On 23 September 2026, President Javier Milei told the 81st UN General Assembly that Argentina wants to be "the most competitive place in the world" for developing AI. He listed the pillars as a cutting-edge legal framework, a commitment not to regulate preventively, limited liability for companies using AI agents, and tax benefits (Buenos Aires Times). "Where many see a threat, we see a frontier," he said. The speech also warned against UN-level AI rules and lined Argentina up with Washington's opposition to a centralised global control scheme.
We think the instinct is right. The details are what will decide whether it works.
The case for the other side
The strongest argument for preventive rules is that harms from AI systems are hard to undo once deployed. Facial recognition, automated credit scoring and synthetic sexual imagery hurt real people before any court can respond. Argentine legislators have filed bills on these problems. The IAPP's summary describes risk-tiered AI frameworks from Senator Silvia Sapag and Deputy Daniel Gollán, which ban behavioural manipulation and harmful social scoring. It also describes a facial-recognition bill from Deputy Martín Yeza that requires impact assessments and prior authorisation from the data-protection authority (IAPP). These bills borrow from the EU AI Act. Their authors are not cranks, and their worry is legitimate.
Why 'no preventive regulation' is still the better default
Risk-tier regimes assume regulators can classify systems that change faster than statutes can. They tend to load compliance costs onto small firms while large incumbents absorb them easily. Argentina is also not starting from a legal vacuum. Ley 25.326, sanctioned on 4 October 2000, already governs personal data. Its Article 20 says administrative decisions that assess human conduct cannot rest solely on automated processing of personal data, and it declares such acts null (Argentina.gob.ar). Consumer-protection and civil-code liability also apply to AI products as they do to any other product. That covers a good deal of the harm the risk-tier bills target, without a new AI-specific bureaucracy.
The agency that enforces the data law, the Agencia de Acceso a la Información Pública, already exists and is responsible for personal-data protection (AAIP). Waiting until a harm is demonstrated, and then legislating narrowly, is a defensible sequence. It is also consistent with a pro-innovation, evidence-based view of regulation.
The pitch rests on real assets
This is more than rhetoric. In October 2025 the government announced a letter of intent between OpenAI and Sur Energy for a data centre of up to 500 MW in Patagonia, described as the first Stargate project in Latin America. The reported US$25 billion investment would run through the RIGI large-investment regime. Coverage noted that OpenAI itself would put in no money and would be the main buyer of compute once the site is operating. The same coverage put Latin America's largest existing data centre at 61 MW, in São Paulo (Buenos Aires Times). A letter of intent is not a built facility. Even so, cheap energy is the scarcest input for AI compute, and Argentina has a real story to tell on it.
Where the pledge gets thin
Three gaps stand out.
- "Limited liability" is itself a regulatory choice. Limiting what an AI-agent operator owes to a harmed user or counterparty means deciding who bears the loss instead. Without a written standard, a court will decide case by case, and uncertainty of that kind is what investors dislike most. A clear safe harbour, tied to concrete conduct such as logging, human escalation and disclosure that a user is dealing with an agent, would serve firms better than a slogan.
- A promise not to regulate is not a legal framework. Milei used both phrases in the same breath. A cutting-edge framework needs statutory text, and Argentina's Congress currently holds mostly EU-style proposals, not a competing pro-innovation one. If the executive does not offer its own bill, the risk-tier drafts remain the only ones on the table.
- Speech and rights still need narrow guardrails. Argentina's data law dates from 2000, and reformers in Congress are working on its update. A no-preventive-rules line should not extend to leaving automated decisions about people's benefits, credit or liberty without any recourse. Article 20 already draws that line, and the government should say plainly that it stays.
What proportionate looks like
The best route is neither the EU model nor a regulatory void. Argentina could keep its horizontal rules minimal and enforce existing law. It could adopt the sandbox idea from Gollán's bill without the risk-tier machinery. It could write a short, explicit liability safe harbour for AI-agent operators, and pair it with a narrow, targeted rule on facial recognition in public security. That would give the country a stronger argument than the speech did, and it would sit comfortably with the "global race" framing Milei chose.
The UN podium is a fine place to set out a philosophy. Investors, though, will read statutes. If Argentina turns "limited liability" and "cutting-edge framework" into short, predictable text, its claim to be the most competitive AI jurisdiction will be credible. If it does not, the pledge will be read as a slogan and the EU-inspired bills will fill the gap.