By Nicolas Le Pays du Teilleul, LL.M. (London), Avocat (Paris), Member, CEIPI Research Laboratory (UR 4375)

     As innovation accelerates, the world is becoming more polarized. Generative artificial intelligence (“AI”) clearly reflects this duality. In an environment marked by instability and complexity, action can at times seem impossible. One path remains: understanding. That endeavor is not definitive; it is a first step toward a more informed form of individual freedom.

     Recent circumstances forced me to stop and observe for some time. I used that opportunity to immerse myself in the legal debates surrounding the European and French regulation of generative AI—the only form of AI at issue here. As I worked through the writings of scholars, policymakers, and institutions, one image gradually emerged: that of a roped climbing team. We move forward together, each according to his or her strength, yet each dependent on the others.

     The rope represents the conviction that law is a lifeline: a code of conduct that distinguishes those who seek freedom from those who yield to sheer power. At a time when law is being tested on the international stage, it is essential to remain within its shelter and to bring it back into view.

     The ice axe is the tool that calls into question certain settled assumptions. Its point allows us to uncover what time and repeated blows may have caused to harden. Cutting through the rhetoric surrounding AI quickly leads to copyright and related rights. Born of the upheavals of the revolutionary era and slowly solidified through case law, the sturdy legislative shell of copyright law—domestic, European, and treaty-based—remains a reliable guide through the labyrinth of the knowledge economy and the digital world.(1)

     Over the course of the past century, the relationship to creation has changed profoundly. Art is now expected to entertain immediately and at a low cost. The public has become a demanding, and at times insatiable, body of consumers, while creators themselves remain marked by a pervasive uncertainty. Bernard Stiegler’s work sheds light on that tension by identifying  the transformation of art as fungible commodity.(2) In this respect, figures remain the clearest way to show just how central art is to what are now commonly described as the “cultural and creative industries” (“CCIs”).(3) 

     Across the European Union, the cultural and creative industries can be valued at t several hundred billion euros. By contrast, the amounts earned through copyright and related rights remains far below the value generated through their use. According to Eurostat and the CISAC 2024 report, the EU’s cultural value added is approximately €200 billion, whereas collections by European collective management organizations amounted to just over €7.1 billion in 2023. The ratio between value created and value captured through rights is about one to thirty.

     In France, the analysis requires greater caution. The CISAC 2024 report records collections of €1.777 billion for France in 2023, making it the world’s second-largest market after the United States of America. If that figure is supplemented by a broader aggregation of the principal French collective management organizations,  the total may rise to approximately €2.5 to €3 billion. That second figure, however, is an aggregate estimate rather than a number directly consolidated by CISAC. The structural imbalance remains, but its intensity varies depending on the scope used.

     That distinction is crucial. Aggregate collection figures make it possible to document the gap between the sector’s overall economic value and the sums effectively captured through rights. Standing alone, however, such figures do not fully reflect the economic reality of authors, which still depends on distribution rules, administrative deductions, and differing institutional boundaries.

Territory	Indicator	Amount (order of magnitude)	Source
European Union	Cultural value added	≈ €199–202 bn	Eurostat
European Union	Collected copyright and related-rights revenue	≈ €7.16 bn	CISAC Global Collections Report 2024
EU – Ratio	Value capture	≈ 1 / 30	Calculated from the above data
France	Cultural and creative industries value added	≈ €43.1 bn	EY, Panorama des industries culturelles et créatives 2025
France	CISAC collections 2023	≈ €1.777 bn	CISAC Global Collections Report 2024
France	Aggregate CMO collections	≈ €2.5–3 bn	Aggregate estimate based on the principal CMOs
France – Ratio	Value capture	≈ 1 / 15 to 1 / 24	Depending on the perimeter used

     Disclaimer: The amounts presented above are totals derived from statistical and professional sources⁴ whose scopes are not fully identical. Nevertheless, they highlight a structural imbalance in the distribution of value.

     Artificial intelligence likewise relies on authors’ works to train and refine its models (the inputs). By contrast, synthetic content (the outputs) generated by artificial intelligence does not, at least for now, enjoy copyright protection in most legal systems absent identifiable human creative intervention.(5) 

     Thus, artificial intelligence acts as an accelerator of existing tensions. Its relationship to law, economics, politics, and philosophy is unstable, sometimes conflicting, and always revealing.

