Buenos Aires province did what Argentina's Congress has not managed in over two years: it wrote a binding rulebook for how government uses artificial intelligence. On July 14, 2026, Governor Axel Kicillof signed Decree 742/2026, creating the Marco Provincial para el Desarrollo, Uso y Gobernanza de la Inteligencia Artificial — a mandatory framework covering every ministry, decentralized agency, and state-owned company in the province. The decree, signed alongside Government Minister Carlos Bianco, designates the Subsecretaría de Gobierno Digital as enforcement authority and bars any provincial body from developing, contracting, or deploying an AI system without its prior sign-off (La Nación).
What the decree actually does
The framework rests on eleven guiding principles — human centrality, transparency, data protection, non-discrimination, and "technological sovereignty" among them — and creates a Registro de Inteligencia Artificial where agencies must log every system's characteristics, status, and modifications (InfoPlatense). It borrows its risk architecture almost directly from the EU's AI Act (Regulation 2024/1689): four tiers — unacceptable, high, limited, and null risk — with outright bans on "unacceptable risk" systems, subliminal manipulation techniques, and social scoring that would condition public benefits on behavioral data (El Cronista). High-risk uses require human sign-off before a final decision, plus impact assessments covering fundamental rights, labor effects, algorithmic bias, and environmental cost. An unpaid interdisciplinary advisory council of specialists and universities will issue non-binding recommendations. Municipalities can opt in voluntarily; none are compelled.
The steelman: this gap was real
The case for the decree isn't hypothetical. Provincial deputy Silvina Nardini pushed a bill earlier this year after reports that AI-generated intimate images of students circulated in provincial schools — a concrete harm with no administrative rulebook governing how the state should even respond (Ámbito). More broadly, provincial agencies already use algorithmic tools to help decide who gets welfare benefits, where police resources go, and which taxpayers get flagged for audit — decisions where an unaccountable model can quietly encode bias at scale, with no registry even showing regulators where the tools operate. A government that puts algorithms in front of citizens without a paper trail, human sign-off, or the option to contest an automated outcome deserves the criticism it would get. Requiring registration and impact assessment before deployment is a genuinely low-cost, high-value guardrail for public-sector use — a different animal from regulating private-sector AI development, and one where the state, as both regulator and user, has every reason to hold itself to a higher standard than it holds industry.
But it also confirms a federal vacuum, and that's the actual story
Argentina's only national AI governance document remains Disposición 2/2023 — non-binding "Recomendaciones para una Inteligencia Artificial Fiable" issued by the Jefatura de Gabinete, built on UNESCO's ethics recommendation, with zero enforcement teeth (Argentina.gob.ar). Since then, deputy Diego Giuliano's proyecto de ley — introduced April 4, 2026, proposing a national AI registry, risk tiers, and a six-month compliance window for existing systems — has sat in the Chamber's Science, Technology and Productive Innovation committee without a floor vote, one of more than a dozen AI bills stuck there since the debate began (HCDN bill text; El Intransigente). Meanwhile President Javier Milei's government is moving the opposite direction at the federal level — pitching Argentina as a deregulated AI laboratory, courting the Stargate Argentina data-center project with OpenAI, and advancing a General Law of Corporations reform that would let "automated societies" run on AI agents with no human employees, explicitly resisting the Giuliano-style regulatory model industry groups like Argencon call "intervencionista y burocrático" (iProUP).
That's the real risk in Buenos Aires province moving alone: not that the decree overreaches — it's narrowly scoped to public administration, not private AI development — but that Argentina now has one province regulating with EU-style risk tiers, a national executive actively deregulating, and a Congress unable to pass anything at all. If Córdoba, Santa Fe, and Mendoza each write their own provincial AI decrees over the next year, government contractors and civic-tech vendors selling across jurisdictions will face 24 incompatible rulebooks before the national government settles on one. That patchwork cost falls hardest on smaller vendors and startups, the opposite of what Milei's investment pitch is supposed to deliver.
The fix isn't for Congress to adopt Giuliano's full private-sector regulatory apparatus, nor for Milei to keep AI governance entirely off the federal agenda. A narrowly scoped national framework — mirroring what Buenos Aires just built for its own government: risk tiers, a registry, and hard bans on social scoring, limited strictly to public-sector use — would preempt the patchwork without touching the private-sector innovation Argentina is trying to attract. Kicillof has effectively field-tested that model at province scale. Congress has the template sitting in front of it; it just needs to act before more provinces write their own.