Latest

AI Agents

Argentina bans its prosecutors from entering case data into ChatGPT and other public AI tools

Argentina bans prosecutors from uploading case data into ChatGPT and other public AI tools and imposes rules for their use

Argentina bans its prosecutors from entering case data into ChatGPT and other public AI tools
Argentina bans its prosecutors from entering case data into ChatGPT and other public AI tools

Argentina's Public Prosecutor's Office has approved its first framework for regulating the use of artificial intelligence. The rule prohibits entering case data, reserved information or personal and sensitive data into public AI tools and establishes human oversight, audits, mandatory training and the progressive development of its own secure systems.

The Attorney General's Office approved the Institutional Guidelines for the Use and Implementation of Artificial Intelligence Tools, a document that for the first time sets out principles, procedures and clear limits for incorporating these technologies into the daily work of prosecutors, officials and employees.

The new rules were approved through Resolution PGN 64/2026, signed on September 4, 2026 by the Attorney General's Office. The decision comes as the use of generative artificial intelligence is already widespread within the body. An internal survey shows that ChatGPT is used by 90.2% of staff who turn to AI tools, far ahead of Gemini and Copilot.

Regulating a technology its professionals are already using

Precisely for this reason, the Public Prosecutor's Office has chosen to regulate a technology its professionals are already using rather than ignore its presence in daily tasks. The measure is part of the Artificial Intelligence Program created in 2025 through Resolution PGN 14/2025, whose goal is to design a strategic and comprehensive approach to the adoption of these tools within the body.

Since then, the program has mapped institutional needs related to AI use and carried out an anonymous internal survey between August and September 2025. Its results, published in August 2026, served as the empirical basis for drafting the new guidelines.

ChatGPT, in the daily work of the Public Prosecutor's Office

The internal survey, which reached about 40% of the body's staff, produced a particularly significant finding: generative artificial intelligence is already widely used among Public Prosecutor's Office staff, both for personal and work tasks. Among those who use these technologies, 90.2% use ChatGPT, followed by Gemini at 40.6% and Copilot at 16.7%.

The main uses relate to information searches, drafting and editing texts, and consulting doctrine and case law. Moreover, much of this adoption is happening outside institutionally contracted tools. Some 73% use free versions, while another 23% pay for personal subscriptions with their own funds.

The results therefore reflect a de facto adoption of artificial intelligence that predates the existence of a specific framework to regulate it. But the survey also showed very broad support for ending that use without clear institutional rules.

Some 93.3% of staff said they favored moving toward institutional use of AI, albeit with important safeguards. Some 77.7% called for guarantees of constant human oversight; 76.3% asked for clear rules on permitted and prohibited uses; 74% demanded strict protection of privacy and sensitive data, and a resounding 90.4% requested a comprehensive practical training plan.

These numbers, as the resolution's own text explains, served as institutional capital to turn that reflective caution among staff into a sustainable technology governance model.

Banned from entering case data into ChatGPT and other public AI tools

One of the most important decisions in the new regulation takes immediate effect and directly affects the use of public artificial intelligence tools. Until a specific regime is issued for individual use of publicly accessible AI tools, Public Prosecutor's Office staff may not enter into them reserved information, data from pending cases, personal or sensitive data, or metadata that could compromise confidentiality or allow individuals to be re-identified.

The prohibition responds to one of the main risks posed by the use of commercial generative artificial intelligence models in the judicial sphere: that especially sensitive information leaves the systems controlled by the institution itself.

The document also expressly warns of another problem: users must not use publicly accessible chatbots as reliable search engines or authorized documentary sources. According to the guidelines, these tools draw on material published on the internet, prioritize linguistic probability over legal accuracy and can generate content that appears convincing but is false, incomplete, outdated or biased.

For this reason, staff must always cross-check any result obtained through artificial intelligence against official sources and may never use it as the sole basis for a decision.

Ten principles for governing artificial intelligence

The normative core of the new framework consists of ten guiding principles inspired by international standards such as those of the OECD, UNESCO, the OAS and the U.S. National Institute of Standards and Technology (NIST), as well as national guides prepared by the Chief of Cabinet of Ministers and the Agency for Access to Public Information.

These principles include legality and respect for fundamental rights; justice, fairness and non-discrimination; public purpose and efficiency, and technology governance based on responsible innovation. One of the fundamental pillars is meaningful human oversight. Artificial intelligence systems may assist professionals in tasks such as analyzing or classifying information, but in no case may they replace legal judgment or regulatory interpretation.