     AI does not create that imbalance; it reveals and accelerates it. The same is true for related rights, which fall outside the main scope of this Article, although some of them may also be affected by the development of AI. If AI-generated outputs cannot be classified as works, they likewise cannot support the legal characterization of a performance and therefore cannot give rise to performer status.

      This Article steps back to examine the forces at work. More specifically, it asks what artificial intelligence reveals about the legal and political order of contemporary society and, more particularly, about our relationship to law generally and copyright specifically. That is the purpose of Part I, which proceeds at the macro level—at the level of the power relations that structure the digital world.

     Next, we will need to return to the technical fundamentals of generative artificial intelligence. Without this examination, legal solutions specifically crafted for AI may miss their target. From this perspective, the general law of intellectual property remains an essential avenue of recourse for authors in potential disputes with AI. This will be discussed in Part II.

1. Artificial Intelligence as an Illustration of U.S.A./E.U. Antagonism

A. From Allies to Adversaries: The Slow Degradation of Transatlantic Legal Relations

     June 6, 2013.

     Sixty-nine years after D-Day, Edward Snowden revealed the extent of surveillance programs implemented by the United States, even against its allies. These revelations occurred at a time when the internet had already entered the Web 2.0 era, which is characterized by the importance of platforms and data.(6) 

     However, this fracture did not begin in 2013. As early as 2010, the European Commission began considering reforming the framework for protecting personal data,(7) a process that culminated in 2012 with a proposed regulation that laid the groundwork for the General Data Protection Regulation (“GDPR”).(8) 

     The European Parliament joined this movement by adopting, in 2014, a resolution on surveillance programs. That resolution marked a significant political shift and reflected a growing desire for independence.(9)

     The Court of Justice of the European Union, for its part, deepened this legal divide by invalidating the mechanisms governing data transfers to the United States in Schrems I (2015) and again in Schrems II (2020).(10)

      The GDPR was thus a turning point. The European Union established a regulatory instrument with no real equivalent. It is built on an expressly extraterritorial ambition—not to be confused with the American position—and is intended to project its standards beyond its borders. This has successfully created a “Brussels effect”, notably in the United States. 

      That momentum continued with the adoption of the Digital Services Act and the Digital Markets Act, (11) which signaled a shift in scale: the European Union is no longer merely protecting data; it is now seeking to regulate the actors and markets themselves.

B. The Intensification of Legal Antagonisms in the Age of AI

      The emergence of generative artificial intelligence is part of a broader reconfiguration of the digital world. This reconfiguration is marked by growing polarization among Western powers and the confirmation of China as a technological and economic rival to the United States.

      The latter dominates the most strategic segments of the AI value chain, including infrastructure, models, and applications. This dominance is based on exceptional investment capacity and deep vertical integration, backed by strong, coherent national political will.

     China has implemented a coordinated national industrial strategy combined with substantial educational efforts structured by the 2017 artificial intelligence development plan.(12)

     The European Union appears to lag behind technologically and industrially. Nevertheless, it possesses a central asset: a dense cultural and informational capital base that constitutes a significant share of the data used to train models. As the French Competition Authority has observed, these models depend on “large volumes of data … mainly derived from publicly accessible sources” to the extent that access to such data may constitute a “barrier to entry” in the market.(13) This gives the European Union indirect power grounded in control of access to these resources.

      The result is a balance of power among increasingly fragmented blocs (“guiding values are diverging in the ‘West’ ”)(25), where  law becomes an instrument of sovereignty.

      The adoption of the AI Act fits within that logic. Built on a risk-based approach, the Act imposes ex ante obligations on market actors and provides for ex post sanctions with a clearly extraterritorial reach.(14) In doing so, the Act extends the European strategy of compensating for a technological deficit by establishing the rules of the game.

      Despite this legislative intensification, European law lags behind the exponential speed at which generative AI systems are advancing. 

II. Artificial Intelligence Tested by Copyright: A Legal Framework Under Strain

A. A Legal Framework Predating the Rise of Generative AI

      Using protected works to train AI systems does not occur in a legal vacuum. It primarily falls within the framework established by the April 17, 2019 Directive on copyright and related rights in the digital single market.(15)

     Adopted when artificial intelligence technologies were substantially developed, but public-facing uses of generative AI were in their infancy, the Directive aimed chiefly at automated data analysis practices for scientific or informational purposes. It was not designed to regulate industrial uses based on the massive ingestion of protected works to train models capable of producing competing content.

