Argentina - Buenos Aires - Generative AI Guidelines

Overview

The Buenos Aires Province Generative AI Guidelines, formally enacted through Resolution 4/2025 by the Undersecretariat of Digital Government, represent a landmark regulatory step in Argentina's provincial governance. As the first jurisdiction in the country to provide specific, binding instructions for the use of Generative Artificial Intelligence (GenAI) within the public sector, the Province of Buenos Aires (PBA) seeks to balance the immense potential for administrative efficiency with the critical need to protect citizen rights and data integrity. The resolution was prompted by the rapid proliferation of large language models (LLMs) and other generative tools among public employees, necessitating a centralized framework to prevent the mishandling of sensitive state information and to ensure that AI adoption aligns with democratic values and provincial legal standards. The guidelines are part of a broader digital transformation strategy aimed at modernizing the state while maintaining technological sovereignty. This initiative recognizes that while GenAI can significantly reduce the time spent on repetitive administrative tasks, it also introduces novel risks related to misinformation, algorithmic bias, and the potential exposure of confidential government data. By establishing these guidelines, the Ministry of Government provides a clear roadmap for public agents, ensuring that innovation does not come at the cost of legal certainty or ethical responsibility. The scope of these guidelines is comprehensive, applying to all personnel within the Provincial Public Administration, regardless of their contractual status or hierarchy. This includes central administration, decentralized bodies, and autarchic entities. The primary objective is to establish a set of 'minimum standards' that guide public officials in the responsible deployment of GenAI for tasks such as drafting documents, summarizing reports, and improving citizen communication. By setting these boundaries, the Provincial Government aims to foster a culture of 'demonstrated responsibility,' where innovation is encouraged but strictly governed by ethical principles, transparency, and the non-negotiable requirement of human oversight to mitigate the risks of algorithmic bias and misinformation.

Definitions

The resolution provides a rigorous technical and legal lexicon to ensure uniform application across different ministries. At its core, it defines Generative Artificial Intelligence (IAGen) as a subset of AI technologies designed to create novel content—ranging from textual prose and computer code to synthetic imagery—derived from patterns learned during training on vast datasets. This definition distinguishes GenAI from traditional analytical AI, emphasizing its creative and probabilistic nature, which introduces unique risks such as 'hallucinations' or the generation of factually incorrect but plausible-sounding information. The guidelines explicitly address these technical nuances, requiring users to understand that GenAI outputs are statistical predictions rather than verified facts. Furthermore, the document defines key operational terms such as 'Human Supervision' and 'Algorithmic Transparency.' Human supervision is characterized as a mandatory 'human-in-the-loop' mechanism where no administrative act or decision can be solely based on an AI-generated output without a prior and posterior review by a competent public official. Transparency is defined not just as the disclosure of AI use, but as the 'intelligibility' of the system—ensuring that the processes, data sources, and logic used by the AI can be explained to and understood by the citizens affected by its use. Other critical definitions include 'Prompt Engineering,' defined as the process of optimizing input text to guide AI models toward desired outputs, and 'Synthetic Content,' referring to any data or media generated or significantly altered by AI. The resolution also clarifies the concept of 'Data Anonymization' within the context of AI, specifying that it involves the irreversible removal of personal identifiers to prevent the re-identification of individuals by the AI model or its providers. These definitions serve as the foundation for the compliance obligations detailed in the subsequent sections of the resolution, providing a common language for technical and legal staff alike.

Governance and Institutional Framework

The governance of AI within the Province of Buenos Aires is centralized under the Ministry of Government, specifically through the Undersecretariat of Digital Government. This body acts as the primary regulatory authority, responsible for interpreting the guidelines, issuing technical updates, and overseeing the implementation of AI tools across the provincial state. Under Resolution 4/2025, the Provincial Directorate of Digital Innovation is tasked with providing the technical support and training necessary for departments to comply with the new standards. This institutional structure ensures that AI adoption is not fragmented or siloed, but rather follows a cohesive provincial strategy focused on technological sovereignty and administrative modernization. A critical component of this framework is the requirement for hierarchical authorization. Before any public agent can utilize a GenAI tool for official duties, they must obtain formal approval from their immediate superior. This creates a chain of accountability, ensuring that department heads are aware of and responsible for the tools being used within their units. Additionally, the resolution paves the way for the 'Provincial Registry of Artificial Intelligence Systems,' a centralized database where all AI implementations must be recorded. This registry allows the Undersecretariat to monitor the provincial 'AI footprint,' evaluate the cumulative impact of these technologies on public services, and conduct audits to ensure adherence to the established ethical and security protocols. The framework also establishes a 'Technical Advisory Committee' composed of experts in data science, law, and ethics, who will provide ongoing recommendations to the Undersecretariat. This committee is responsible for reviewing high-risk AI applications and ensuring that the province stays abreast of rapid technological shifts. The governance model emphasizes a top-down policy setting combined with bottom-up implementation, allowing individual ministries to innovate within the safety parameters established by the central digital authority.

