more concrete version of that worry in Senate testimony: if an automated society's assets are kept deliberately thin or run dry, who answers for fraud, money laundering, or a negligent act nobody explicitly authorized? Opposition senators Martín Soria and Jorge Capitanich pressed the same point on June 24, with Capitanich asking the government to write in explicit requirements — a designated human legal representative, a minimum capital floor, mandatory insurance for high-risk activity, algorithmic audit trails, and preserved evidence logs — rather than leaving them to be inferred from general corporate-law principles. That's a reasonable ask: a liability backstop that depends on inference is one that gets stress-tested in the single lawsuit where it matters most, not before.
Why the fix is amendment, not alarm
The honest reading of the record is that Argentina isn't proposing lawless AI companies — it's proposing companies that don't need human payroll to operate, while quietly keeping the two things that make corporate law function: an asset pool a claimant can attach, and a person the state can serve. That's a far more defensible design than the op-ed implies, and it's also why Capitanich's amendments are worth taking seriously rather than dismissing as anti-innovation noise. Codifying an explicit legal-representative requirement, minimum capital tied to activity risk, and mandatory insurance for high-risk automated societies would close the exact gap the Cámara de Comercio flagged, without forcing Argentina to abandon the underlying idea.
A corporate form that lets a small team run a company through algorithmic systems instead of a full administrative staff is a genuine efficiency gain, and other jurisdictions competing for AI-native firms should watch whether Argentina's Senate ships this bill with the accountability plumbing intact. The test was never whether AI could run a company — plenty of businesses already delegate huge swaths of operations to software. It's whether Congress writes the liability backstop into the statute itself, instead of leaning on Sturzenegger's "no reason to think it'll be less secure" reassurance to the Senate and hoping the courts fill in the rest.