      In this context, the text and data mining (“TDM”) exceptions were introduced in Articles 3 and 4. Subject to certain conditions, these exceptions allow for the reproduction and extraction of works and other protected material lawfully accessible to the user. This is permitted either for the benefit of research organizations and cultural heritage institutions acting for scientific research purposes, or where rightsholders have not exercised their right to opt out.(16)

      Notably, the French TDM regime is more restrictive. The exceptions in Article L. 122-5-3 of the Intellectual Property Code provides as follows: TDM is tied to “the implementation of an automated technique for analyzing texts and data in digital form in order to extract information therefrom, including patterns, trends, and correlations” and it is further provided that “copies and reproductions shall be destroyed once text-and-data mining has been completed”.(16)

     This mechanism is a key part of the balance established by the framework. In theory, it enables rightsholders to regain control over the use of their works by excluding their exploitation for text and data mining purposes.

     However, in the context of generative artificial intelligence, that balance has been deeply eroded. The public release of systems in 2022 revealed the extent to which content is collected and used online. The legal implications of these practices have emerged gradually, primarily because of the systems’ technical complexity and the opacity of their operation.

     In this context, the opt-out mechanism is largely ineffective. Designed for an environment in which uses were relatively limited, the mechanism assumes rightsholders can effectively express their refusal and that such refusal can be technically detected and respected. However, the dissemination of works, the absence of standardized reservation methods, large-scale web scraping, and the opacity of datasets make this requirement difficult to fulfill.(17)

     The adoption of Regulation (EU) 2024/1689 on June 13, 2024, on artificial intelligence (“AI Act”) did not replace that framework. On the contrary, it implicitly consolidated it. The Regulation does not establish an independent copyright regime but imposes duties of transparency and compliance with Union law on providers of general-purpose AI models, including in the field of intellectual property.(18)

     The AI Act assumes the framework inherited from the 2019 Directive and fully integrates the TDM mechanism into applicable artificial intelligence law. In practice, it ratifies a regime not designed for its current applications, as the European Parliament succinctly stated: “A simple way to remember the difference is: TDM finds patterns; GenAI synthesizes new expressions”.(17)

     This is all the more significant given that the copyright-related provisions were introduced late in the legislative process under mounting pressure from rightsholders and European institutions. This illustrates the difficulty of articulating technical, economic, and legal principles that are fundamentally heterogeneous within a single instrument.

B. The Necessary Detour Through Technical Reality

     Any legal analysis of artificial intelligence systems requires at least a minimal understanding of how those systems work technically.

     As the French Senate’s informational report on artificial intelligence and creation explains, these systems rely on a technical chain consisting of several stages: data collection, preparation, training, and deployment.(19)

      The process begins with the large-scale collection of data, often carried out through automated web-scraping techniques. Drawn from online content, these data constitute the models’ “fuel” or, as Alexandra Bensamoun puts it, the “ingredients of the recipe.”(19)(22) Their accessibility, however, says nothing about whether they may be used freely; they may be protected by copyright, related rights, or other legal regimes.

     The collected data are then processed through several technical steps, including tokenization, which breaks content into elementary units and renders it machine-processable. More broadly, the training process involves the copying and transformation of works, allowing the model to capture certain structural, stylistic, and statistical regularities, even though their perceptible form is not retained as such. Model training thus relies on identifying regularities and probabilistic relationships within the data. That technical transformation, however, does not mean that no human ever perceives the works in their original or partially original form. At other stages of the pipeline—including data selection, curation, annotation, quality control, and fine-tuning—human operators may still view works or portions of works as perceptible content.

     This chain of human responsibility in the design and operation of such systems may support a finding of copyright infringement, as it will be developed hereinafter.

      Ultimately, artificial intelligence merely recombines data in response to a prompt. AI does not create in the etymological sense; rather, it operates based on preexisting works, which it transforms and recombines.(20) Thus,  AI-generated content does not appear to be eligible for copyright protection under the dominant state of law absent identifiable human creative intervention. Therefore, it is appropriate to call such material “synthetic content”.(21)

C. The Late but Welcome Reassertion of Copyright

     Over time, copyright advocates have reasserted the relevance of copyright law in response to generative artificial intelligence through expert reports, parliamentary hearings, and normative initiatives.

