Argentina has no comprehensive national AI law. One of its most concrete governance instruments is a provincial decree that applies only to the government of Buenos Aires Province itself. Decree 742/2026, signed by Governor Axel Kicillof and countersigned by Government Minister Carlos Alberto Bianco, approves a framework for the development, use and governance of AI across the provincial public administration. Coverage first appeared in mid-July 2026. The official text on the province's norms portal is dated July 6, 2026, so readers should treat that as the signing date.
The decree should be read for what it is. It is a rule about how the state buys and deploys AI. It does not regulate private companies.
What the decree actually does
The official text covers central, decentralized and autonomous provincial bodies, constitutional bodies, and majority-state-owned enterprises. Municipalities are invited to adhere voluntarily. Some summaries describe six thematic pillars, but the published text sets out eleven guiding principles. They include transparency, explainability and traceability, human supervision in decision-making, non-discrimination, accountability, technological sovereignty, interoperability, and data quality.
Three mechanisms carry the weight:
- Human supervision. The text calls for human supervision in decision processes, especially where citizens' rights are affected, and bars automating decisions that negatively affect fundamental rights. BAE Negocios reports that AI is defined as a tool of assistance and support, so no algorithm makes a final determination alone.
- A mandatory AI registry. Article 3 creates a Registry of Artificial Intelligence. Public bodies must record each system's characteristics, status and modifications.
- A central gatekeeper. The Digital Government Undersecretariat is the implementing authority, with power to set operational standards and supervise compliance. BAE Negocios reports that bodies must obtain its approval, after a risk assessment, before developing, contracting or deploying AI systems.
The strongest case for the decree
The case for this design is serious. Automated administrative decisions affect benefits, licences, fines and enforcement. When an agency adopts a scoring tool without a record of what it is or who answers for it, citizens cannot contest the outcome and auditors cannot reconstruct it. A registry plus a named human accountable for each decision is a modest, low-cost remedy. It also follows a pattern seen elsewhere: governments set rules for their own use of AI first, because they can do so without constraining private innovation.
I think that logic is largely right. Traceability and human review of rights-affecting decisions are proportionate. They target the point where state power meets the individual, and they do not presume that AI is dangerous in itself.
Where the risk lies
The weaker element is the pre-approval gate. Centralized authorization of every AI development, contract or deployment can turn into a bottleneck. A small agency that wants to use an off-the-shelf drafting assistant, or a municipal hospital that wants a transcription tool, would wait in the same queue as a high-stakes eligibility engine. Proportionality means scaling scrutiny to risk. A flat approval requirement for everything works against that. If the Undersecretariat does not publish clear tiers and fast turnaround targets, public servants will either avoid useful tools or adopt them informally, outside the registry the decree is meant to build.
There is also a capacity question. The decree concentrates supervision in one office. Its value depends on that office having the staff to review requests quickly, and the text, as summarized, does not say how it will be resourced.
The ban on automating decisions that negatively affect fundamental rights is a broad phrase. Many routine administrative steps touch rights in some way, such as a tax assessment or a permit denial. The Undersecretariat's standards will need to say which uses are covered. Otherwise the line will be drawn case by case, which favors caution over experimentation.
The national vacuum
The decree matters mostly because the national level is empty. Researchers from CONICET and UBA's TecnocenoLab found that 53 AI-related initiatives were filed in the two chambers of Congress in 2025, and only 11 aimed at a comprehensive AI law. Earlier, Deputy Maximiliano Ferraro introduced what Diario Judicial described as Argentina's first AI regulation bill in March 2024. It proposed a registry of significant-risk automated decision systems, a certification program for good algorithmic practices, and an advisory council. The bill file is on the Chamber of Deputies site.
None of that has produced a law. In December 2025, DPL News reported that a bill under debate faced resistance from both industry and the Milei government, and that the Buenos Aires precedent was pressuring Congress to avoid a patchwork of jurisdiction-by-jurisdiction rules. I could not verify the details of the Executive's own deregulatory AI proposal that some outlets have described, so this analysis does not rely on it.
What a national strategy should borrow, and what it should not
A national framework can take three things from Buenos Aires. First, a public-sector AI registry. It is cheap, it creates the data a future regulator will need, and it does not touch private firms. Second, a human-accountability rule for decisions affecting rights. Third, voluntary adherence for sub-national governments, which lets municipalities opt in without a federal mandate.
It should not copy the blanket pre-approval model. A national strategy that aims to attract AI investment should separate high-risk uses, which warrant review, from low-risk productivity tools, which should be adopted freely and logged afterward. It should also keep private-sector obligations out of any public-administration instrument.
If provinces each write their own rules, developers selling to the state will face a different compliance regime in each jurisdiction. That raises costs for the local startups the country says it wants to grow. A short federal baseline, with the Buenos Aires registry and oversight rules as a reference, would be better than 24 separate regimes.
Bottom line
Decree 742/2026 is a sensible first step on state use of AI, and its human-oversight and registry provisions deserve to be copied. Its authorization requirement needs risk tiers before it helps rather than hinders. Argentina's real gap is still national: no strategy, and no law that the 11 comprehensive bills have managed to become.