On 23 September 2026, President Javier Milei told the 81st UN General Assembly that, on AI, Argentina "se ubicará en la vanguardia de la libertad." According to the Casa Rosada transcript, he promised "un marco jurídico de vanguardia," a commitment "de no regular de manera preventiva," "responsabilidad limitada para las empresas que operen con agentes de IA y beneficios fiscales." He listed critical minerals, energy and talent as Argentina's advantages.
The passage was brief. Chequeado counted nine references to AI in a 20-minute speech in which Malvinas took about seven minutes. It still works as a national AI strategy, because it restates a legislative package already sent to Congress.
The strongest case for the other side
Critics have a serious argument. Liability rules are how societies make firms internalize the harms their products cause. An autonomous system can deny credit, give bad medical advice or move money. If the operator's exposure is capped in advance, victims bear the residual risk. Precaution is also not fear. Milei himself conceded in the speech that AI "exige prudencia," and added that "prudencia no significa miedo." The question is whether the bills' design delivers prudence by other means.
What is actually on the table
The speech echoes two June 2026 initiatives. DPL News reported on 4 June that the government sent Congress articles that would create a "Sociedad Automatizada," a company run without employees. By DPL's account, the text says such a company must "responder con su patrimonio" to third parties for damage caused by its AI systems. That is narrower than the phrase "limited liability" suggests. Shareholder protection and liability to victims are different things, and the drafting will matter more than the slogan.
The second piece is a "súper RIGI." Emol reports a minimum investment of US$1 billion, a 15% income tax rate for adhering projects, and accelerated depreciation. It builds on the existing RIGI, created by Ley 27.742. According to the official RIGI page, that regime offers a flat 25% income tax rate, 30 years of regulatory stability, and export duty relief from the third year. The page gives a general investment floor of USD 200M.
The sources I reviewed report that the bills were sent to Congress but do not report a committee vote or approval. Nothing here is law yet.
Why the pro-innovation instinct is right
The argument against preventive regulation has real support. Rules written before a technology's uses are known tend to lock in incumbents and today's assumptions. Argentina is a mid-sized economy that cannot out-subsidize the US or EU. A predictable tax regime and a refusal to license models before deployment are things it can credibly offer. Compared with ex-ante regimes that classify systems by risk tier before the market has shown where harm occurs, a light-touch approach is a defensible bet.
The stated advantages are also real inputs for compute-heavy industry. Milei named energy, critical minerals and talent, and the transcript confirms it (it says "talento humano," not "recursos humanos"). Data centers need cheap power and cooling, and lithium-rich Argentina has both land and energy projects to offer.
Where the strategy is exposed
First, credibility beats incentives. The RIGI's headline feature is 30-year stability, which is a promise about future governments. Argentina's history of abrupt policy reversals is why investors discount such promises. A law that depends on a presidential agenda is weaker than one with cross-party support. The Super RIGI and the automated-company rules have not been reported as passed, and their durability past 2027 is untested.
Second, liability design will decide whether the offer is attractive or toxic. A regime that is too soft invites the "impunity" warning in the joint statement reported by teleSUR. Oscar Soria warned it "puede convertirse en una cueva jurídica para esconder responsabilidades." teleSUR is the only outlet I found carrying this statement, so its breadth is unconfirmed. Still, the concern is not frivolous. A jurisdiction seen as a liability haven risks reputational costs, and counterparties and insurers may avoid it. The better design is clear, proportionate liability: operators answer for foreseeable harm, with caps and insurance requirements rather than blanket immunity.
Third, "no preventive regulation" is not "no regulation." Argentina already has data protection law, consumer law and civil liability rules. Whether a Sociedad Automatizada falls under them is exactly what firms and victims will need answered. Silence is not a regulatory strategy. Courts will fill it, case by case, and that is slower and less predictable than a short, clear statute.
Fourth, the pitch leaves out the unglamorous inputs. The UN speech lists minerals and energy. It says little about grid capacity, connectivity, or the university pipeline. Tax rates are the cheapest lever to announce and rarely the binding constraint on where data centers and AI labs locate.
What would make it credible
Three steps would turn rhetoric into a strategy. Legislators should pass the bills with a defined liability floor, such as mandatory insurance or asset backing for automated companies, so that "responsabilidad limitada" does not become an excuse. The government should publish an AI-specific enforcement guide stating which existing laws apply to AI agents, so there is a clear baseline in place of regulatory silence. And it should lock the incentive terms into statute rather than decree, extending the RIGI's 30-year stability logic to AI.
Milei is right that fear should not become obedience. But for investors, the test is whether the law will still read the same in 2035. Argentina's AI strategy will be judged by that, not by the speech.