AI Regulation Tracker  /  Courts and liability

Argentina Appellate Court Strips Lawyers of Fees and Refers Them to the Bar Over AI-Fabricated Case Citations

On September 19, 2025, the Cámara de Apelaciones in General Roca, Río Negro found 26 nonexistent case citations across two lawyers' briefs in a consumer appeal, called the conduct a grave irregularity, denied both lawyers their fees for the appeal as inoficiosa, and referred them to the bar's ethics tribunal. The court held that generative-AI output must be checked against real sources before it goes into a filing. This is a real appellate ruling from last year, not a new one, and I am flagging it as an evergreen marker of where the duty is heading, not as breaking news.

By Anthony Guerriero  ·  The Leveraged Years Regulatory Research  ·  Published July 20, 2026  ·  Last updated July 20, 2026

The underlying dispute was ordinary. A consumer sued over a zero-kilometer vehicle that turned out to be defective, naming Peugeot Citroën Argentina and the official dealership, DMX2. The case went up on appeal, and that is where it stopped being ordinary. When the appellate judges went to check the authorities the parties had cited, the authorities were not there. Not misquoted, not taken out of context. They did not exist.

Twenty of the phantom rulings came from the plaintiff's lawyer. Six came from the dealership's side. Twenty-six fabricated citations in a single appeal, from lawyers on opposite ends of the table. The court did not have a signed confession that a chatbot wrote them, and it said so plainly. But it treated unverified generative AI as the most probable source, which is the sensible read when two independent briefs both cite a pile of cases that were never decided.

What the court actually did

This is the part that separates the ruling from a stern footnote. The court did not simply frown at the lawyers and move on. It reached for consequences that a working attorney feels.

First, it declared the lawyers' professional intervention in the appeal inoficiosa. In practice that means their work on that stage of the case is treated as having no legitimate value, so they earn no fees for it. In a fee-shifting system where the appeal is often where the money is, that is a real financial hit, and the court applied it to both sides. Second, it issued a formal reprimand. Third, and most durable, it referred the lawyers to the Colegio de Abogados, the bar association, and its ethics tribunal, so the professional-discipline process could take its own look. The court also asked the bar to push its members toward good practice, meaning rigorous verification of every citation before it goes into a filing.

The holding worth quoting

The reasoning is what makes this exportable. The judges did not treat AI as forbidden. They treated it as a tool that produces confident text you cannot trust on sight. In the language reported from the decision, chatbot results "deben ser indefectiblemente cotejados," they must be inescapably cross-checked, because these systems often produce what the court called "las denominadas 'alucinaciones,'" so-called hallucinations, that surface as "fuentes lisa y llanamente inexistentes," sources that are flatly nonexistent and invented.

From there the court drew the line that matters for every lawyer reading this. Even assuming good faith, even assuming the lawyers did not intend to deceive anyone, that does not relieve them of the professional duty to verify that the sources they hand a court actually exist and say what they claim. Good faith is not a substitute for checking. The output of the machine is not the end of your work, it is the beginning of it.

Why a US lawyer should care

An Argentine provincial appellate court does not bind anyone in the United States, so the direct legal effect on your practice is zero. The reason to read it is that the reasoning is already your reasoning, and it is converging across systems.

US lawyers have watched this exact story unfold at home since Mata v. Avianca, where a Manhattan federal judge sanctioned lawyers who filed a brief full of ChatGPT-invented cases. Since then the sanctions orders have piled up under Rule 11 and the courts' inherent authority, in trial courts and now at the appellate level, with fee awards, striking of filings, referrals to disciplinary bodies, and public naming. What the General Roca court adds is not a new idea, it is confirmation that the idea is not a US quirk. Put the two systems side by side and the rule is the same: you sign it, you own it, and an AI hallucination in your citation is your professional problem, not the model's.

The practical translation is boring and that is the point. If you use a chatbot for research or drafting, and plenty of good lawyers now do, then verification is not an optional finishing step, it is the job. Pull every case the model gives you and read it in a real reporter or database before it touches a brief. Confirm the citation, the holding, and that the case exists at all. Keep a record of how you checked. None of this is exotic, and all of it is now the difference between a competent AI-assisted filing and a sanctions exhibit. The courts in two countries have told you which side of that line they will put unverified output on.

Browse the full AI Regulation News tracker

Informational analysis for working professionals, not legal advice. This briefing summarizes a 2025 foreign appellate ruling and compares it with US practice. Confirm how any development applies to your situation with qualified counsel in the relevant jurisdiction.