Notes

• [1]
  While the EU AI Act (Regulation [EU] 2024/1689) contains a definition of a General-Purpose AI (GPAI) model (Art. 3(63) AI Act) that includes GenAI models (not legally defined though), the now repealed U.S. Executive Order under President Biden contained the following definition of GenAI: ‘The term “generative AI” means the class of AI models that emulate the structure and characteristics of input data in order to generate derived synthetic content. This can include images, videos, audio, text, and other digital content.’ (Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence, October 30, 2023, Sec. 3(p)). The term is further explained by some authors as ‘a catch-all name for a massive ecosystem of loosely related technologies, including conversational text chatbots like ChatGPT, image generators like Midjourney and DALL·E, coding assistants like GitHub Copilot, and systems that compose music, create videos, and suggest molecules for new medical drugs. Generative-AI models have different technical architectures and are trained on different kinds and sources of data using different algorithms. Some take months and cost millions of dollars to train; others can be spun up in a weekend. These models are also made accessible to users in very different ways. Some are offered through paid online services; others are distributed open-source, such that anyone could download and modify them.’ (K. Lee, A. Feder Cooper, and J. Grimmelmann, ‘Talkin’ Bout AI Generation : Copyright and the Generative-AI Supply Chain’ (July 27, 2023). (https://ssrn. com/abstract=4523551).
• [2]
  G. Marcus, Generative AI’s Continuing Copyright Problems, an Essay in Memory of Suchir Balaji, 1998 – 2024’, December 14, 2024.
• [3]
  The U.S. knows the safety valve of ‘fair use’ (in Article 107 of the U.S. Copyright Act), not the EU, the UK, Australia and many other countries.
• [4]
  O. Demirci, J. Hannane and X. Zhu, ‘Research : How Gen AI Is Already Impacting the Labor Market, Harvard Business Review’, November 11, 2024 (at https://hbr. org/2024/11/research-how-gen-ai-is-already-impacting-the-labor-market).
• [5]
  Ibid.
• [6]
  According to Nicole Hendrix, co-founder of the Concept Art Assn (which commissioned the study of CVLEconomics, ‘Future Unscripted : The Impact of Generative Artificial Intelligence on Entertainment Industry Jobs’, January 2024) quoted by Christi Carras, ‘Which entertainment jobs are most likely to be disrupted by AI? New study has answers’, Los Angeles Times, January 30, 2024 (at https://www.latimes.com/entertainment-arts/business/story/2024-01-30/ai-artificial-intelligence-impact-report-entertainment-industry).
• [7]
  According to the Harvard Business Review above-mentioned study, ‘automation-prone jobs like writing and coding saw significant declines in demand while new types of work requiring AI-related skills began emerging.’
• [8]
  Relying on a study from Goldman Sachs (dated from 2023), a review by the World Economic Forum ‘suggests that generative AI has the potential to automate 26% of work tasks in the arts, design, entertainment, media and sports sectors’ (‘How might generative AI change creative jobs ?’ May 9, 2023, at https://www.weforum.org/stories/2023/05/generative-ai-creative-jobs/ relying on Jan Hazius, Joseph Briggs, Devesh Kodnani and Giovanni Pierdomenico, The Potentially Large Effects of Artificial Intelligence on Economic Growth, March 2023 at https://www.key4biz.it/wp-content/uploads/2023/03/Global-Economics-Analyst_-The-Potentially-Large-Effects-of-Artificial-Intelligence-on-Economic-Growth-Briggs_Kodnani.pdf). Another study for the U.S. stresses that for the film and TV industry, the jobs most likely impacted are ‘3-D modelling, character and environment design, voice generation and cloning and compositing, followed by sound design, tools programming, script writing, animation and rigging, concept art/visual development and light/texture generation’, while for the music and sound-recording industry, ‘the tasks most likely to be affected by AI are voice generation and cloning, music generation and recording and lyrics composition, followed by mastering, mixing and tools programming’, and for the gaming industry, ‘3-D modelling and concept art/visual developing are the tasks most vulnerable to AI, followed by character and environment design, sound design, tools programming and voice generation and cloning’ (see the study of CVLEconomics, op. cit., quoted by Christi Carras, Which entertainment jobs are most likely to be disrupted by AI?, op. cit.). In general, based on conversations with creative workers and feedback from observers of the creative economy, among the jobs which are and will be primarily affected, one can single out voice actors (the cloning of voice has made incredible progress), graphic designers, photographers, journalists (several press publishers have started to use automatic tools for certain categories of news such as sports) and translators (beyond the DeepL effect for translations of common, utilitarian texts, the post edition of automatic translations for literary works is less remunerated).
