A Court Draws Its Own Line
On June 24, 2026, the Suprema Corte de Justicia de la Provincia de Buenos Aires (SCBA) — the top court of Argentina's most populous province — issued Resolution SC 1719/26, publishing a draft "Reglamento para el desarrollo y uso responsable de la Inteligencia Artificial" under Expediente Nº 3000-21862-2026. A companion resolution, RP 747/26, opened a public consultation running through August 7, 2026, inviting judges, lawyers, bar associations, and academics to weigh in before the court finalizes the rule (justicia.ar; SCBA).
The timing is notable. President Javier Milei has spent 2026 marketing Argentina as a deliberately unregulated AI haven — telling the Financial Times in a June 4 op-ed that his government would keep AI "sin la mano mortal de una regulación prematura y mal comprendida" and pitching Buenos Aires as "lo que Ámsterdam fue para la era de la navegación" (Buenos Aires Times). Three weeks later, a provincial court — a separate branch of government with its own constitutional authority over judicial administration — moved in the opposite direction for the one domain it controls directly: its own bench.
What the Draft Rule Actually Does
The SCBA's text is narrower than early headlines suggested. It does not touch AI use in the economy Milei is courting; it governs only how judges and court staff may use AI inside Buenos Aires provincial proceedings. Three provisions matter most:
- No delegated adjudication. The rule states the "función jurisdiccional es indelegable e insustituible" — judges cannot let an AI system weigh evidence, resolve legal questions, or draft the substantive reasoning of a ruling when that output becomes the "único o determinante" basis for the decision.
- Disclosure in the case file. Whenever generative AI assists drafting, the operator must record the system name, the model used, and which specific parts of the document were AI-assisted — creating a paper trail parties can challenge (La Nación).
- A flat ban on predictive-recidivism tools in detention decisions. Algorithmic risk-scoring systems cannot bind or determine rulings on pretrial liberty or detention (Infobae).
A new Comisión de Gobernanza y Uso de Inteligencia Artificial, drawn from the court's Planning, IT, and Jurisdictional Services secretariats, will collect consultation feedback and administer the rule once adopted. Until then it is explicitly "orientador y no vinculante" — guidance, not law.
The Case for the Rule
The strongest argument for this approach doesn't require distrusting AI generally — it requires taking seriously what pretrial detention and evidentiary rulings actually are: liberty-restricting state action that due process requires a named, accountable human to own. Recidivism-scoring tools have a documented history of encoding racial and socioeconomic bias into ostensibly neutral risk scores — the criticisms leveled at COMPAS in U.S. state courts are the reference case the SCBA is plainly writing against. A defendant facing detention cannot cross-examine a black-box model. Disclosure requirements exist for the same reason expert witnesses must be named: unattributed reasoning cannot be tested.
Why This Is Good Regulatory Design, Not a Contradiction
Where the SCBA gets it right is precision. It does not ban AI from courts — it bans AI from functions where the harm of an unreviewable error is irreversible (a person detained, a case wrongly decided) while leaving low-risk administrative uses, like docket management or legal research, largely untouched. That risk-tiered structure is the opposite of the blunt, precautionary bans that chill legitimate AI adoption elsewhere. It is also, notably, compatible with Milei's national posture rather than a rebuke of it: a government can credibly tell foreign AI firms "we won't regulate your product" while a court credibly tells its own judges "you will still sign your name to your rulings." Commercial AI deployment and judicial due process are different risk categories, and treating them identically — either by banning AI everywhere or regulating nowhere — would be the actual policy failure.
Spain's Consejo General del Poder Judicial adopted a broadly similar instruction on January 28, 2026, also requiring human oversight and barring AI from substituting judicial reasoning (CGPJ Instruction 2/2026) — suggesting this is converging judicial consensus, not a provincial outlier.
What to Watch
The consultation closes August 7, with the governance commission's recommendations due by August 31. Two things will determine whether this stays a model rule or becomes a cautionary one: whether the disclosure mandate gets specific enough to be enforceable ("which parts were AI-assisted" invites gaming without a stricter drafting standard), and whether the Comisión de Gobernanza is resourced to actually audit high-risk systems rather than rubber-stamp them. A rule that draws a sharp, narrow line around adjudicative functions while leaving everything else alone is exactly the kind of proportionate regulation this publication argues for — Buenos Aires's courts should be watched to see if they can make it stick.