     In France, the work of the Conseil supérieur de la propriété littéraire et artistique (“CSPLA”) and the hearings conducted before Parliament established a convergent diagnosis of the following issues: the mass harvesting of protected content; opacity surrounding training data; the practical ineffectiveness of the opt-out mechanism; and structural evidentiary asymmetry.(22) Those findings were reiterated in the Senate’s information report on artificial intelligence and creation.(19)

     The Ministry of Culture launched an attempt at negotiated regulation, but it failed due to structural divergences among the involved parties and the absence of effective legal constraints.

     Then, a legislative response was contemplated in the form of a bill designed to regulate the use of works by artificial intelligence systems23, notably by creating presumptions of use of protected works. However, when asked for its opinion, the Conseil d’État substantially limited the scope of these mechanisms, emphasizing the constraints imposed by European Union law. (24)

     At the European level, the issue was incorporated into the debates surrounding the AI Act only belatedly before becoming the subject of expert studies. These studies highlighted a structural “legal mismatch” between the regime established by Directive 2019/790 and the uses associated with generative artificial intelligence.(17)

     The European parliamentary process led by Axel Voss culminated in the delivery of a report.(25) Subsequently, on March 10, 2026, a resolution was adopted26  which recognizes the fundamental nature of intellectual property rights. It also proposes the use of statutory presumptions, among other things. 

     These initiatives reflect a convergence toward a common objective: restoring copyright’s effectiveness in the face of systems characterized by opacity and informational asymmetry. Their implementation, however, remains uncertain.

D. In the Face of the Fragility of AI-Specific Norms, the Persistence of Copyright Fundamentals

     At first glance, the mechanisms contemplated in France and at the European Union level—chief among them presumptions concerning the use of protected works—display a meaningful degree of legal coherence.

   In France, however, the Conseil d’État has significantly narrowed their scope, confining them to civil proceedings and economic rights.(24) The criminal arm of infringement law is therefore excluded—an unsurprising result, given the centrality of the presumption of innocence and the allocation of the burden of proof in criminal matters. The corresponding exclusion of moral rights is far more debatable, given the central role those rights play in the structure of French copyright law.

     Moreover, the effectiveness of these mechanisms runs into major technical obstacles. Identifying the works used in training remains extraordinarily difficult because of the scale of data collection, the transformation of those data into mathematical representations, and the opacity of the models themselves.

     However attractive—and, in some respects, necessary—these instruments may be, they still require further refinement before they can become fully operational.

     In the meantime, and in any event, the general law of copyright remains fully available to rights holders in disputes involving AI systems, whether the allegedly unlawful exploitation occurs at the input or output stage, provided, of course, that the TDM regimes are found inapplicable.

     Economic rights may be enforced through identifiable human acts occurring along the value chain, including collection, selection, annotation, training, generation, and exploitation of content. Depending on the circumstances, those acts may implicate the rights of reproduction and communication to the public, subject, of course, to questions of territoriality.

     By contrast, moral rights may be implicated even absent any directly identifiable human intervention. Because these systems operate through the absorption, transformation, and recomposition of works, the operation of the system itself may affect the author’s right of integrity—in both its material and spiritual dimensions—and, in some cases, the right of attribution. The harm arises less from any single discrete act than from the system’s architecture. Such harms may become legally cognizable as soon as the system, or its outputs, is made accessible within the territory of the Union.

     Technical and evidentiary difficulties, however, do not require abandoning copyright. Rather, they call for a shift in its points of attachment: economic rights apply where human intervention can be identified, whereas moral rights remain an inherent limit on the operation of the systems themselves.

      In this respect, the AI Act provides only an uncertain basis for any attempt to reclassify copyright-infringing generative AI systems as high-risk systems. To be sure, Recital 48 makes clear that infringements of fundamental rights, including intellectual property rights, fall within the broader logic of the regime. Chapter III, however, remains organized around a closed classification tied to the uses identified in Article 6 and Annex III. Copyright noncompliance therefore does not appear, per se, sufficient to bring a generative AI system within the high-risk regime unless its concrete use already falls within one of the relevant categories—for example, in education or vocational training.²⁷ Member States remain involved in the implementation of sanctions which are not necessarily exclusively pecuniary. One might discern in that arrangement a limited opening for authors. That hope, however, should be tempered. In an environment dominated by actors with considerable financial power, the balance of power remains overwhelmingly economic. The bees of Staten Island and the disciples of Shaolin put it more simply: “Cash Rules Everything Around Me”.

