Folha v. OpenAI: copyright, news archives and AI training in Brazil

Generative AI turned archives into strategic infrastructure. Newsrooms, academic publishers, image banks and legal databases now ask whether their work became raw material for model training without permission. The dispute between Folha de S.Paulo, one of Brazil's most influential newspapers, and OpenAI places that global question inside the Brazilian legal system.

For a foreign reader, the important point is that the Brazilian debate is not a carbon copy of the New York Times litigation in the United States. Brazil has a civil-law tradition, strong constitutional language around authorship, and a copyright statute that does not use the same open-ended fair use doctrine familiar to US lawyers.

That does not mean every AI training activity is automatically unlawful. It means that the legal analysis must be built from Brazilian categories: authorization, economic exploitation, moral and patrimonial rights, unfair competition, market substitution and the proportionality of innovation against protected creative labor.

A publisher in Folha's position will usually present the archive as a protected body of journalistic production, not as an anonymous mass of public text. The claim is that articles, headlines, reports and editorial organization represent investment, selection, authorship and market value.

The legal theory tends to combine three lines of argument. First, the use of protected content for AI training may require authorization or remuneration. Second, an AI system can compete with the publisher by summarizing, reproducing or substituting access to the original content. Third, the extraction of content at scale may create an unfair advantage if one party builds a product with another party's investment.

The strongest version of the claim is not that every machine reading is forbidden. It is that industrial-scale ingestion of journalistic work, when used to create a commercial AI service, changes the legal and economic nature of the act.

An AI developer will typically answer that model training is transformative, statistical and non-expressive. In that view, the system does not store or republish the newspaper as a substitute archive; it learns language patterns, factual associations and probabilistic structures from a large corpus.

The developer may also argue that strict licensing for every source would make general-purpose AI impossible, locking innovation inside the hands of a few companies able to negotiate with every rights holder. This is the policy argument behind much of the global defense of large-scale training.

In Brazil, however, the decisive question is not whether the argument sounds plausible in Silicon Valley. It is whether Brazilian courts accept that type of transformative-use reasoning without an equivalent statutory fair use doctrine, and how they balance innovation with the economic rights of authors and publishers.

Brazilian copyright law protects both economic rights and certain personal links between the author and the work. These concepts do not map perfectly onto Anglo-American copyright. A foreign reader should avoid assuming that an American fair use analysis will decide the Brazilian outcome.

The Brazilian framework also interacts with constitutional values: freedom of expression, access to information, technological development, free enterprise and protection of intellectual creation. AI litigation will force courts to weigh all of those interests at once.

This is why the Folha dispute is strategically relevant. It may help define whether Brazil treats AI training mainly as innovation infrastructure, as copyright exploitation, as unfair competition, or as a hybrid problem requiring negotiated licensing models.

Publishers should map their archives, licensing history, technical blocking measures, API terms and evidence of substitution. The dispute will be stronger when the archive's market value and the AI system's commercial use are documented in concrete terms.

AI companies should maintain training-data governance, source documentation, opt-out policies, rights-management procedures and output monitoring. Even when the legal theory is uncertain, poor documentation makes the company look careless before regulators and courts.

Lawyers advising either side should not reduce the issue to a slogan such as copyright blocks innovation or AI steals content. The serious work is to design licensing, transparency and accountability models that preserve both technological development and the sustainability of professional knowledge production.

Folha v. OpenAI is part of a wider legal turning point. The question is no longer whether AI can process human knowledge at scale. It can. The question is who bears the cost of that process, who receives compensation, and how legal systems prevent innovation from becoming extraction.

For Brazil, the case may become a reference point for copyright, journalism and AI governance. For international readers, it is a reminder that the global AI market will not be governed only by US doctrines or European regulations. Civil-law jurisdictions will write part of the next chapter.