Key Focus Areas

The guidelines prioritize four key pillars: Ethics, Transparency, Data Protection, and Human Oversight. Ethically, public agents are mandated to use GenAI in a manner that respects the dignity, equality, and fundamental rights of all individuals. This includes a proactive duty to identify and mitigate algorithmic biases that could lead to discriminatory outcomes in public service delivery. The resolution emphasizes that AI should be a tool for inclusion, not a mechanism for automating existing social or systemic prejudices. Transparency is operationalized through the requirement that any interaction between a citizen and an AI-mediated system (such as a chatbot) must be clearly labeled, ensuring that the individual is aware they are not communicating with a human in the first instance. Data protection and confidentiality represent perhaps the most stringent focus area. The guidelines strictly prohibit the input of sensitive personal data, classified state information, or confidential legal documents into public or third-party GenAI platforms (e.g., ChatGPT, Gemini) unless they are part of a secured, provincial-controlled environment. Before any data is shared with these systems, it must undergo a rigorous anonymization process to ensure that no individual can be identified from the prompts or training inputs. This focus area is directly linked to Argentina's National Personal Data Protection Law (Law 25.326), ensuring that provincial AI use does not circumvent established privacy safeguards. Furthermore, the guidelines address 'Intellectual Property,' stating that AI-generated content used in official documents must be properly attributed and that the state must ensure it has the rights to use the training data or the resulting outputs. Another focus area is 'Environmental Sustainability,' encouraging the selection of AI models and providers that demonstrate energy efficiency and a commitment to reducing the carbon footprint of large-scale computation. These focus areas collectively ensure that AI adoption is holistic, addressing not only technical performance but also social, legal, and environmental impacts.

Implementation Framework

Implementation is designed as a phased and supervised process. The resolution outlines that public entities wishing to integrate GenAI must first conduct an internal assessment of the tool's necessity and potential impact. This 'Impact Assessment' is crucial for identifying whether the intended use case falls into a high-risk category, such as those involving the allocation of social benefits or the processing of legal evidence. The framework encourages the use of pilot programs and 'sandboxes' where AI tools can be tested in a controlled environment before full-scale deployment, allowing the Provincial Directorate of Digital Innovation to evaluate performance and security vulnerabilities. To support this implementation, the Province has committed to a continuous 'Digital Literacy' program. Recognizing that the primary risk in AI adoption is often human error or over-reliance, the guidelines mandate ongoing training for all public employees. This training covers prompt engineering, ethical evaluation of outputs, and data security best practices. The goal is to move beyond mere usage and toward a sophisticated understanding of AI's limitations. By empowering public agents with these skills, the Province ensures that GenAI serves as a genuine productivity multiplier rather than a source of administrative liability or technical debt. The implementation framework also includes specific 'Technical Guidelines for Procurement,' which provide procurement officers with the criteria necessary to evaluate AI vendors. These criteria include data residency requirements (preferring local or secure cloud storage), transparency of training datasets, and the availability of API-based controls that allow the province to maintain oversight of data flows. By integrating these requirements into the procurement process, the province ensures that third-party tools meet the same rigorous standards as internally developed systems.

Monitoring and Evaluation

Monitoring is established as a continuous, rather than periodic, obligation. The Undersecretariat of Digital Government is empowered to conduct regular audits of AI systems registered within the provincial administration. These audits focus on 'Traceability' and 'Auditability,' requiring that the logs of AI interactions and the data used for fine-tuning models be preserved for review. Evaluation metrics include the accuracy of AI outputs, the frequency of human corrections required, and the impact on service delivery times. If a system is found to be producing biased or consistently erroneous results, the Undersecretariat has the authority to order its immediate suspension or modification. Furthermore, the resolution encourages a feedback loop involving both public employees and the citizens they serve. Public agents are required to report any 'anomalous behavior' or significant errors produced by GenAI tools to the Provincial Directorate of Digital Innovation. This bottom-up reporting mechanism is vital for identifying systemic issues that might not be captured by top-down technical audits. On the citizen side, the transparency requirements ensure that individuals have the right to challenge decisions made with the assistance of AI, triggering a mandatory human review process that serves as both a quality control measure and a safeguard for administrative due process. The evaluation process also includes an 'Annual AI Impact Report,' which will be published by the Ministry of Government. This report will summarize the state of AI adoption across the province, highlighting successful use cases, identifying common challenges, and proposing updates to the guidelines based on observed performance. This commitment to public reporting ensures that the government remains accountable for its technological choices and provides a basis for evidence-based policy adjustments in the future.

