Lex Inside – ‘Can artificial intelligence be trained on unlicensed protected content?’ with Anne-Marie Pecoraro

Arnaud Dumourier invited Anne Marie Pecoraro, partner at UGGC Avocats, to BSMART 4Change to discuss the question: ” Can artificial intelligence be trained on protected content without a license?” She explained the application of the U.S. Copyright Act and Disney and NBC Universal’s recent complaint against #Midjourney. Once a work is protected, the Copyright Act prohibits its reproduction, distribution or adaptation, unless an authorization or license has been obtained. Reproduction must be characterized. Anne Marie Pecoraro clarified the “fair use” exception and the four criteria for its application, before addressing the US Copyright Office’s position on the reproduction process.

Shortly after this interview, two institutions unveiled the results of their studies on copyright and artificial intelligence.

At EU level, the European Parliament (the European Union’s highest legislative body) has released the final version of its study on how to reform copyright in order to adapt it to generative AI[1]. The institution recommends implementing a reform that gives priority to the consent of the rightful owner prior to training AI on protected works, and remunerating creators by taxing AI providers. This second suggestion is in line with the draft report by German MEP Axel Voss (EPP), published on July 4, 2025.

In France, the Senate Committee for Culture, Education, Communication and Sport published a report[2] on the subject. This standing Senate committee, set up by resolution on January 19, 1959, is chaired by Senator Laurent Lafon (Union Centriste). It issues information and legislative reports on communication, culture, education, the French-speaking world, youth, research, sport and community life. In this case, the commission’s mission was to analyze the links between artificial intelligence and artistic creation.

As the Senate’s rapporteurs point out, generative AI raises questions about the legal status of the works it creates, since it is no longer simply a tool that extends the will of a human creator; it sometimes becomes the source in its own right. Copyright law is deeply rooted in the notion of a work of the mind, and a fully generative AI necessarily comes up against the condition of originality inherent in copyright law.

AI is particularly disruptive for copyright holders. Indeed, the learning methods of these models involve massive data collection. The rapporteurs insist on the importance of remunerating the use of data within the framework of licenses, so that a balance can be struck between those involved in new technologies and copyright holders.

As far as legislation on this subject is concerned, the 2019 European Union Directive[3] on copyright and related rights has very quickly been overtaken by generative AI. A provision (transposed into French law)[4] of the directive opened a loophole for ” text and data mining ” or TDM, by authorizing the exploitation of protected works for the development of AI models. Developers were able to use this data, unless right holders objected via the opt-out system. The European Union Regulation of June 13, 2024[5] more recently introduced an obligation for AI model providers to take measures to respect copyright, notably through two provisions:

• The first requires compliance with the opt-out clause provided for in the 2019 directive for the benefit of rightsholders, and the establishment of a non-binding code of good practice. This code, drafted by the European Commission (the European Union’s second legislative body), was published on Thursday July 10 after a final review. Created to help AI providers comply with European AI legislation, it will come into force on August 2, 2025. Before that date, member states and the Commission will assess its suitability, and once approved, providers of general-purpose AI models who sign up to it will benefit from a lighter administrative burden and greater legal certainty. Rightsholders remain dissatisfied with the new wording, due to the lack of control over the data used by providers, and the prioritization of rights reservation methods in favor of “robot.txt”. The drafters have stayed within the framework of the AI regulation, but have taken a step towards rightsholders by replacing the notion of “reasonable efforts” by moving towards a commitment to apply these best practices for AI providers. By the end of July, the Commission will complete this code with guidelines on key concepts related to general-purpose AI models, including who falls in and out of the scope of AI legislation.
• The second concerns the issue of transparency regarding the data used by AI providers. It obliges AI providers to make available to the public a “sufficiently detailed summary ” of the content used to train their systems. This summary has yet to be drafted, and will have to reconcile the protection of rights holders with respect for suppliers’ business secrecy.

The Senate’s information report aims to bring about a French law that would establish a presumption of use of cultural content by AI providers, particularly if the consultation launched between AI providers and cultural rights holders fails to find a suitable solution by next autumn. The senators could table a bill which, after being studied by the relevant committees and potentially amended, would be put to the vote by the National Assembly and the Senate. Should this law fail, the option of taxing the sales made in France by AI providers has been raised, with the aim of compensating the cultural sector.

BIBLIOGRAPHY

• https://www.senat.fr/travaux-parlementaires/commissions/commission-de-la-culture-de-leducation-et-de-la-communication/intelligence-artificielle-et-creation.html
• https://www.senat.fr/travaux-parlementaires/textes-legislatifs/la-loi-en-clair/proposition-de-loi-relative-a-la-reforme-de-laudiovisuel-public-et-a-la-souverainete-audiovisuelle.html
• https://www.legifrance.gouv.fr/codes/article_lc/LEGIARTI000044363192
• https://eur-lex.europa.eu/legal-content/FR/TXT/PDF/?uri=CELEX:32019L0790
• https://eur-lex.europa.eu/legal-content/FR/TXT/PDF/?uri=OJ:L_202401689
• https://www.culture.gouv.fr/fr/nous-connaitre/organisation-du-ministere/Conseil-superieur-de-la-propriete-litteraire-et-artistique-CSPLA/travaux-et-publications-du-cspla/missions-du-cspla/le-cspla-publie-le-rapport-de-mission-relative-a-la-remuneration-des-contenus-culturels-utilises-par-les-systemes-d-intelligence-artificielle
• https://www.europarl.europa.eu/RegData/etudes/STUD/2025/774095/IUST_STU(2025)774095_EN.pdf
• https://www.contexte.com/fr/actualite/medias/le-parlement-publie-ses-recommandations-sur-lia-et-le-droit-dauteur_233790
• https://www.contexte.com/fr/actualite/medias/les-senateurs-menacent-de-passer-la-seconde-pour-obliger-les-fournisseurs-dia-a-payer-les-ayants-droit_233787
• https://www.contexte.com/fr/actualite/medias/le-cspla-publie-les-rapports-definitifs-sur-la-remuneration_228084
• https://www.legipresse.com/011-52749-droit-dauteur-quelle-protection-des-contenus-utilises-par-lia.html
• Rightsholders are already thinking about what to do next with the Commission’s “AI and copyright” code section | Context
• Lobbies prepare the “after” of the code of good practice on AI | Contexte
• https://www.bsmart.fr/video/29455-lex-inside-11-juillet-2025

---

[1] Generative AI and Copyright – European Parliament – July 2025

[2] Mission d’information relative à l’intelligence artificielle et la création – Published on July 9, 2025

[3] European Directive 2019/790 of April 17, 2019 on copyright and related rights

[4] Article L122-5-3 of the French Intellectual Property Code

[5] Regulation 2024/1689 of June 13, 2024 on artificial intelligence, articles 53, 1, c) and 53, 1, d)