Perfil Sues OpenAI and Microsoft in Argentina — Where There Is No Fair Use Defence

Last updated: 27 August 2026. Figures, dates and quotes below are sourced to the primary publications linked inline, and each was checked against that source before publication. Where a claim comes only from one party to the litigation, it is labelled as such. Quotes from Spanish-language sources are given in the original with a translation.

The 60-second version

• Editorial Perfil, the Argentine publisher behind Perfil and Noticias, has filed suit against OpenAI, L.L.C., Microsoft Corporation and Microsoft de Argentina S.R.L. in Argentina’s federal civil and commercial courts.
• The claims are intellectual property infringement, unfair competition, abuse of a dominant market position, and daños y perjuicios — damages. Perfil alleges its journalism was taken through “procesos masivos de scraping, tokenización y vectorización” and used as “materia prima gratuita” — free raw material — for models now sold commercially, including material behind its paywall.
• Perfil says it is the first Spanish-language news publisher anywhere to bring such a case — the publisher’s own claim, echoed independently by the EFE wire, which we could not contradict.
• The suit follows a failed mediation. EFE reports that OpenAI and Microsoft took part in a pre-suit mediation and put arguments justifying their practices; it did not settle. Neither company has issued a public statement on the filing itself.
• Neither the damages figure nor the filing date has been disclosed.
• The important detail is the venue. Argentina has no fair use doctrine: Law 11.723 offers a short, closed list of exceptions, and the nearest analogue caps quotation at 1,000 words.
• Most coverage frames this as “another publisher sues OpenAI”. Read it instead as a test of whether the training-data defence US courts are still chewing over exists at all in civil-law jurisdictions.

Key dates and figures

What Perfil has actually alleged

The complaint, as described by Perfil’s own judicial desk, rests on infringement of intellectual property, unfair competition and daños y perjuicios; the Buenos Aires Times adds abuse of a dominant market position to that list. The factual allegation is that articles, investigations and exclusive interviews from Perfil’s newsroom were tracked, downloaded and stored at scale, then processed into training corpora, without licence or payment.

Two allegations are worth separating from the general noise. The first is that content behind Perfil’s paywall was reached. That is a materially different claim from scraping open pages: it implies either circumvention of an access control or ingestion of copies obtained downstream. Publishers who have spent the last two years tuning flexible sampling setups so that paywalled content stays indexable without giving the whole article away will recognise why this matters — the sampling window that keeps you in Google is also the window a crawler sees, and the gap between “what the indexer gets” and “what a subscriber gets” is exactly where this argument will be fought.

The second is the substitution claim. The Buenos Aires Times report names Bing, Copilot, Microsoft 365 Copilot, GitHub Copilot and Azure AI as products that draw on journalistic content to answer user queries directly, and argues this “reduces the need for users to visit news sites”. That is the unfair-competition leg, and it is the one that does not depend on winning the training-data argument at all.

A necessary disclosure: the Buenos Aires Times is published by Grupo Perfil. The fullest English-language account of this lawsuit is therefore the plaintiff’s own outlet reporting on the plaintiff. It is careful reporting, but it is not independent, and none of the coverage we read flagged it. Independent reporting does exist — an EFE wire story carried by swissinfo, ABC Color and others — but it is in Spanish, and it is the only place we found the detail that a mediation was tried first and failed.

Why the jurisdiction is the whole story

Every substantial analysis of AI training and copyright over the past three years has been written with American law in the background — fair use, four factors, transformativeness. Argentina does not have any of that.

A closed list, not a balancing test

Argentina’s copyright statute is Law No. 11.723 of 1933, as amended up to Law No. 27.588, published in the Boletín Oficial on 16 December 2020. It contains no fair use provision and no open-ended balancing test. What it has instead is a short set of specific carve-outs: an educational performance exception, an exception for people with disabilities that was broadened in December 2020 when Law 27.588 brought the Marrakesh Treaty into Argentine law, and — closest to the point here — Article 10, which permits publishing comments, critiques or notes on a work for didactic or scientific purposes, including up to 1,000 words of a literary or scientific work, and in every case “sólo las partes del texto indispensables a ese efecto” — only the parts of the text indispensable for that purpose.