Conclusion

     The regulation of artificial intelligence must be considered  within a broader context of global polarization—geopolitical, economic, and social. That polarization has been accompanied by widening inequality, increasing concentration of resources, and growing fragmentation of regulatory and economic models. As a core technological infrastructure of the twenty-first century, artificial intelligence both reflects and accelerates those dynamics.

     Artificial intelligence is neither inherently virtuous nor inherently harmful. It is a tool. Left insufficiently constrained, however, it can reinforce existing imbalances through the concentration of value, the appropriation of content, and the marginalization of creators. Properly understood and regulated, by contrast, it may become a tool of rebalancing, enabling a fairer distribution of value and a renewed affirmation of the place of creative activity.

      It is in this light that current initiatives, at both the national and European levels, should be understood. They reflect an effort to regain control over a technological environment structured by global actors, using law as an instrument of sovereignty.

     That effort, however, cannot be limited to a simple set of standards, One of the principal lessons of this research is that law, if it is to remain effective, cannot disregard the technical reality of the systems it seeks to govern. The principles presently under consideration—transparency, traceability, and evidentiary presumptions—display a substantial degree of legal understanding. Their effectiveness, however, will depend on whether they can be implemented in environments defined by opacity and complexity.

    In that respect, the development of complementary tools appears essential: registries of training data, watermarking mechanisms, and the use of trusted third parties. Many of these instruments remain to be developed, and they will determine whether the law can produce concrete effects.

     At the same time, artificial intelligence invites a reaffirmation of the foundations of copyright law. Far from weakening copyright, it underscores its continuing relevance. The machine does not operate ex nihilo; it learns, transforms, and recombines. That reality confirms that authorship remains inseparable from human creative intervention and reinforces an author-centered conception of copyright. From this perspective, it may be appropriate for both European Union and national legislation to recognize more explicitly the connection between the originality of a work and the creative intervention of a human person, and perhaps even expressly to exclude an AI “operator,” within the meaning of Article 3 of the AI Act, from the scope of the concept of a collective work.

     In this context, European initiatives—particularly those advanced by Parliament—point toward the possibility of further harmonization. Such harmonization is desirable, provided it is not adopted with haste. Recent experience in the digital field shows that regulations adopted with urgency often struggle to achieve their intended objectives.(28) Effective harmonization requires a precise understanding of the technical mechanisms at issue and the economic equilibria they shape.

     Beyond law, the response must also be cultural and educational. Artificial intelligence should not result in the eclipse of the human person, but rather in a redefinition of the human role. That requires a sustained educational effort, enabling individuals to understand these tools, make use of them, and retain a meaningful degree of freedom in relation to them. In that framework, culture must be reaffirmed as a strategic concern at both the national and European levels.

     In this evolving landscape, the role of lawyers—and of attorneys in particular—is bound to change. The litigation to come will be marked by technical uncertainty and informational asymmetry. It will require collective, structured, and genuinely deterrent enforcement mechanisms capable of shifting the balance of power in favor of copyright and related-rights holders.

     Artificial intelligence does not mark the end of copyright. It marks a new stage in its development. Like any major technological transformation, it places existing categories under strain, but it also offers an opportunity to reinforce them. The French copyright tradition, the product of a gradual construction over more than a century, has already demonstrated its capacity for adaptation. It must now confront a challenge of unprecedented scale without relinquishing its fundamental principles. So long as a sustained dialogue is maintained among technology, law, and politics, this evolution may constitute not a threat, but an opportunity: an opportunity to reaffirm the central place of the human person at the heart of both creation and law.

     By way of conclusion, the emergence of quantum computing may represent a further challenge by calling into question the very foundations of digital security.  Rendering vulnerable certain methods of encryption that are now widely used raises significant issues of sovereignty and serves as a reminder that digital regulation operates within a technological environment in constant transformation.(29) Artificial intelligence places copyright under pressure. Quantum computing may, in time, place trust itself under pressure.

