A 1972 law meets a 2026 problem
Argentina's Senate Commission on General Legislation, chaired by Senator Nadia Márquez, has spent the winter working through a bill that would replace the country's General Companies Law No. 19,550 — in force since 1972 — with a framework built for firms that don't need a payroll to operate. Federico Sturzenegger, the Minister of Deregulation and State Transformation, has been the government's lead witness, appearing before the committee on June 24 and again on August 18, 2026, to defend the bill's most novel provision: the Sociedad Automatizada, or Automated Society.
Article 14 of the draft defines it as an entity that carries out its corporate purpose through autonomous algorithmic systems or AI agents, without needing employees in a dependent relationship for ordinary operations. The bill also creates a second, more radical vehicle — the Sociedad Descentralizada Autónoma Operativa, Argentina's version of a DAO — built on blockchain protocols and token-based participation that can run partially or entirely without human input.
The liability claim
Sturzenegger's core argument to the committee is that none of this requires inventing new liability law. An Automated Society, he has said, responde con su patrimonio — answers with its assets — for damages its algorithms or AI agents cause to third parties, exactly as any corporation does today. The bill explicitly blocks a company from disclaiming an obligation by pointing to the "autonomy" or "unpredictability" of its own software. "No hay motivo para pensar que va a ser más insegura," he told senators in June — there's no reason to think it will be less safe than a conventional company.
He has also been candid about the experimental nature of the bet. Asked whether the model would actually work, he answered: "¿Va a funcionar? No lo sabemos. ¿Vale la pena? Sí" — will it work? We don't know. Is it worth trying? Yes.
Where critics have a real point
That candor is worth taking seriously, because the legal objections aren't just partisan noise. Ricardo Nissen, who ran the Inspección General de Justicia — Argentina's corporate registry — under two previous governments, warned the committee the bill risks becoming "un sistema sin domicilio social, sin actividad empresarial, sin capital social": a system without a registered address, without business activity, without paid-in capital. A separate legal critique circulating among corporate lawyers raises a more basic problem: Argentine law does not recognize AI systems as legal subjects, so an algorithm cannot itself be a founder, administrator, or signatory — someone still has to sign, and the bill doesn't fully say who. Judge Julián Benito Flores put the tension succinctly before the committee: "Modernizar no es desproteger; simplificar no es desregular" — modernizing isn't the same as stripping protection, simplifying isn't the same as deregulating.
These aren't fringe complaints. A liability regime that can't identify, in a specific dispute, which named person authorized the algorithmic decision that caused harm isn't really a liability regime — it's a disclaimer with better branding.
Why "same as any corporation" is still the right starting point
Even so, the underlying instinct — extend the existing corporate-veil framework rather than draft a bespoke AI-liability statute from scratch — is sound, and probably the least disruptive way to bring AI-operated firms into a functioning legal system. Limited liability backed by traceable assets has absorbed novel business forms before, from Argentina's own stripped-down SAS to fintechs. The alternative — refusing to let any AI-run entity incorporate until a perfect bespoke statute exists — would just push that activity into jurisdictions with weaker oversight, not stronger. Sturzenegger's comparison to trusting Google Maps' routing decisions is glib but not wrong on the merits: people already delegate consequential choices to algorithms without a special legal category for it. The real question isn't whether algorithmic delegation is acceptable — it's whether the entity behind it can actually be reached when something goes wrong.
The check that's already working
That's precisely the gap the legislative process is now closing on its own. On August 19, Patricia Bullrich — an ally whose bloc's votes the government needs — announced she would push amendments making the liability of automated societies "mucho más pétrea," far more solid, by requiring clearly identified physical people standing behind each one. That isn't the bill collapsing; it's the ordinary business of a legislature hardening a proposal before voting on it, exactly where that kind of scrutiny belongs. A framework built to survive committee hearings, floor amendments, and eventual courtroom testing is more durable than one drafted in isolation and never pressure-tested. Argentina doesn't need to resolve every hypothetical about AI personhood to let algorithmically-run firms incorporate; it needs a traceable human on the hook when the algorithm gets it wrong. The amendment now on the table suggests the Senate is converging on exactly that, without abandoning the underlying wager on attracting AI-native capital.