On 23 September 2026, President Javier Milei told the UN General Assembly that Argentina wants to be the most competitive place to develop AI, with a commitment not to regulate preventively. "Prudence does not mean fear," he said, and he called the dystopian scenarios used to justify control economically senseless. According to El Economista, he also announced a legal framework with tax benefits for AI developers. The offer is paired with a bill that would let companies run by algorithms or AI agents operate under their own legal personality.
The pledge deserves a serious reading, not a reflexive one. Its core is defensible. The weakest part is the corporate-law vehicle, not the decision to hold back on new AI-specific rules.
The strongest case against Milei's approach
The best argument for regulating early is that harms from automated systems are hard to reverse once they are embedded in credit, hiring or public services, and that clear rules can build the trust that adoption needs. The European Union took that route. Its AI Act sorts systems into four risk levels, from prohibited to minimal, and entered into force on 1 August 2024. General application began on 2 August 2026, and some high-risk obligations are now scheduled for December 2027 and August 2028. A critic can fairly say that Argentina is skipping a protective layer that other jurisdictions consider basic.
Oscar Soria, quoted by Perfil, put the point sharply: there is "no evidence" that less regulation necessarily produces more innovation or investment.
Why the no-preventive-rules stance is still reasonable
The EU's own timetable shows the cost of writing detailed rules before the technology settles: obligations have already been pushed back more than once. A middle-income country without a large domestic model industry gains little from copying a compliance regime built for Brussels. Argentina is also not a legal vacuum. Ley 25.326, the personal data law sanctioned on 4 October 2000, already governs how personal data may be processed. Its Article 20 voids judicial or administrative acts that assess human conduct based solely on automated processing of personal data that defines a person's profile. That is a real, technology-neutral safeguard against the most obvious algorithmic harm in public decision-making.
A proportionate approach would therefore enforce existing consumer, data and labour law against concrete harms, and legislate narrowly where gaps show up. That is a stance worth defending, and it is consistent with an open, pro-innovation internet policy.
Where the plan gets shaky: automated companies
The harder question is the Companies Law reform. Press reports say the Executive sent a comprehensive reform of the Ley General de Sociedades to Congress around 29 May 2026. As summarised by Marval, it would create companies that run through "autonomous algorithmic systems or AI agents" with no payroll employees. These would have to carry the word "Automatizada" in their names. The bill also creates a DAO company type. Business coverage says such entities would have their own legal personality and limited liability.
Legal personality for a business that automates its operations is not itself radical. Companies are already legal fictions, and software has long run much of their daily activity. What matters is who answers when something goes wrong. In a June 2026 Financial Times piece summarised by the Buenos Aires Herald, Milei and Deregulation Minister Federico Sturzenegger called limited liability "a precondition" for these entities. The Herald reports that shareholders could choose their governance law, but beneficial owners would have to be disclosed. Critic Ariel Garbarz warned of "programmed impunity" in which responsibility shifts onto machines.
That concern is legitimate, and the government appears to have heard it. Perfil reports that the ruling party said it would add a "responsible natural or legal person" after receiving objections. Soria argues that naming a responsible person is not enough if the structure lets central decisions be made autonomously. Specialists have also flagged legal fragmentation, since bylaws would take precedence over the statute and creditors could struggle to know which rules apply, according to press summaries of the bill.
What a better design would look like
Three fixes would keep the innovation upside while closing the accountability gap:
- A named, liable human or entity. Every automated company should have an identifiable controller who answers for its acts, with limited liability protecting shareholders' capital but not the controller's duty of care.
- Disclosure at the point of dealing. The "Automatizada" label is a good start. Counterparties and consumers should be able to see who stands behind the entity and what recourse exists.
- Mandatory-minimum rules for creditors and third parties. Contractual freedom suits sophisticated shareholders. Outside parties, including workers, tax authorities and consumers, did not sign the bylaws.
None of this requires an AI-specific regulator or a risk-tier regime. It requires the ordinary discipline of corporate law.
The investment narrative needs scrutiny too
The UN pitch sits beside a separate investment push. The Herald notes a letter of intent between OpenAI and Sur Energy for a Patagonian data center estimated at US$20-25 billion, not yet formally launched. Critic Enrique Viale told Perfil that the Super RIGI regime's 30-year regulatory stability amounts to a clause lasting until 2056. Long-term stability is what data-centre investors ask for. But locking in rules for three decades while promising to avoid regulation is a tension: a state that cannot adjust later has to get the first design right.
Bottom line
Argentina's refusal to copy the EU's front-loaded model is a legitimate policy choice, and its existing data-protection law already supplies a floor. The credibility of the strategy, though, depends on the Companies Law reform. If automated entities come with real human accountability, they could become a useful experiment. If they become a liability shield, the first scandal will hand regulation-first critics the argument they need.