Penalties, Liability, and Appeals

Resolution 4/2025 clarifies that the use of AI does not absolve public officials of their administrative, civil, or criminal responsibilities. The principle of 'Human Responsibility' is paramount: any error, data breach, or discriminatory act resulting from the use of GenAI is legally attributed to the human official who authorized or failed to supervise the system. This ensures that the 'black box' nature of AI cannot be used as a legal defense for administrative failures. Violations of the guidelines—such as the unauthorized use of AI tools or the input of sensitive data into unsecured systems—can lead to disciplinary sanctions under the provincial public employment statutes (Law 10.430 and related regulations). These sanctions range from formal warnings and suspensions to dismissal in cases of gross negligence or intentional data leaks. Regarding citizen rights, the guidelines establish a clear pathway for appeals. Any administrative act that has been significantly influenced by a GenAI tool must be explicitly identified as such, and the affected party has the right to request a full human re-evaluation of the decision. This right to 'human intervention' is a cornerstone of the provincial AI strategy, ensuring that the automation of the state does not lead to the dehumanization of administrative justice. The appeals process follows the standard provincial administrative procedure, but with the added requirement that the state must be able to explain the logic of the AI's contribution to the final decision in an understandable manner. If the state cannot provide a clear explanation for an AI-influenced decision, that decision may be deemed arbitrary and subject to annulment by the provincial courts. This legal framework places the burden of proof on the administration to demonstrate that AI use was both necessary and compliant with all ethical and procedural safeguards.

Relationship to Other Instruments

The Buenos Aires Province Generative AI Guidelines do not exist in a vacuum but are part of a broader legal ecosystem. They are designed to complement the National Law on Personal Data Protection (Law 25.326) and the Provincial Law on the Modernization of the State (Law 14.828). While the guidelines focus specifically on GenAI, they are reinforced by the subsequent Resolution 9/2025, which established a broader risk-based framework for all AI systems, categorizing them from 'unacceptable risk' to 'minimal risk' in alignment with international standards like the EU AI Act. This layered approach ensures that the specific challenges of generative models are addressed without losing sight of the general principles governing all automated systems. Additionally, the resolution references the 'Malvinas Argentinas' Normative and Documentary Information System, which serves as the official repository for all provincial regulations. By integrating the AI guidelines into this system, the Province ensures that they are easily accessible to all public officials and legal researchers. The guidelines also interact with the provincial Cybersecurity Protocols, which provide the technical infrastructure requirements for hosting AI models on provincial servers. This inter-instrumental alignment ensures that AI adoption is consistent with the Province's broader goals of technological sovereignty, data security, and administrative transparency. Furthermore, the guidelines are designed to be compatible with the 'Federal Digital Agenda' of Argentina, ensuring that provincial efforts contribute to a unified national approach to AI governance. This coordination is essential for ensuring that data shared between provincial and national entities remains protected under a consistent set of rules, regardless of the specific AI tools employed by different levels of government.

International Alignment

In drafting Resolution 4/2025, the Province of Buenos Aires explicitly looked toward international best practices to ensure that its local regulations are globally competitive and ethically sound. The guidelines are heavily influenced by the OECD Recommendation on Artificial Intelligence and the UNESCO Recommendation on the Ethics of Artificial Intelligence. These international instruments emphasize the importance of human-centric AI, transparency, and the protection of human rights—principles that are woven throughout the provincial resolution. By aligning with these standards, the Province signals its commitment to a 'responsible innovation' model that is recognized by international partners and investors. Furthermore, the resolution mirrors the risk-based approach pioneered by the European Union's Artificial Intelligence Act (EU AI Act). Although the PBA guidelines are focused on the public sector, they adopt similar classifications for high-risk applications and mandate impact assessments that resemble the EU's requirements for algorithmic transparency. This alignment is not merely theoretical; it serves a practical purpose by ensuring that AI tools developed or procured by the Province meet high international standards, facilitating cross-border cooperation and the potential for the Province to act as a regional leader in AI governance within Latin America. The guidelines also incorporate elements from the 'G7 Hiroshima AI Process,' particularly regarding the prevention of disinformation and the protection of intellectual property in the age of generative models. By participating in this global regulatory dialogue, the Province of Buenos Aires ensures that its public administration is equipped with tools that are not only efficient but also resilient to the global challenges posed by rapid AI advancement. This international perspective helps the province avoid 'reinventing the wheel' and instead builds upon a foundation of globally recognized ethical and technical norms.

Implementation Timeline

Sources and References