These competencies will remain the exclusive responsibility of Public Prosecutor's Office staff. The internal survey shows that this oversight is already part of how most workers currently use AI.

Some 96% of staff use responses generated by artificial intelligence as input that they subsequently review or adapt. Specifically, 60.5% draft from scratch based on the information obtained and 35.5% adapt what is generated. Barely 0.4% use AI responses directly without review.

The guiding principles are completed by requirements for transparency and traceability of systems, proportionality and security in their deployment, protection of personal data in accordance with Law 25.326, reliability of the data used and institutional responsibility. This last point means that each staff member will be responsible for their use of artificial intelligence tools, even when these are provided by third parties.

Every AI project will have a regulated life cycle

The new guidelines are not limited to setting out general principles. They also establish a concrete procedure that any institutional project based on artificial intelligence must follow. The process begins with identifying a specific need. Before developing or procuring an AI-based solution, it must be assessed whether this technology is really the most appropriate tool to solve the problem at hand or whether simpler alternatives with lower risks exist.

A planning stage then begins in which it will be mandatory to classify the data to be used, analyze its sources and quality and carry out an impact assessment. That analysis will determine the project's risk level, which may be classified as unacceptable, high, limited or minimal, following categories inspired by international standards such as those used in the European Artificial Intelligence Regulation.

Once the project is approved, the system may be developed internally or acquired from external providers. The resolution considers internal development preferable, since it allows greater control over data. When an external solution is chosen, contractual terms must guarantee the possibility of carrying out audits and establish effective human oversight mechanisms.

Nor does the regulation end when the tool begins to be used. After implementation, the system must go through continuous stages of maintenance, oversight and auditing, under the responsibility of the General Directorate of Organizational Development and New Technologies and, at later control stages, of the Internal Audit Unit.

Mandatory training to learn to question AI

The Public Prosecutor's Office does not limit itself to establishing prohibitions, controls and procedures; it also considers specific training for its workers necessary. The resolution tasks the General Directorate of Training and the Public Prosecutor's Office School, in coordination with the Secretariat of Institutional Coordination, with designing a Permanent Program of Practical and Applied Training in Artificial Intelligence.

The training must provide staff with concrete operational skills for working with these technologies. These include formulating prompts, critically validating the results provided by the models and identifying possible algorithmic biases.

The training also responds to a widespread demand within the institution itself: 90.4% of participants in the internal survey called for a comprehensive practical training plan.

Its own AI tools

Another relevant element of the resolution is the commitment to progressively reducing dependence on commercial platforms when working with especially sensitive information. The Attorney General's Office tasks the General Directorate of Organizational Development and New Technologies with the progressive development of its own secure environments and tools.

These systems must meet the data analysis and document management needs of the body's different departments and comply with the standards set out in the new guidelines. The underlying commitment is clear: reduce dependence on publicly accessible commercial tools wherever sensitive data related to judicial cases is at stake.

The decision is especially relevant given that most of the body's AI users currently rely on free versions and that some staff even pay for subscriptions to these platforms out of their own pockets.

Periodic review and updating

The Public Prosecutor's Office acknowledges that these first rules will hardly remain unchanged given the speed at which artificial intelligence evolves. The document itself warns that the guidelines should not be understood as closed and definitive regulation, but as the starting point of a process projected to be disruptive for the body's practices.

The Attorney General's Office assumes that the pace of technological developments will require periodically reviewing and updating the regulatory framework. However, it establishes from the outset a boundary meant to hold regardless of how technology evolves: artificial intelligence may assist prosecutors and officials, but not replace their responsibility, their legal judgment or the guarantees surrounding the functioning of justice.

The premise running through the document sums up that philosophy: technology must always serve justice, not justice serve technology.

With Resolution PGN 64/2026, Argentina's Public Prosecutor's Office thus joins a growing trend among judicial and justice bodies in the region seeking to organize, rather than prohibit, a use of generative artificial intelligence that has already spread among their professionals.

Alfredo Collosa

Author

Alfredo Collosa

Alfredo Collosa is a tax specialist. Public accountant. Bachelor's degree in Business Administration. Official Master's in Public Finance and Tax Administration from UNED/IEF in Spain. AMLCA Anti-Money Laundering Certificate (Florida University, US) and member of MetaLaw

The latest
From Huang to Botín, from Zara to Byron: corporate responsibility in global capitalism

Keep reading