Is There a Legal Limit to Training AI with Journalistic Content? What the Ruling Against Meta Reveals

The rapid advance of generative artificial intelligence has put technology companies and copyright holders on a collision course. On one side, platforms that need massive volumes of data to train their models. On the other, news outlets, content creators, and companies that invest their own resources in producing original material and do not necessarily agree to see it absorbed, without compensation or clear consent, by the algorithms of those very platforms.

This conflict gained a concrete chapter in early November 2026, when the 44th Civil Court of São Paulo granted a preliminary injunction ordering Meta to suspend a contractual clause that conditioned the continued use of news outlets' accounts on its platforms upon authorization to collect data and content, including copyright protected material, with use also intended for training artificial intelligence systems. It's worth stating upfront that this ruling did not originate from our firm, does not involve our clients, and, since it is a preliminary injunction, is not yet final nor does it represent a consolidated legal position on the matter. It is a provisional decision, subject to appeal and further review as the case proceeds, but it already works as a useful gauge for business owners paying attention to the topic.

Is content available online free to use for training AI?

The first confusion that tends to come up in this debate is equating public access with authorization to use. The fact that a piece of content is available online, visible to anyone who visits a website, does not mean it is free of copyright protection. The Brazilian Copyright Law (Lei de Direitos Autorais), Law 9.610 of 1998, grants the author or rights holder control over the reproduction, distribution, and transformation of their work, except in cases of limitations expressly provided for by law, such as quotation for purposes of criticism or use in a classroom setting.

Training an artificial intelligence model with journalistic texts, images, or other protected material technically involves acts of reproduction and, in many cases, transformation of the work into structured data that feeds the machine's learning process. This does not automatically fall under any of the existing legal exceptions, which leads much of the legal scholarship and emerging case law to treat this use as potentially subject to the rights holder's authorization, especially when there is clear economic exploitation behind the trained model.

The Meta case illustrates this point well. The controversy did not revolve around the existence of a website open to the public, but around the way the platform allegedly sought to obtain, through contract terms, broad authorization to reuse journalistic content published by outlets on its own social networks, including for purposes of developing artificial intelligence.

Does accepting terms of use amount to authorizing exploitation of the work?

This is the second central point of the discussion and perhaps the most sensitive one for companies operating on digital platforms. Terms of use and privacy policies tend to be accepted en masse, often without individual negotiation and without real bargaining power on the part of the user, whether that user is an individual or a company that depends on that platform to reach its audience. In the case at hand, the contested clause tied the continuation of the news outlet's account to acceptance of an authorization to use data and content that included AI training purposes, a sort of all or nothing deal: either the outlet accepted the condition, or it risked losing access to its own account and to the audience it had built over time.

The injunction granted by the São Paulo court recognized, on a provisional basis, that this type of conditioning deserves closer scrutiny, given the possible configuration of abuse of contractual position. This does not mean that every adhesion contract is void, nor that every authorization obtained through terms of use is invalid. It means that, when authorization to use protected content is obtained as a condition imposed for access to a service essential to the user's activity, without any real room for negotiation, the courts may find that there is an imbalance that justifies intervention, even on an urgent, provisional basis.

It's essential to stress that this ruling addresses a specific contractual clause, analyzed within a particular factual context, and does not represent a blanket ban on using journalistic content to train artificial intelligence under any circumstance. The matter is still taking shape in Brazilian courts, with no settled position yet from the Superior Court of Justice (STJ) or the Supreme Federal Court (STF), and rulings from higher courts, including on any eventual appeal of this very case, could well go a different way.

What Brazilian legislation and the international debate point to

Brazil still does not have a specific law governing the use of protected works to train artificial intelligence models. Bill 2338 of 2023, currently moving through Congress, proposes a regulatory framework for AI and touches on sensitive points such as transparency about data used in training and the rights of content holders, but it has not yet been enacted into law. In the meantime, what applies is the existing legal framework: the Copyright Law, the Civil Code, with respect to abuse of rights and contractual good faith, and the General Data Protection Law (LGPD), Law 13.709 of 2018, when the content involves personal data.

Internationally, the debate is heading in a similar direction. Several news outlets and publishers in other countries have already sought, through litigation or direct commercial agreements, ways to be compensated when their content is used to train language models. This movement suggests a trend toward regulatory maturity: instead of unrestricted use of data available online, the shift is toward models of licensing, compensation, or at least informed and negotiated consent, rather than consent imposed through a standardized clause.

How companies and creators can protect their assets against AI

For companies that produce original content, whether journalistic, advertising, technical, or editorial, and also for companies that develop or use artificial intelligence tools, some practical precautions are becoming increasingly relevant. The first is a careful reading of digital platforms' terms of use before accepting them, paying special attention to clauses dealing with content licensing, assignment of rights, or authorization for use in training automated systems. These clauses are often drafted in generic terms, buried within lengthy texts, precisely so they go unnoticed.

The second precaution is to formally document and register, whenever possible, ownership of the works produced by the company, whether through registration with the competent authorities or through clear internal contracts with employees and service providers regarding the intellectual property generated. The third is to assess, before entering into commercial partnerships with major platforms, whether there is room to negotiate clauses related to data and content use, rather than simply accepting standardized terms, especially when the company's activity structurally depends on that distribution channel.

Companies that develop or use artificial intelligence solutions, for their part, would do well to adopt a preventive stance regarding the origin of the data used in training, mapping out whether protected content is involved and whether there is a clear legal basis for that use. Given a regulatory landscape that is still incomplete, exposure to legal challenges tends to grow, and companies that already operate with transparency about their data sources tend to face lower reputational and legal risk.

The case involving Meta and Brazilian news outlets is a sign that the courts are willing to scrutinize closely how tech giants obtain authorization to exploit third party content, even though the final outcome will depend on higher courts and on a legislative maturing process that is still underway. For business owners and managers, the practical lesson is not to wait for a specific law to emerge, but to review their contracts, their content use policies, and their exposure to digital platforms now, with support from legal counsel specialized in intellectual property and digital law, so as to reduce risk in a regulatory environment that is still being built, one ruling at a time.