A defendant cannot argue in Buenos Aires that mass ingestion for model training is transformative and therefore permitted, because the statute does not offer that shape of argument. It must instead find a listed exception that fits, or argue that what it did was not a reproduction in the legal sense at all. Neither is a comfortable position. This is the substantive reason to pay attention to a mid-sized Argentine publisher’s filing rather than filing it under “more of the same”.

Microsoft de Argentina is not decorative

Naming Microsoft de Argentina S.R.L. alongside the US parent is deliberate, and coverage has largely skipped past it. A local operating subsidiary gives an Argentine court a defendant with assets, staff and a registered address inside its jurisdiction. It shortens the service-of-process problem and makes any eventual order enforceable without a cross-border recognition fight. Whether that entity did anything relevant to training or inference is a separate question — but the strategic logic is clear, and publishers in similar markets should read it as a template.

What most coverage is getting wrong

The framing error is treating this as one more entry on a lawsuit tracker. Three things distinguish it.

The licensing asymmetry is the actual grievance. OpenAI has held a Spanish-language content deal with Prisa Media — publisher of El País, Cinco Días and As — since 13 March 2024, and signed its first Brazilian deals, with Grupo Folha and Grupo UOL, on 25 May 2026. So the position is not that Spanish- or Portuguese-language journalism has no value to these models; it is that journalism from Madrid and São Paulo was licensed and journalism from Buenos Aires apparently was not. That asymmetry, more than any doctrinal point, is what a court will find easy to understand.

The substitution claim survives even if the copying claim fails. Coverage keeps merging the two. They are separable. If the court finds that ingestion was lawful or unprovable but that direct answer-generation diverts readers who would otherwise have arrived at perfil.com, the unfair-competition leg still stands. We covered the underlying mechanism when we looked at how AI chatbots are reshaping news search and consumption; what has changed since is that the diversion is now being pleaded as a quantifiable commercial harm rather than described as an industry trend.

Argentine law may contain a defence nobody is discussing. Article 28 of Law 11.723 provides that “las noticias de interés general podrán ser utilizadas, transmitidas o retransmitidas” — news of general interest may be used, transmitted or retransmitted — requiring only that the source be stated when the original version is republished. Whether that reaches machine ingestion of full archives, as opposed to a rival paper reprinting a wire item, is genuinely unsettled, and we have seen no analysis testing it. If the defendants are well advised, it is where they will start. The statutory text is verified; the argument built on it is our inference, not a reported fact.

Practical takeaways if you publish, build or run software

• Extend your access-log retention now. The evidentiary core of every one of these cases is server-side records showing what fetched what and when. Policies that discard logs after 30 days quietly destroy the best evidence a publisher owns.
• Separate “blocked” from “provably blocked”. A disallow directive helps in litigation only if you can show it was live and correctly formed at the relevant time. Version-control your robots.txt and keep dated copies. Our guide to robots.txt rules that accidentally de-index news is written for the SEO failure mode, but the same discipline turns a policy into evidence.
• Audit the gap between your indexable sample and your paid article. If your paywall serves the full body to any agent claiming to be a search crawler, you have a bypass story waiting to be told about you.
• If you build on these APIs, read your indemnity clause rather than assuming it. Enterprise AI agreements vary widely on whether the vendor covers third-party IP claims arising from model output, and several cap that cover well below the exposure.
• Stop optimising solely for the click. Whatever the courts decide, answer engines are where a growing share of your audience meets your reporting. Being the cited source is now a distinct discipline from ranking — see how AI Overviews choose their sources and our ChatGPT Search visibility checklist for publishers. Litigation is a slow lever; citation is a fast one.

What we still don’t know

• The filing date. Perfil reported it on 24 August, ADEPA on 25 August. Neither states when the complaint was actually lodged, and we could not locate a docket entry.
• The amount claimed. No damages figure is public. Argentine practice often permits quantification later, so the absence is not necessarily meaningful.
• The evidence of paywall access. Perfil asserts protected content was reached. What it has — logs, model outputs reproducing subscriber-only text, something else — is not public.
• Whether an injunction is sought. None of the accounts we read mentions a medida cautelar. A prohibitory order would change the commercial stakes far more than damages.
• The defendants’ substantive position. Both companies argued their case in the failed pre-suit mediation, according to EFE, but the content of those arguments is not public and neither has issued a statement on the filing. As of 27 August 2026 this piece still largely reflects one side’s pleading.
• Whether the “first Spanish-language publisher” claim holds. It is Perfil’s own characterisation. We found no earlier example, which is not the same as there being none.
• How Argentine courts will treat foreign training conduct. If ingestion occurred on infrastructure outside Argentina, the applicable-law question arrives before any copyright analysis does.