• [9]
  E. Newton-Rex, ‘The UK’s AI & copyright proposals would irreparably harm the country’s creators’, post on LinkedIn, December 17, 2024 (https://www.linkedin.com/pulse/uks-ai-copyright-proposals-would-irreparably-harm-ednewton-rex-w0vhc/).
• [10]
  This might exacerbate already the decline of earnings by self-employed authors and artists (as showed by a recent survey for the U.K. provided by Martin Kretschmer, CREATe research group, University of Glasgow, 2025):
  

  
• [11]
  And some filmmakers are indeed going in this direction : M. Singh, ‘Indian filmmaker Ram Gopal Varma abandons human musicians for AI-generated music’, TechCrunch, September 19, 2024 (at https://techcrunch.com/2024/09/19/indian-filmmaker-ram-gopal-varma-abandons-human-musicians-for-ai-generated-music/). In particular, the film industry is keen in having free hands to use those GenAI tools (and this explains why, for instance, the motion picture association considers that, in some cases, the use of GenAI tools falls under the ‘fair use’ exception under U.S. law (see above note 3 and https://www.motionpictures.org/policy-statement/mpa-comments-to-the-u-s-copyright-office-on-artificial-intelligence-and-copyright/).
• [12]
  OpenAI-written evidence (LLM0113) before the House of Lords Communications and Digital Select Committee inquiry, Large Language Models, December 2023, p. 4 (at https://committees.parliament.uk/writtenevidence/126981/pdf/).
• [13]
  P. Cervini, C. Farronato, P. Kohli and M. W. Van Alstyne, ‘Is AI the Right Tool to Solve That Problem ?’, Harvard Business Review, December 18, 2024 (at https://hbr.org/2024/12/is-ai-the-right-tool-to-solve-that-problem).
• [14]
  See the explanation of the term ‘AI slop’ on Wikipedia (https://en.wikipedia.org/wiki/AI_slop accessed on 28 May 2025). Please note that the references to Wikipedia’s extracts might have to be updated or modified as the content made available on this community platform is likely to change over time.
• [15]
  Ibid. Google explains the process in a blog post entitled ‘Supercharging Search with generative AI’: https://blog.google/products/search/generative-ai-search/.
• [16]
  J. Cox, ‘Google News Is Boosting Garbage AI-Generated Articles’, January 18, 2024 (at https://www.404media.co/google-news-is-boosting-garbage-ai-generated-articles/?ref=mail.cyberneticforests.com).
• [17]
  Based on the Wikipedia entry on ‘AI slop’ (accessed on May 28, 2025).
• [18]
  K Knibbs, ‘Yes, That Viral LinkedIn Post You Read Was Probably AI-Generated’, Wired, November 26, 2024 (at https://www.wired.com/story/linkedin-ai-generated-influencers/?ref=mail.cyberneticforests.com).
• [19]
  See the Meta news of September 25, 2024, at https://about.fb.com/news/2024/09/metas-ai-product-news-connect/.
• [20]
  C. Criddle and H. Murphy, ‘Meta envisages social media filled with AI-generated users’, Financial Times, December 27, 2024.
• [21]
  Ibid.
• [22]
  K. Knibbs, ‘AI Slop Is Flooding Medium’, Wired, October 28, 2024 (at : https://www.wired.com/story/ai-generated-medium-posts-content-moderation/).
• [23]
  Ibid.
• [24]
  And the recent announcement by Meta that its Facebook and Instagram platforms will collect the public data shared by their users, unless they opt-out, to train its Meta AI tool is significative of this need to tap into the user- (now often AI-) generated content shared online. OpenAI, a U.S. leading provider of GenAI tools, also develops a social network so as to be able to swallow lots of data for training its models (see K. Robison and A. Heath, ‘OpenAI is building a social network. Is Sam Altman ready to up his rivalry with Elon Musk and Mark Zuckerberg ?’, The Verge, April 15, 2025, at https://www.theverge.com/openai/648130/openai-social-network-x-competitor). The convergence between social networks and AI platforms is under way, and, at their junction, what has been termed ‘social AI’ is coming.
