Argentina - AI Trustworthiness Guidelines (2/2023)

Overview

The Disposición 2/2023 approves the annex "Recomendaciones para una Inteligencia Artificial Fiable", a guidance manual intended for national public sector actors who lead, develop, procure or operate systems that incorporate artificial intelligence. The instrument was sanctioned on 1 June 2023 and published in the Boletín Oficial on 2 June 2023. The annex frames AI as a tool that can improve public service delivery and administrative efficiency but also notes the potential for risks to rights and public values if projects are not governed appropriately. The Recommendations adopt a lifecycle approach, emphasize human-centred design, and take inspiration from international instruments and soft-law sources. For the official publication and text, see the national normativa portal and InfoLEG archival entry: Disposición 2/2023 – Argentina.gob.ar and InfoLEG – Anexo: Recomendaciones para una inteligencia artificial fiable.

Definitions

The Recommendations define and clarify key concepts used throughout the annex, including "Artificial Intelligence" (broadly as systems that use data and algorithms to perform tasks that would otherwise require human cognitive functions), "project lifecycle" (the stages from conception through operation and maintenance), "human oversight" (mechanisms to ensure humans retain control and responsibility), "risk assessment" and "impact assessment" (procedures to identify, evaluate and mitigate social, ethical and legal risks), and "data governance" (policies for collection, storage, access and provenance). The document relies on internationally recognised phrasing while adapting definitions to the needs of public administration projects. The definitions are designed to be operational for procurement, design and audit purposes.

Governance and Institutional Framework

Disposición 2/2023 is issued by the Subsecretariat of Information Technologies under the Jefatura de Gabinete de Ministros; the annex situates responsibility for implementation within the internal governance structures of each public body. It recommends establishing clear institutional roles—such as a designated project lead and responsible technical and ethical reviewers—and internal oversight mechanisms (e.g., ethics committees or review boards) to supervise AI initiatives. The guidance encourages coordination with existing national oversight actors, including data protection authorities and auditing bodies, to ensure compliance with privacy, administrative and financial rules. The document also urges that governance arrangements incorporate multidisciplinary capabilities (legal, technical, social, domain experts) and provide channels for stakeholder and citizen engagement. For the issuing authority and official publication, consult the government normativa page: Disposición 2/2023 – Argentina.gob.ar.

Key Focus Areas

The annex organises recommendations around several interlocking focus areas: ethical principles, risk and impact assessment, team composition, data and model governance, transparency and documentation, human oversight, testing and validation, operational controls and cybersecurity. Ethical principles include human-centredness, non-discrimination, fairness, explainability, privacy protection and proportionality. Risk management guidance stresses early-stage impact assessments, continuous monitoring, and documented mitigation plans. On data governance the annex sets out best practices for provenance, quality control and minimisation consistent with personal data protections. Transparency recommendations include producing explanatory documentation (e.g., model cards, datasets descriptions), disclosure of limitations and known biases, and accessible user information. Human oversight guidance prioritises human-in-the-loop or human-on-the-loop controls for high-stakes decisions, clear allocation of responsibilities and training for operators. Testing and validation recommendations call for pre-deployment evaluation in realistic conditions, robustness testing, and continuous validation during operation. Cybersecurity guidance addresses model security, access controls, logging and incident response measures. The manual organises these measures across four lifecycle stages—design and data modeling; verification/validation; implementation; and operation and maintenance—so that each focus area is applied where most relevant in the project timeline. The lifecycle and stage descriptions are summarised graphically in the second annex. The Recommendations have been discussed and summarized by legal and policy commentators and industry analysts, reflecting their influence on public-sector AI adoption practices.

Implementation Framework

The document proposes an implementation framework oriented around: (1) project initiation (needs assessment; designation of roles; preliminary risk screening); (2) design and procurement (requirements that embed ethical and legal safeguards; tendering language that requires documentation and auditability from vendors); (3) verification/validation (independent or internal testing, user acceptance tests, bias evaluations and robustness checks); (4) deployment (operational safeguards; human oversight; staff training); and (5) maintenance and monitoring (periodic re-evaluation, patching, retraining controls and incident response). The framework recommends integration with existing public procurement rules and internal controls, and encourages use of standardized artifacts (e.g., impact assessment forms, model cards, data inventories) to facilitate auditability and inter-agency comparability. Legal analyses note the document's emphasis on operationalizing abstract principles into checklist-style deliverables for public teams.

Monitoring and Evaluation

Monitoring guidance calls for continuous performance tracking, post-deployment audits, maintenance of logs and data lineage, and scheduled re-assessments of social and technical risk. The annex recommends metrics for fairness, accuracy, reliability and resilience, and suggests establishing internal reporting channels for identified harms or anomalies. It encourages public bodies to publish non-sensitive summaries of monitoring results to enable external oversight and build public trust. The Recommendations also signal the importance of aligning monitoring approaches with data protection obligations and internal audit cycles.

Penalties, Liability, and Appeals

As an advisory instrument, the Recommendations do not create explicit administrative fines or criminal penalties within their text; rather, they provide non-binding standards to guide public administration behavior. Non-compliance may still carry practical consequences: procurement decisions could be affected, internal disciplinary or corrective actions may follow, and responsible officials may face scrutiny in administrative audits. The annex emphasises assigning clear responsibility and maintaining documentation to support accountability and facilitate appeals or remediation when adverse impacts occur.

Relationship to Other Instruments

The annex draws on international soft-law and standards, positioning itself in relation to national laws and transversal instruments such as data protection rules (Law No. 25.326 and subsequent regulatory developments), public procurement and administrative law, and sectoral regulatory frameworks. It is intended to complement—rather than supplant—binding legal obligations, and to serve as an operational bridge between high-level principles and everyday project management practice in the public sector. The document explicitly cites international references and seeks interoperability with broader policy initiatives. (servicios.infoleg.gob.ar)

International Alignment

Disposición 2/2023 references international frameworks and best practices (drawing on UNESCO, OECD and other recognised sources) and recommends alignment with evolving regional and international standards. The guidance is designed to be compatible with cross-border cooperation, data-sharing arrangements and international procurement norms, facilitating interoperability and increasing the likelihood that public-sector projects will meet external expectations for trustworthy AI. Analysts highlight the instrument's role in signalling Argentina's policy alignment with global ethical AI trends and its potential to shape regional approaches.

Implementation Timeline

Sources and References

Disposición 2/2023 – "Recomendaciones para una Inteligencia Artificial Fiable" (Argentina.gob.ar) - Primary Source.
InfoLEG – Anexo: Recomendaciones para una inteligencia artificial fiable (InfoLEG) - Primary Source.
Boletín Oficial de la República Argentina (Boletín Oficial) - Official Gazette.