Footnotes

1. See, in particular, on the history of French copyright since the Act of 19 July 1793: A. Lucas, H.-J. Lucas and A. Lucas-Schloetter, Traité de la propriété littéraire et artistique, LexisNexis, 5th ed., p. 3 ff.; Fasc. 1110: Histoire du droit d’auteur, JurisClasseur Propriété littéraire et artistique, first publication: 1 Sept. 2010, Laurent Pfister.

2. B. Stiegler, La société automatique. 1. L’avenir du travail, Paris, Fayard, 2015, esp. pp. 98–103 and p. 160. See also:

   ;

   ;

   ;

   . 

3. Ministry for Europe and Foreign Affairs, “Supporting cultural and creative industries”: https://www.diplomatie.gouv.fr/fr/le-ministere-en-action/assurer-la-presence-de-la-culture-francaise/soutenir-les-industries-culturelles-et-creatives

4. Eurostat, Culture statistics – cultural enterprises: https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Culture_statistics_-_cultural_enterprises ; CISAC, Global Collections Report 2024: https://www.cisac.org/Newsroom/global-collections/global-collections-report-2024 ; EY, Panorama des industries culturelles et créatives 2025: https://www.ey.com/fr_fr/insights/tmt/panorama-icc-2025

5. A. Bensamoun and G. Loiseau (eds.), Droit de l’intelligence artificielle, 3d ed., LGDJ, para. 527 ff.

6. European Parliament resolution of 12 March 2014 on the US NSA surveillance programme, surveillance bodies in various Member States and their impact on EU citizens’ fundamental rights and on transatlantic cooperation in Justice and Home Affairs (2013/2188(INI)): https://www.europarl.europa.eu/doceo/document/TA-7-2014-0230_EN.html

7. EU Commission, A comprehensive approach on personal data protection in the European Union, COM(2010) 609 final, 4 Nov. 2010: https://eur-lex.europa.eu/LexUriServ/LexUriServ.do?uri=COM:2010:0609:FIN:EN:PDF

8. European Commission, Proposal for a Regulation of the European Parliament and of the Council on the protection of individuals with regard to the processing of personal data and on the free movement of such data (General Data Protection Regulation), COM(2012) 11 final, 25 Jan. 2012: https://eur-lex.europa.eu/legal-content/EN/TXT/PDF/?uri=CELEX:52012PC0011

9. CJEU, 6 Oct. 2015, Maximillian Schrems v Data Protection Commissioner (“Schrems I”), Case C-362/14: https://eur-lex.europa.eu/legal-content/FR/ALL/?uri=CELEX:62014CJ0362 ; CJEU, 16 July 2020, Data Protection Commissioner v Facebook Ireland Ltd and Maximillian Schrems (“Schrems II”), Case C-311/18: https://eur-lex.europa.eu/legal-content/FR/ALL/?uri=CELEX:62018CJ0311

10. Regulation (EU) 2016/679 of the European Parliament and of the Council of 27 Apr. 2016 (GDPR): https://eur-lex.europa.eu/legal-content/FR/ALL/?uri=celex%3A32016R0679

11. Regulation (EU) 2022/1925 of the European Parliament and of the Council of 14 Sept. 2022 on contestable and fair markets in the digital sector (DMA): https://eur-lex.europa.eu/legal-content/FR/ALL/?uri=CELEX:32022R1925 ; Regulation (EU) 2022/2065 of the European Parliament and of the Council of 19 Oct. 2022 on a Single Market for Digital Services (DSA): https://eur-lex.europa.eu/legal-content/FR/TXT/?uri=CELEX%3A32022R2065

12. State Council of the People’s Republic of China, New Generation Artificial Intelligence Development Plan, 20 July 2017: https://digichina.stanford.edu/work/full-translation-chinas-new-generation-artificial-intelligence-development-plan-2017/

13. French Competition Authority, Opinion No. 24-A-05 of 28 June 2024 on the competitive functioning of the generative artificial intelligence sector, esp. p. 4: https://www.autoritedelaconcurrence.fr/sites/default/files/integral_texts/2024-07/24a05_merged.pdf

14. Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 (AI Act): https://eur-lex.europa.eu/eli/reg/2024/1689/oj?locale=fr

