Argentina - AI Use Guidelines (206/2025)

Overview

The Resolución de Presidencia 206/2025 approves, as an annex, the "Guía de Recomendaciones y Directrices para el Uso de Sistemas de Inteligencia Artificial (IA) en el Poder Judicial de la Ciudad Autónoma de Buenos Aires." The Guide adopts a risk-based, human-centric framework that permits use of AI as an assistant in judicial processes while preserving the non‑delegable role of judges and judicial personnel. It requires prior classification of use-cases, mandatory impact assessments for high‑impact systems, and documentation and auditability obligations for deployed models. The decision and its annex are published and referenced on the Council's official portal; see the Council's committee page for the AI standardization committee where the resolution and its annex are listed as the normative basis for the Committee's work: Consejo de la Magistratura (CABA) — AI committee.

Definitions

The Guide defines core concepts for judicial application: "Artificial Intelligence System" (any software or tool using machine learning, statistical methods, or rules/heuristics to perform tasks normally requiring human intelligence); "Generative AI" (models producing novel text, images, audio or other content); "Human Oversight" (designated, documented human control over deployment, outputs, or final decisions); "High-impact Use" (deployments likely to materially affect fundamental rights, procedural guarantees, liberty, economic interests or access to justice); "Audit Trail" (immutable logs of inputs, prompts, outputs and system versions); and "Data Protection Impact Assessment / DPIA" (systematic analysis of privacy risks and mitigations). The Guide clarifies that multipurpose models used beyond their intended scope must be re-evaluated prior to new deployments.

Governance and Institutional Framework

The resolution assigns governance responsibilities primarily to the Consejo de la Magistratura and its Secretariat of Administration and Budget, while establishing an inter‑disciplinary Committee for AI Standardization that includes representatives from Innovation, IT, key Chambers and invited public prosecutors and the Tribunal Superior de Justicia. The Guide requires that every AI initiative must have a sponsoring judicial authority, a named technical owner, and a documented implementation plan. Procurement processes must include AI-specific vendor due diligence, contractual clauses for documentation, model explainability and access to logs, and obligations to notify data-protection authorities when personal data processing is involved. Implementation oversight is performed through internal registries and prior authorization for high-impact systems; the Committee reviews technical/economic risk matrices and issues recommendations. For the Committee's formation and legal basis, the Council's page lists the Guide as Annex 1 to Resolución 206/2025: Committee integration and normative framework.

Key Focus Areas

The Guide concentrates on several interlocking areas: (1) Human oversight and the prohibition on delegating adjudicative functions to AI; (2) Risk management — including classification of use-cases, impact assessments, mitigation plans, and thresholds for authorization; (3) Transparency and user information — clear notices in proceedings when AI tools are used and publication of high‑level technical descriptions for public scrutiny; (4) Data protection and privacy — DPIAs, data minimization, purpose limitation and secure deletion; (5) Safety, testing and validation — pre-deployment testing datasets, bias and robustness tests, and continuous performance monitoring; (6) Documentation and accountability — model cards, data lineage, versioning, prompt registries and audit trails; (7) Cybersecurity and model/infrastructure protections — access controls, encryption and incident response; (8) Capacity building — mandatory training for judicial staff; and (9) Vendor management and procurement safeguards. The Guide also addresses generative-model specific concerns (hallucinations, provenance, misuse) and requires additional controls for such systems. Legal and technical recommendations in the Guide reflect comparators such as provincial judicial protocols and international guidance noted in legal commentary (see legal analyses linked in Sources).

Implementation Framework

The Guide prescribes a staged implementation model: (A) Identification and classification of proposed AI use-cases (neutral/low/medium/high impact); (B) For medium/high impact, completion of a DPIA and a technical safety assessment; (C) Procurement with contractual obligations (documentation, logs, remedy clauses, data-processing agreements); (D) Pilot testing in controlled environments with defined acceptance criteria; (E) Authorization by the competent Council authority (and registration in the internal AI registry); (F) Production deployment under monitoring, with mandatory logging and reporting; and (G) Periodic re-assessment or re‑authorization upon significant change in the model or use-case. The Guide requires that all systems be auditable and that outputs used operationally be accompanied by human review and retention of the decision rationale to enable challengers to seek review.

Monitoring and Evaluation

Monitoring responsibilities are divided between the AI Committee (policy and oversight), IT/security units (technical monitoring and incident response), and sponsoring judicial units (operational compliance and user training). Required monitoring includes continuous performance metrics (accuracy, false positive/negative rates), bias and fairness metrics for protected classes, security monitoring, and periodic re‑testing after model updates or drift. The Guide mandates maintenance of an internal registry of authorized systems and annual public reporting on high-impact deployments and their oversight outcomes. It also contemplates independent audits when necessary and recommends publication of sanitized model cards for public transparency.

Penalties, Liability, and Appeals

The Guide establishes an enforcement ladder: remedial measures (suspension or rollback of a deployment), administrative sanctions in accordance with Council disciplinary rules, referral to disciplinary bodies for judicial staff where obligations are breached, contractual remedies against vendors (including termination and damages), and obligations to remediate affected data subjects. The instrument reiterates that legal responsibility for judicial decisions remains with human decision‑makers; AI cannot be the sole basis for a final adjudication. It also creates procedural mechanisms for affected parties to request human review and to file complaints with internal oversight units; these complaints may trigger audits and corrective measures.

Relationship to Other Instruments

The Guide situates itself within the existing legal architecture: it references the Constitution of the City of Buenos Aires, Law No. 31 (functions of the Council), prior Council instruments such as Resolución CM N°108/2021 (Program for AI Modernization), and provincial judiciary protocols. It is designed to operate in conformity with Argentina's data-protection regime (Law No. 25.326) and with national procurement rules, while aligning with international guidance (e.g., UNESCO draft guidelines and other judicial AI protocols). The Guide functions as internal, sectoral regulatory guidance for the CABA judiciary and can be complemented by legislation or further Council resolutions as needed.

International Alignment

The Guide explicitly cites international standards and comparative initiatives as inspiration and reference points (UNESCO, provincial judicial protocols, EU AI Act principles in commentary). It prioritizes best-practice alignment on human oversight, risk classification, DPIAs and transparency, aiming to make the City judiciary's approach interoperable with commonly proposed international safeguards. Where contradictions arise between domestic law and international guidance, the Guide emphasizes conformity with Argentine law and judicial independence while recommending voluntary alignment with high‑quality international models for technical controls and auditing.

Implementation Timeline

Sources and References