Buenos Aires Province, Argentina's largest and most populous jurisdiction, published Decree 742/2026 in its Official Bulletin on July 14, 2026, creating the first binding artificial intelligence governance framework by any Argentine government body. The decree requires every provincial agency that develops, buys, or deploys an AI system to log it in a new Artificial Intelligence Registry, run a prior impact assessment before rollout, and preserve human oversight over any decision the system informs. It bans systems that pose "unacceptable risk," explicitly naming social scoring and behavioral classification tools among the prohibited uses. The Ministry of Government, through its Digital Government Subsecretariat, will administer the registry and issue interpretive rules (La Nación; El Cronista).
Why Kicillof's government moved
The trigger was concrete, not theoretical. Earlier this year the municipality of Zárate assigned a chatbot named "ZARA" apparent directorial authority over municipal functions, an episode that alarmed provincial officials watching AI deploy faster than any oversight structure could track it, per El Cronista's reporting on the decree. Governor Axel Kicillof's administration frames the response as risk-tiered and EU-inflected: an "unacceptable risk" category is banned outright, high-risk systems require impact assessments and ongoing monitoring, and citizens must be told when they are interacting with an automated system rather than a person.
The steelman
The case for binding this now, rather than waiting for a comprehensive statute, is genuinely strong. Public administration is exactly where AI failures compound fastest: a benefits-eligibility algorithm or a permitting chatbot can affect thousands of residents before anyone outside the agency notices a defect, and unlike a private company's product decision, a government misstep isn't disciplined by exit or market competition — citizens can't switch providers. A registry that forces agencies to document what a system does before it goes live, and a flat ban on social-scoring tools, are proportionate, low-cost interventions squarely inside the province's own administrative authority. Nothing in Decree 742/2026 requires private companies to change how they build or sell AI products; it only constrains how the government itself may use them. That is the correct place to start, and it is a narrower, more defensible move than premature private-sector mandates would have been.
Where the analogy breaks down
The risk is that this decree gets read, in Buenos Aires and beyond, as evidence that Argentina now "has" an AI governance framework — when in fact it governs one layer of one province's own bureaucracy and says nothing about the platforms, models, and algorithmic content-ranking systems Argentines actually interact with every day. At the federal level, the gap is real. Deputy Diego Giuliano's bill to create a National Registry of Artificial Intelligence Systems, with EU-style risk tiers and prohibited-use categories, was filed in the Chamber of Deputies on October 17, 2024, and referred to three committees; it has not advanced since (Cámara de Diputados). The one federal body with a mandate spanning the topic, the Interministerial Board on Artificial Intelligence, was restructured under Administrative Decision 899/2024 to span ten ministries plus two coordinating secretariats — but it is a coordinating forum, not a rulemaking authority, and it issues no binding obligations on anyone (Boletín Oficial).
ENACOM's mandate stops short of platforms
That gap matters because Argentina's most obvious existing regulator, ENACOM, has no general competence over platform liability or AI-driven content decisions. Its statutory mandate under Ley 27.078 ("Argentina Digital") and Ley 26.522 covers telecommunications licensing, spectrum, and broadcasting; its 2026 rulemaking agenda has focused on equipment certification and RAMATEL registration for online telecom-equipment sellers, not on how platforms moderate, rank, or generate content. Argentina has never adopted an intermediary-liability statute comparable to the EU's Digital Services Act or Brazil's Marco Civil, leaving courts to improvise from general civil-liability doctrine on a case-by-case basis when platform disputes reach them. A provincial decree about internal government AI use does nothing to close that hole, and shouldn't be mistaken for progress toward it.
The proportionate path forward
Decree 742/2026 is a sound model precisely because of what it doesn't attempt: it doesn't try to regulate model training, impose liability on platforms for third-party content, or freeze private AI deployment behind a permitting wall. If Congress uses it as a template for the Giuliano bill's committee work, the lesson to take is the scoping discipline, not the registry mechanic — start with government's own use of AI, where the state is both regulator and user and can hold itself to a higher standard without chilling the broader market, before reaching for platform-wide mandates that would need much more evidentiary grounding. Bolting an untested liability regime onto ENACOM's existing telecom mandate, without the sector-specific analysis Congress hasn't yet done, would risk exactly the kind of overreach this decree carefully avoided.