15. Directive (EU) 2019/790 of the European Parliament and of the Council of 17 Apr. 2019 on copyright and related rights in the Digital Single Market: https://eur-lex.europa.eu/eli/dir/2019/790/oj?locale=fr

16. See, in particular, Directive (EU) 2019/790, recitals 8, 18 ff., Arts. 3 and 4; French Intellectual Property Code, Art. L. 122-5-3: https://www.legifrance.gouv.fr/codes/article_lc/LEGIARTI000044363192

17. European Parliament, Generative AI and Copyright, PE 774.095, July 2025: https://www.europarl.europa.eu/RegData/etudes/STUD/2025/774095/IUST_STU(2025)774095_EN.pdf

18. Regulation (EU) 2024/1689, Art. 53(1)(c) and (d): https://eur-lex.europa.eu/legal-content/FR/TXT/HTML/?uri=OJ%3AL_202401689

19. Senate, Information Report No. 842, submitted on behalf of the Committee on Culture, Education, Communication and Sport by the fact-finding mission on artificial intelligence and creation, by Agnès Evren, Laure Darcos, and Pierre Ouzoulias, registered at the Presidency of the Senate on 9 July 2025, esp. pp. 22–28, 42–49, 60–61, and 77–90: https://www.senat.fr/rap/r24-842/r24-8421.pdf

20. Dictionnaire de l’Académie française, 9th ed., entry “créer”; CJEU, 11 June 2020, SI and Brompton Bicycle Ltd v Chedech / Get2Get, Case C-833/18: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=celex:62018CJ0833

21. A. Bensamoun and G. Loiseau (eds.), Droit de l’intelligence artificielle, 3d ed., LGDJ, para. 527 ff.

22. CSPLA, Rapport de mission présenté à la réunion plénière du CSPLA du 9 décembre 2024: https://www.culture.gouv.fr/Media/medias-creation-rapide/cspla_rapport_ia_template_dec_.2024.pdf ; see also:

    ;

    ;

    ;

23. Bill No. 220 establishing a presumption of exploitation of cultural content by artificial-intelligence providers, introduced by Senators Laure Darcos, Agnès Evren, Pierre Ouzoulias, Laurent Lafon, Catherine Morin-Desailly, and Karine Daniel: https://www.senat.fr/dossier-legislatif/ppl25-220.html

24. Conseil d’État, Opinion on a bill establishing a presumption of exploitation of cultural content by artificial-intelligence providers, 19 March 2026, No. 410652: https://www.conseil-etat.fr/content/download/237335/document/410652.pdf

25. European Parliament, A. Voss, Report on copyright and generative artificial intelligence – opportunities and challenges, 25 Feb. 2026, doc. A10-0019/2026, 2025/2058(INI): https://www.europarl.europa.eu/doceo/document/A-10-2026-0019_EN.html

26. European Parliament, resolution of 10 March 2026 on copyright and generative artificial intelligence – opportunities and challenges, P10_TA(2026)0066, A10-0019/2026: https://www.europarl.europa.eu/doceo/document/TA-10-2026-0066_EN.html

27. Regulation (EU) 2024/1689, recital 48, Art. 6, and Annex III: https://eur-lex.europa.eu/eli/reg/2024/1689/oj?locale=fr

28. European Commission, Communication from the Commission to the European Parliament and the Council – Data protection as a pillar of citizens’ empowerment and the EU’s approach to the digital transition – two years of application of the General Data Protection Regulation, COM(2020) 264 final, 24 June 2020: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=celex:52020DC0264 ; European Commission, Second Report on the application of the General Data Protection Regulation (GDPR), 25 July 2024: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=celex:52024DC0357

29. European Commission, Quantum Technologies Flagship; NIST, Post-Quantum Cryptography: https://digital-strategy.ec.europa.eu/en/library/recommendation-coordinated-implementation-roadmap-transition-post-quantum-cryptography

AI and Copyright: Symptoms of a Polarized World

By Nicolas Le Pays du Teilleul, LL.M. (London), Avocat (Paris), Member, CEIPI Research Laboratory (UR 4375)

Thanks to Elise Mariage, student at CEIPI, for her patience; to Elodie Migliore, doctoral researcher at CEIPI, for her time and insightful guidance; to Yann Basire for his trust and careful reading, and to Kevin Tottis, with Tottis Law, for his expertise in US law.