Training an AI with protected works: is it legal? Who holds rights over what it generates? Legal keys on monetization and authorial recognition

23 May, 2025
–
Intellectual property
–
Lucas Charnet


In the history of human creativity, few revolutions have been as fast and unsettling as the emergence of generative artificial intelligence. Just as the printing press, photography, or cinema once marked a before and after, today AI forces us to rethink the very foundations of authorship, originality, and, of course, intellectual property.

Authors, artists, publishers, and record labels are all asking the same question: Can a tech company train an artificial intelligence model using works protected by copyright? And is the output itself protectable under copyright? Moreover, if the result of that training is monetized, shouldn’t there be financial compensation—and moral recognition—for the creators whose works were used as raw material?

This article explores the most relevant legal aspects of this issue from the perspective of European and Spanish law and proposes steps toward a fair and sustainable regulation in a context where technology is evolving faster than the law.

Is it legal to train an AI with protected works?

The first legal battle takes place during the training phase. For an AI model to be capable of generating coherent images, texts, or music, it must analyze millions of examples. Many of those examples are works protected by copyright. Is that legal? As we’ve discussed in various posts on this blog and in our podcast “Authors with Rights,” authors and rightsholders have the exclusive right to authorize reproduction, transformation, public communication, and distribution of their works. Therefore, the key question is: when an AI is trained, is it “reproducing” a work the way a radio station or advertisement would—and thus using that right without the artist’s permission?

Here we find two opposing positions:

Big tech companies argue that training an AI is akin to human learning: we read, listen, and then create something new. They claim there is no reproduction or literal copying, but mere “inspiration.” They argue that the AI model doesn’t store or reuse literal fragments, but rather extracts patterns, structures, and styles from the analyzed data. In this logic, the training process would not amount to direct commercial exploitation of the original work.

Authors and collecting societies, on the other hand, argue that there is reproduction—even if temporary or partial—and therefore authorization is required. Any reproduction, even ephemeral, is subject to the exclusive rights of the holder.

In legal terms, the issue is whether training infringes the right of reproduction (Art. 18 LPI), and consequently whether a prior license is needed or if it might be covered under certain legal limitations to copyright.

What does the law say? Europe vs. the United States

In the United States, the doctrine of fair use applies. This general clause allows certain unauthorized uses of protected works if they are transformative, serve a legitimate purpose (education, criticism, parody…), and do not unjustifiably harm the market of the original work. This has given tech companies some leeway to defend training as fair use—an expansive interpretation of the principle.

Europe, in contrast, operates under a system of enumerated limitations. The most relevant here is text and data mining (Art. 4 of Directive 2019/790). This allows the use of works to train AI provided that:

1. Access to the work was lawful.
2. The rights holder has not explicitly objected.

And here lies the crux: authors can exclude their works from this use via an express reservation—for example, through robots.txt files or legal disclaimers on their websites. But in practice, this “opt-out” is difficult to enforce and its effectiveness is limited. The problem is evident: how can an author know if their work has been used without permission?

And if the result is a new work? Is it protected?

Let’s assume the use of protected works for training is permitted—or that tech companies pay a fee to authors for using their works to train AI. The next issue is what happens with the works that the AI generates: a song, an image, a text, a graphic design… Do they enjoy rights as if created by a human? Can they be commercialized?

From a legal standpoint, as we have discussed repeatedly on this blog, a work must result from human creative activity to be eligible for copyright protection in Europe. This is established by Article 5 of the Intellectual Property Law. An AI cannot be an author. Nor a rightsholder. Therefore, under continental European copyright doctrine, a creation made by AI cannot be considered a legal “work.”

However, this does not mean that everything generated by AI lacks protection. If there is significant human input—as when an artist uses AI as a tool—the resulting work could be protected. It would be no different from using a paintbrush, a camera, or a synthesizer.

The problem arises when there is no discernible human contribution. In those cases, the work would not be eligible for protection… but in practice, these works are already being monetized in the market. And that’s where a new conflict begins.

Monetization without compensation: the new injustice against creators

Today, major tech companies are generating millions using systems trained with protected works. Music, illustrations, voices, scripts, narrative styles… all of it forms the data digested by algorithms.

But while the economic profits are concentrated in a few companies, the authors whose works were used as source material receive no compensation or recognition. This imbalance is not only unfair—it also threatens the sustainability of professional artistic creation.

As happened with the music industry during the era of piracy, without a fair compensation system, professional creativity may become unviable.

Legal solutions: from remunerated exceptions to extended collective licenses

Faced with this scenario, lawyers, legislators, and rights organizations are proposing legal solutions to regulate this mass use of protected works:

1. Remunerated exceptions: amending the law so that the use of works for AI training purposes is subject to a fee, similar to private copying levies. This would allow continued use of the works, but with compensation for rightsholders.
2. Extended collective licensing: a collecting society grants a license on behalf of all rightsholders, including those who are not members, unless they explicitly opt out. This solution has been proposed in Spain and already exists in countries like Sweden or Denmark.
3. Transparency obligations: the EU Artificial Intelligence Regulation mandates the publication of summaries of the datasets used to train models. But we are still far from real traceability that would allow authors to know if their works have been used.

Moral recognition: it’s not all about money

Beyond the economic question, one right cannot be overlooked: the moral right of attribution. This right, inalienable and non-renounceable, requires acknowledgment of the author as the creator of their work.

If an AI generates a work based on the style, voice, or material of a real author—even if it does not literally infringe the original work—it erodes their artistic identity. Recognizing that contribution, even if it does not entail exploitation rights, is a matter of cultural justice.

One step further: synthetic works and mandatory transparency

The European Commission is already working on proposals to require that any AI-generated work be properly labeled—with metadata or watermarks—as non-human content. This measure, part of the AI Regulation, does not solve all the problems, but it helps avoid confusion and protects consumers.

The goal: to ensure that synthetic creations are not sold as human works and that the public—and markets—can tell the difference.

Conclusion: toward a new creative pact

AI is here to stay, and its use in the creative sector will continue to grow. But if we want professional artistic creation to survive—and not be reduced to a “romantic hobby”—we need new rules to rebalance the scales.

It’s about building a pact between technology, rights, and culture. A pact that recognizes the importance of creators in the digital ecosystem, ensures fair compensation, and preserves the integrity of their works from unauthorized use.

Because if history has taught us anything, it’s that every technological revolution requires new rules. And this will be no exception.

At MES Advocats, we specialize in intellectual property and the legal impact of artificial intelligence in the creative industries. We offer comprehensive advice and tailored solutions to protect the rights of authors, publishers, and companies in this new landscape. If you want to know whether your work may have been used without your consent, or how to regulate AI use in your contracts, contact us through this link. We’re here to help.

###


Derecho tecnológicoCopyrightInteligencia artificial generativaIntellectual PropertyRegulación europea IA

---