• [25]
  See K. Martineau and R. Feris, ‘What is synthetic data ?’, IBM blog, February 8, 2023.
• [26]
  I. Shumailov, Z. Shumaylov, Y. Zhao et al., ‘AI models collapse when trained on recursively generated data’, Nature 631, p. 755–759 (2024). https://doi.org/10.1038/s41586-024-07566-y.
• [27]
  See M. Gerstgrasser et al., ‘Is Model Collapse Inevitable ? Breaking the Curse of Recursion by Accumulating Real and Synthetic Data’, submitted on April 1, 2024, available on https://arxiv.org/abs/2404.01413; H. Chowdhury and H. Langley, ‘The AI world’s most valuable resource is running out, and it’s scrambling to find an alternative : ‘fake’ data’, Business Insider, August 9, 2024, at https://www.businessinsider.com/ai-synthetic-data-industry-debate-over-fake-2024-8?utm_source=chatgpt.com. This mid-2024 study prompted my argument made in the present article, and the same view was later confirmed by the French Study for the Ministry of Culture (Note d’étape économique, Mission sur la rémunération des contenus culturels utilisés par les systèmes d’intelligence artificielle, J. Farchy et B. Blain, December 2024 available at 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/mission-relative-a-la-remuneration-des-contenus-culturels-utilises).
• [28]
  I focus here on the related right of performers as it complements the human dimension of copyright which, according to the Continental European tradition (of author’s rights), protects solely original works made by individual authors (not corporate authors).
• [29]
  Art. 53(1)(c) and (d) AI Act require the providers of General-Purpose AI (such as the GenAI tools available on the market) to put in place a copyright policy and a way to identify and comply with the TDM opt-outs, as well as a ‘sufficiently detailed summary about the content used for training’.
• [30]
  C. Geiger and V. Iaia, ‘The forgotten creator : Towards a statutory remuneration right for machine learning of generative AI’, Computer Law & Security Review, 52, 2024.
• [31]
  According to Art. 4(3) of the 2019/790 (CDSM) Directive, TDM is permitted if the use of the works has not been ‘expressly reserved by their right holders in an appropriate manner, such as machine-readable means in the case of content made publicly available online.’
• [32]
  Art. 5(2) of the 2001/29 InfoSoc Directive allows for exceptions or limitations to the reproduction right in a limited number of cases, including mainly, ‘(b) in respect of reproductions on any medium made by a natural person for private use and for ends that are neither directly nor indirectly commercial, on condition that the right holders receive fair compensation which takes account of the application or non-application of technological measures’. Art. 5(2)(a) and (e) InfoSoc dir. similarly allows to introduce a ‘fair compensation’ system ‘in respect of reproductions on paper or any similar medium’ (the reprography exception/levy) and ‘in respect of reproductions of broadcasts made by social institutions pursuing non-commercial purposes’.
• [33]
  M. Senftleben, ‘Generative AI and author remuneration’, IIC-International Review of Intellectual Property and Competition Law 54.10 (2023), p. 1535-1560.
• [34]
  To demonstrate this, a review of the CJEU case law on the communication to the public right should be made, this would require some developments that go well beyond this short article.
• [35]
  M. Senftleben, ibid., p. 1537.
• [36]
  This complex issue is not addressed in this short contribution.
• [37]
  See the Draft Royal Decree, ‘to regulate the granting of extended collective licenses for the massive exploitation of works and other subject matter protected by intellectual property rights for the development of general-purpose artificial intelligence models’ presented by the Spanish Minister of Culture on November 19, 2024, and submitted to a public consultation.
• [38]
  S. Carre, S. Le Cam, and F. Macrez, Buyout contracts imposed by platforms in the cultural and creative sector, Study Requested by the JURI Committee, European Parliament, November 2023; French Presidency Report, Effectivité du cadre européen du droit d’auteur (2022), available at : https://data.consilium.europa.eu/doc/document/ST-10629-2022-INIT/x/pdf ; European Commission, Directorate General for Communications Networks, Content and Technology, Study on contractual practices affecting the transfer of copyright and related rights and the creators and producers’ ability to exploit their rights, No. 2023-031, under Framework Contract CNECT/2022/ OP/0036, Brussels (2023), October 19, 2023, p. 5-7.
• [39]
  Many developments are happening within this space, including with various coalitions such as CP2A, the Coalition for Content Provenance and Authenticity, the progress of standardisation bodies such as JPEG Trust (a framework for establishing trust in media), and, at EU level, the Copyright Infrastructure Task Force (or CITF referred to for instance in the Note of the Working Party on IP of the Council of the EU, December 20, 2024, doc. 16710/1/24, annex 2, p. 36 and ff. available at https://data.consilium.europa.eu/doc/document/ST-16710-2024-REV-1/en/pdf). As co-founder (with Philippe Rixhon) and legal counsel of the Estonian company Valunode (see https://www.valunode.com/), I am directly involved with the development of an innovative marketplace for verifiable rights data. Valunode aims to build a marketplace enabling ‘rights holders to declare their creative works, establish their rights across sectors, channels and jurisdictions, and receive registration certificates of machine-readable rights data’.
• [40]
  At the end of the 1990s, people started to refer to TPMs (for technological protection measures) and RMIs (right-management information systems). The 2001/29 InfoSoc Directive contains already two provisions (Art. 6 and 7) dealing with the protection of respectively TPMs and RMIs, but their widespread use never took off. The closed dimension of those private systems developed independently by various stakeholders seriously limited their usefulness and attraction.
• [41]
  This additional remuneration right under collective management should be accompanied with an improved system for governing the collective bodies (CMOs) and prevent the lack of transparency, inefficiencies or even, in some cases, the corruption that has characterised some CMOs in the past.
• [42]
  This dual system already exists in Spanish, German and Belgian laws for certain types of exploitation of protected works. This could be further detailed, see otherwise the presentation by : M. Senftleben and E. Izyumenko, ‘Author Remuneration in the Streaming Age – Exploitation Rights and Fair Remuneration Rules in the EU’, Joint PIJIP/TLS research paper, October 16, 2024, p. 35 and ff. at https://digitalcommons.wcl.american.edu/cgi/viewcontent.cgi ?article=1136&context=research.
• [43]
  See for ex. P. Goethals, ‘De grootste intellectuele hold-up uit de geschiedenis’, De Standard, January 24, 2023. See also : A. Strowel, ‘ChatGPT and Generative AI Tools : Theft of Intellectual Labor ?’, IIC (2023) 54, p. 491-494.
• [44]
  The investments in AI companies have skyrocketed since the release of ChatGPT in November 2022 (see A Strowel and F. Wéry, ‘L’Intelligence Artificielle pour les juristes’, Larcier, 2025, p. 20 et s. with many additional references). The available data shows a strong preference of the financial markets towards companies focusing on AI. This first reflects in the astronomical capitalisation of Big Tech (the ‘Magnificent Seven’), but the amount of money going to smaller AI companies is also a good sign of a market trend in favour of the AI sector. Regarding the Big Tech, it is important to realise that two of the new titans (the companies with a market capitalisation over 1,000 billions USD) are directly linked to the AI market. Indeed, among the top three with a capitalisation over 3,000 billions USD, we now have Nvidia (which was several times during 2024 the most valuable company in the world), i. e., the semiconductor giant that provides the best GPU (Graphic Processing Units) needed to train the Large Language Models (LLMs) as the engine of GenAI. Tesla also belongs to the so-called ‘Magnificent Seven’ as its investments in self-driving and connected cars is a promise for future benefits. Among the other five Big Tech (that were qualified as the FAMGA in English—or GAFAM in French : Facebook (now Meta), Apple, Microsoft, Google [now Alphabet] and Amazon), most of them are heavily investing in AI, except Apple which has remained more cautious so far (but Apple Intelligence—or AI—is coming).
• [45]
  Indeed, an EU regulation that would create a uniform EU copyright could make sense to solve other problems linked to the existing legal architecture of copyright in Europe; see for ex. A. Strowel, ‘Advocating an EU Copyright Title’, in P. Torremans (ed.), EU Copyright Law : A Commentary (2^nd ed.), Edward Elgar, 2021, p. 1104-1117.
• [46]
  And of course, some legal or administrative mechanism would have to be put in place to check and ensure the conditions remain or not in effect.
• [47]
  Such system has its place within copyright (and the related right of performers), although it could as well be designed outside copyright, and promoted through public policies for the creative sectors, but this would put the burden on States.