Argentina - AI Regulation Bill (1747/23)

Overview

Expediente 1747/23, introduced in the Argentine Senate on 14 August 2023 by Senator Juan Carlos Romero, proposes a national regulatory framework to "control the development, implementation and utilization of systems based on Artificial Intelligence" inside Argentina. The bill embraces a risk-based approach to AI governance, establishes five guiding principles (inclusive growth and welfare; transparency and explainability; robustness, security and protection; collaboration and human supervision; responsibility and accountability), and creates institutional tasks for a national authority to manage a National Registry of AI Systems and to perform pre-deployment Impact Assessments. The draft targets both public and private actors (developers, deployers, providers) and distinguishes between limited/minimal-risk, high-risk and unacceptable uses. It places specific constraints on biometric identification and techniques that manipulate behaviour without consent, while requiring documentation, audits, human oversight mechanisms and data governance safeguards. The primary registry and the bill text are published on the official Senate portal and the expediente was later reproduced under expediente 71/25 for continued handling by commissions in 2025. Official Senate metadata and the downloadable parliamentary text are accessible on the Senate website (original text stored as parlamentaria PDF).

Definitions

The proposed law defines key terms to anchor scope and obligations: "Artificial Intelligence systems" (systems that process data and produce outputs such as classification, prediction, recommendation or decision-support by algorithmic means); "provider" (the natural or legal person that develops or produces an AI system and places it on the market); "operator/deployer" (entity that deploys, operates or uses an AI system); "impact assessment" (a structured pre-deployment evaluation that examines risks, bias, discrimination, privacy and safety issues); "registry" (the National Registry of AI Systems where providers/deployers must register systems above predefined risk thresholds); and risk categories (limited/minimal, high, unacceptable). Definitions align conceptually with international instruments referenced in the bill, including OECD and UNESCO guidance, to ensure that risk and harm analyses consider social, economic and human rights dimensions.

Governance and Institutional Framework

The draft designates the Ministry of Science or a similarly empowered body as the primary enforcing authority, tasked with maintaining the National Registry, conducting or overseeing Impact Assessments, authorizing high-risk system deployment, and coordinating cross-sectoral supervision. Technical advisory committees and inter-agency coordination (including data protection authorities and sectoral regulators) are anticipated by the text to ensure sector-specific expertise for decisions affecting critical infrastructure, public safety, health or education. The authority may require conformity assessments, audits, transparency reports and may issue binding remedial orders. The bill foresees rulemaking powers to develop implementing regulations, procedural rules for impact assessments, and operational guidance for compliance; it also contemplates public consultation steps for regulatory norms. For official record and legislative progress see the Senate parliamentary entry and downloadable text on the Senate portal (senado.gob.ar - Expediente 1747/23).

Key Focus Areas

The bill concentrates obligations and controls across several domains: (1) Risk classification — systems are categorized by potential for harm and public impact, with stricter controls on high-risk systems and prohibition of unacceptable uses; (2) Registration and pre-market evaluation — a National Registry and mandatory Impact Assessments for systems above a threshold; (3) Transparency and documentation — providers must supply model descriptions, training data provenance, performance metrics, and interpretability explanations to the authority and, where appropriate, to impacted individuals; (4) Human oversight and operational constraints — mandatory human-in-the-loop/ human-on-the-loop arrangements for certain automated decision processes and explicit limits on fully autonomous decisions in sensitive domains; (5) Data protection and privacy — obligations to align with Argentina’s data protection norms, minimize personal data usage, and adopt privacy-preserving design choices; (6) Robustness, cybersecurity and model security — technical controls to prevent tampering, leakage and adversarial manipulation; (7) Conformity assessment, auditability and recordkeeping — periodic independent audits and retention of logs for accountability; (8) Restrictions and prohibitions — rules against subliminal manipulation and tight limitations on remote real-time biometric identification in public spaces except under narrowly defined exceptions. These focus areas reflect and adapt internationally recognized trustworthy AI requirements while being tailored for national institutions.

Implementation Framework

Implementation relies on a multi-stage administrative regime: registration of covered systems, submission of an Impact Assessment for review, a clearance/authorization or conditional approval for high-risk systems, and ongoing monitoring through periodic reporting obligations and audits. Providers and deployers must maintain technical documentation, internal risk management processes, and channels for redress. The authority is empowered to set templates for Impact Assessments, to require corrective actions (model retraining, limits on deployment), and to coordinate with sector regulators (health, transport, education, justice). The bill contemplates capacity-building measures within the public administration to evaluate technical dossiers and a phased rollout of registry and assessment obligations to allow time for compliance by small and medium enterprises.

Monitoring and Evaluation

The proposal establishes supervisory powers including on-site inspections, requests for information, periodic reporting obligations, and the ability to commission independent audits. Performance indicators for oversight will include rates of registration, outcome of impact assessments, incidents reported (bias/discrimination/security breaches), and compliance with corrective orders. The authority must publish aggregated statistics and non-confidential summaries of impact assessments to promote transparency and public trust. Coordination with the National Data Protection Authority and sectoral supervisors is required for cases implicating personal data or regulated sectors.

Penalties, Liability, and Appeals

Enforcement tools include administrative sanctions (fines and orders to suspend or withdraw systems from the registry or the market), mandated corrective measures (model re-design, data remediation), and public naming/notification of non-compliant actors. The draft also affirms that legal responsibility remains with natural and legal persons (providers, deployers), not with the AI as an entity; liability and redress frameworks are to be enforced through administrative procedures supplemented by civil or criminal remedies where existing law applies. The bill foresees evidentiary obligations to facilitate investigations and preserves legal avenues of appeal against administrative decisions in accordance with Argentine administrative law.

Relationship to Other Instruments

The project positions itself to complement existing national instruments: Argentina’s data protection regime, national science & technology policies, sectoral safety rules, and consumer protection law. It calls for alignment with the National Strategy or programs on AI, and coordination with the Data Protection Authority for privacy-intensive matters. It also contemplates regulatory cross-referrals to sectoral regulators (health, transport, financial authorities) where overlapping competencies exist. The bill is explicit that it does not supplant sector-specific safety rules but provides baseline obligations and a supervisory architecture for algorithmic systems across sectors.

International Alignment

Expediente 1747/23 references international standards and instruments (OCDE, UNESCO recommendations, European regulatory approaches) as normative guidance to ensure interoperability and to attract compliance alignment for providers operating across borders. The approach mirrors risk-tiering and pre-deployment assessments found in EU AI regulatory debates and emphasizes reciprocity in conformity and assessment mechanisms to facilitate trade and technical cooperation. The draft encourages cooperation with international bodies and inclusion in multilateral AI governance fora to exchange best practices for audits, impact assessment methodologies and cross-border enforcement assistance.

Implementation Timeline

Sources and References