FAQ

Is this the same as the New York Times case against OpenAI?

The allegations rhyme; the legal terrain does not. The NYT case turns substantially on US fair use. Perfil’s does not, because Argentina has no such doctrine. A ruling in Buenos Aires would tell you little about the American outcome, and vice versa.

Has any Latin American publisher already tried this?

Yes, and the outcome is instructive. Folha de S.Paulo sued OpenAI in Brazil on 20 August 2025 for unfair competition and copyright infringement, seeking an order that OpenAI stop collecting and using its content without authorisation or payment. Folha did not win that argument in court: in September 2025 the São Paulo judge refused a preliminary injunction on the ground that irreparable harm had not yet been demonstrated, and granted OpenAI’s request to seal the case. The dispute then ended commercially rather than judicially: OpenAI announced a content partnership with Grupo Folha and Grupo UOL on 25 May 2026, which LatAm Journalism Review reports ended the 2025 lawsuit. Financial terms were not disclosed. The regional precedent, in other words, is not a ruling on the merits — it is a settlement.

Is anything else happening in the region right now?

Yes, and it is closer to Perfil’s paywall argument than the OpenAI case ever got. On 24 August 2026 — the same week as Perfil’s filing — Folha sued Perplexity AI in São Paulo for unfair competition, copyright infringement and breaking its paywall, saying it had documented hundreds of thousands of improper accesses and had tried and failed to negotiate a licence first. The court gave Perplexity 72 hours to respond, and Folha asked for daily fines. Two Latin American publishers, in the same week, both pleading circumvention of paid access rather than training alone.

Does Perfil want money or a licence?

Publicly, damages and unfair-competition relief. But Perfil’s digital director Agustino Fontevecchia told EFE that what the company wants is “que nos retribuyan correctamente por nuestro trabajo” — to be properly paid for its work — while describing the AI companies as “verdaderos parásitos porque toman el contenido y lo exprimen”: true parasites, because they take the content and squeeze it. The regional precedent points the same way, since Folha’s suit against OpenAI ended in a licence rather than a judgment. Read Perfil’s framing as a demand to be paid rather than to be left alone.

Should smaller publishers be doing anything differently today?

Extend access-log retention, keep dated copies of your robots.txt, and check whether your paywall serves full text to anything presenting as a crawler. Those three steps cost little and are the difference between a provable claim and an assertion.

Sources

• Perfil — “Perfil demanda a OpenAI y Microsoft por el uso de contenidos periodísticos sin autorización” (24 August 2026; plaintiff’s own account)
• Buenos Aires Times — “Editorial Perfil sues OpenAI, Microsoft over unauthorised use of journalistic content” (24 August 2026; a Grupo Perfil title — the plaintiff’s own outlet)
• ADEPA — Asociación de Entidades Periodísticas Argentinas (25 August 2026)
• EFE via swissinfo — “Periódico argentino Perfil demanda a OpenAI y Microsoft por uso indebido de sus contenidos” (25 August 2026; independent wire report, source for the failed mediation and the Fontevecchia quotes)
• Infoleg — Argentina, Ley 11.723, official consolidated text as amended to Ley 27.588 (B.O. 16/12/2020)
• LatAm Journalism Review (Knight Center) — Folha de S.Paulo files lawsuit against OpenAI (filed 20 August 2025)
• Daniel Law — Folha de S.Paulo v. OpenAI, update on Brazil’s first generative AI copyright dispute (September 2025; injunction refused, case sealed)
• LatAm Journalism Review — OpenAI signs first Brazilian content licensing deals with Folha and UOL (26 May 2026; reports the deal ended the 2025 lawsuit)
• OpenAI — Grupo Folha and Grupo UOL strategic content partnership (25 May 2026)
• Fast Company Brasil — “Folha processa Perplexity por uso de reportagens e quebra de paywall” (25 August 2026)
• OpenAI — Global news partnerships: Le Monde and Prisa Media (13 March 2024)
• Press Gazette — running tracker of publisher AI deals and lawsuits

Please follow and like us: