I - AI tools, new tools for creators...
If we put the question "AI tools, new tools for creators ... what do you think?" to one of the best-known AI models, we get the following answer:
"Hello! I find this expression, "AI tools, new palettes available to creators", very telling and inspiring. It highlights the extent to which artificial intelligence offers creators a rich and varied range of tools, similar to a painter's palette of colours. These tools make it possible to explore new ideas, push back the boundaries of creativity, and innovate more fluidly and efficiently. AI thus becomes a partner that enriches the creative process, offering infinite possibilities for imagining, experimenting and realizing original projects. In short, it's a real revolution that opens up new perspectives for all creative people, whether they're artists, designers, writers or innovators. If you'd like to discuss this further or find out about some concrete examples, I'm here to help!"
According to AI itself, its very existence should revolutionize creation and innovation by speeding up processes, expanding creative possibilities and transforming working methods.
The use of artificial intelligence by artists is a fast-growing phenomenon with far-reaching implications for artistic creation. AI can be a powerful tool for artists, helping them to push back the limits of their creativity, automating certain repetitive tasks or proposing new solutions.
For example, the artist Refik Anadol, "fascinated by the memory and creative potential of machines, has made data his favourite material, and artificial intelligence his main collaborator" (free English translation), presented an immersive installation entitled "Machine Hallucination - Dream of Nature" at the Centre Pompidou-Metz in August 2022.
Whether in the arts or in innovation, creators have not hesitated to test and appropriate these new technologies.
AI undoubtedly offers new possibilities for artistic exploration, while provoking discussion about the nature of creativity, the authorship of creation or innovation or its originality, and questioning the collaboration between man and machine.
Here again, as AI pointed out when we asked: "The artist does not disappear in the process of AI-assisted creation. On the contrary, he often becomes a curator or guide for the machine, selecting parameters, refining the results or combining elements generated by the AI with personal touches. AI can be seen as a tool that broadens the artist's creative horizon."
Of course, AI does not condemn itself and is careful not to approach the question from the angle of using previous creations in the process of generating a "new" creation through it.
If we compel it to do so by asking it again: "What do you think about using pre-existing works in the creative process via generative AI? "The machine acknowledges that "the use of pre-existing works in creation with a generative AI raises a number of fascinating and complex questions, from an artistic, legal and philosophical point of view", and goes on to offer a series of reflections on the difference between inspiration and plagiarism, the question of intellectual property rights attached to pre-existing works, ethics and the evolutionary process of artistic creation.
Quite correctly, it sums up its position as follows: "the use of pre-existing works in generative AI creation is an interesting frontier between innovation and respect for the rights of original creators. There is a great opportunity to enrich the creative process, but also significant challenges to be faced regarding the legality and ethics of this practice. It's an area that will continue to evolve, in line with technology, legal debates and cultural perceptions of what it means to be an 'artist'."
II - A powerful technology that may infringe the copyright of third parties: a colouring box for counterfeiters?
The rapid development of generative artificial intelligence (generative AI) is profoundly transforming the way we create. Capable of producing texts, images, music or even code from simple instructions, these tools arouse both fascination and concern. While they offer new prospects for creators, they also raise major legal issues, particularly regarding copyright.
In some cases, generative AI acts as a veritable colouring box for forgers, making it possible to reproduce styles, shapes or elements of protected works without respecting their rights.
One of the main legal issues is the possible use of vast databases of protected works by AI models during their training. In the absence of authorization from the rights holders, this use could constitute copyright infringement. In addition, the content generated could incorporate characteristic elements of existing works, creating an increased risk of counterfeiting.
Throughout the world, numerous legal proceedings illustrate these tensions:
- In the UK, Getty Images1, a leading image bank, and other companies in the group, as well as a freelance photographer, are suing Stability AI, a UK-based artificial intelligence company. They accuse it of using their protected images without authorization to train its AI model, Stable Diffusion. They accuse Stability AI of copying images from Getty Images' websites without their permission, thereby infringing their copyright and databases, as well as constituting a deceptive commercial practice.
Getty Images is now seeking 1.7 billion dollars in damages, compared with the initial 1 billion, for the unauthorized use of 11,383 works, instead of 7,300 images in the initial complaint. At this stage, the proceedings are confined to issues relating to the admissibility of one of the plaintiffs (Thomas M Barwick Inc), and the first judgment handed down on 14 January 2025 concerned a request for Stability AI to exclude the possibility of this plaintiff representing a class of copyright holders whose works have been licensed exclusively to Getty. The court dismissed Stability AI's claim on the basis that the class of claimants was ill-defined and refused the option of proceeding without including all exclusive licensors on the basis of insufficient evidence. The court also raised difficulties in identifying copyrighted works used by Stable Diffusion, suggesting sampling as a method of resolving infringement issues. However, there was no consensus on this approach. The technical complexities of analyzing massive data were acknowledged, and it became clear that implementing such a process would be virtually impossible without significant resources. Small companies in the AI sector also face compliance hurdles. The proceedings on the merits, scheduled for June 2025, could have significant implications for the use of data to train AI models and redefine how creators protect their works. This case could set legal precedents influencing the future of copyright and the development of AI.
- In the US, several lawsuits concerning copyright infringement by generative AI applications have been launched recently. For example, a class action against Stability AI, Midjourney, and DeviantArt is currently pending in the US District Court for the Northern District of California. In addition, image bank Getty Images has also sued Stable Diffusion in the US, accusing it of using its images (over 12 million works) to train its AI model without permission or financial compensation.
In December 2023, the New York Times2 filed a copyright infringement suit against OpenAI and Microsoft in the US District Court for the Southern District of New York. The newspaper accuses them of having used, without authorization or compensation, its articles to train their AI models, in particular ChatGPT and Bing Chat, and of reproducing extracts of its content in the responses generated. In their defense, OpenAI and Microsoft invoked the "fair use" provisions, arguing that the use of the content was transformative and that certain extracts were the result of atypical practices known as "prompt hacking". They also attempted to demonstrate that the New York Times itself used generative AI technologies, but the court rejected their request to produce documents, deeming these elements irrelevant.
On 22 November 2024, the court reiterated that the "fair use" analysis focuses primarily on the use made of the protected work and its impact on the market in which it is exploited, and not on the commercial or technological practices of the plaintiff. In refusing to produce documents relating to the New York Times' own AI projects, the court made it clear that this type of peripheral argument was not likely to justify unauthorized use of protected works. This decision marks a turning point: it confirms that the "fair use" defense in the context of generative AI must be firmly based on traditional copyright criteria - the nature of the work used, the extent of the borrowing, the actual transformation of the content, and the effect on the market - and not on general considerations linked to technological developments or the economic strategy of the parties. It could thus foreshadow stricter supervision of AI training on protected databases.
There is also the Sarah Andersen case3, which pits three American artists against Stability AI, DeviantArt and Midjourney in the United States District Court for the Northern District of California in San Francisco. These companies are accused of having trained their generative AIs on protected works without authorization, in violation of copyright, trademark law and the rules of fair competition.
In October 2023, Judge Orrick dismissed some of the claims but gave the artists the opportunity to amend their complaint. In an important decision in August 2024, the judge allowed copyright and trademark infringement actions to proceed. Companies will now have to disclose precise information about the data used to train their AIs and whether any protected works are present. This decision is a major step towards recognizing the rights of creators in the face of the abuses of generative AI.
Finally, very recently, on 11 February 2025, the United States District Court for the District of Delaware handed down an important decision4 in favor of Thomson Reuters in its dispute with ROSS Intelligence, a start-up specializing in artificial intelligence applied to the legal field. ROSS was accused of using, through a third party, copyrighted content from Thomson Reuters' "Westlaw" database to train its AI system. Although the AI in question was not a generative AI as such, the court ruled that the use of protected works was not covered by the "fair use" exception. The court noted that the use made by ROSS was for commercial purposes and was in direct competition with Thomson Reuters' services. It also noted that the AI developed by ROSS did not significantly transform the original content, an essential condition for fair use. This ruling marks a turning point in the application of copyright to the data used to train AIs, by affirming that the commercial purpose and the absence of substantial transformation of the works can preclude upholding the fair use defense. It could thus have a lasting influence on the way technology companies, beyond the legal sector, envisage access to protected databases to train their artificial intelligence systems. It should be noted, however, that an appeal was lodged by ROSS Intelligence against this decision on 14 April 2025, so the case is far from over.
- Things are moving in this direction in China too. On 8 February 2024, the Guangzhou Internet Court handed down an important ruling5 on the use of generative AI systems, one of the first court cases to deal with copyright infringement caused by the automatic generation of works.
The owner of the copyright in a series entitled "Ultraman" had noticed that images representing his character were being generated by an artificial intelligence system in response to the use of the keyword "Ultraman" as a prompt. An infringement action was brought before the Guangzhou Internet Court, which recognized the copyright infringement. The court ruled that the first image generated constituted an unauthorized reproduction of the character, thereby infringing the exclusive right of reproduction. It also held that other images, obtained via different prompts, constituted unlawful derivative works, in the absence of authorization from the rights holder, thereby infringing the right of adaptation. Finally, the court held the provider of the generative AI service liable for allowing these acts of infringement. This decision could become a benchmark in the supervision of generative AI, including in other jurisdictions such as the United States and Europe, where similar litigation is underway.
- In Germany, copyright issues relating to artificial intelligence are also undergoing major developments. In a decision dated 27 September 2024, the Hamburg Regional Court provided a noteworthy interpretation of the provisions and data mining (TDM) exceptions provided for in Directive (EU) 2019/790 on copyright and related rights in the digital single market. The case before the court concerned the aggregation of data extracted from protected works with a view to constituting "datasets" intended for training AI systems. The main issue at stake was the possibility of benefiting from the exception provided for scientific research, even when these "datasets" are subsequently used in projects with a commercial dimension.
In its decision6, the court adopted a pragmatic and flexible approach. It found that the training of AI on datasets created for scientific purposes did not affect the normal exploitation of the works concerned and thus complied with the conditions of the so-called "three-step test" required by the Directive. The court emphasized that the immediate purpose of the use was scientific research, irrespective of any subsequent commercial exploitation. This analysis marks a turning point. By adopting a favorable interpretation of TDM activities for research purposes, the Hamburg court is strengthening the legal certainty of those involved in the development of AI in Europe. However, it specifies that this interpretation only benefits not-for-profit organizations operating within a scientific research framework. The ruling leaves open the sensitive question of the purely commercial use of the data extracted, leaving this issue for future judicial arbitration.
Ultimately, this decision could influence all the European courts called upon to rule on the delicate balance between protecting the rights of creators and encouraging technological innovation, particularly in the field of generative AI.
- Finally, in France, the first decisions have been slow in coming, but a large number of proceedings have been initiated.
On 6 March 2025, the Syndicat national de l'édition (SNE), the Société des gens de lettres (SGDL) and the Syndicat national des auteurs et des compositeurs (SNAC) summoned Meta Platforms Inc. before the 3ᵉ chamber of the Paris Tribunal judiciaire. They accuse "Meta" of having used, without authorization, works protected by copyright to train its generative artificial intelligence models, as part of the development of its "LLaMA" models. The plaintiffs are seeking compensation for the infringement of their economic rights, claiming in particular unlawful reproduction and communication to the public of their works.
In June 2024, several organizations representing the French press, including the Alliance de la Presse d'Information Générale (APIG) and the Syndicat des Editeurs de la Presse Magazine (SEPM), sent formal notices to twenty-five developers of generative AI models, including OpenAI, Google and Mistral AI. These organizations are calling for negotiations to be opened with a view to obtaining fair remuneration for the use, without authorization, of their protected content in training AI models. They are relying on the press-related rights regime provided for in Directive (EU) 2019/790 on copyright and related rights in the digital single market, transposed into French law. In the absence of a satisfactory response or agreement, the publishers reserve the right to take legal action to obtain compensation for any infringements found.
In a ruling handed down on 7 May 2025 (RG 25/02294)7, the Paris Magistrates' Court ordered French ISPs to block the "news.dayfr.com" website, which uses indexing and rewriting technologies to illegally reproduce press articles. The action was brought by a group of press publishers, including La Montagne, Libération and Le Télégramme, on the grounds of copyright infringement under the French Intellectual Property Code. The Court found that the site reproduced protected articles without authorization, with minor modifications and without any legal notice, using generative AI, which showed that the site was unlawful. The Court ruled that the site was manifestly unlawful and imposed an 18-month block on the ISPs. This decision could pave the way for other actions against platforms that use automated models to distribute protected content.
These decisions illustrate the ability of copyright to act as a solid bulwark against the excesses of unbridled generative AI, by reaffirming that technological innovation must not be at the expense of creators' fundamental rights.
Footnotes
3 SARAH ANDERSEN, v. STABILITY AI LTD., Case 3:23-cv-00201-WHO, August 12th, 2024
5 Guangzhou Internet Court, 8 February 2024, SCLA v Tab.
6 Tribunal de Hambourg, 27 sept. 2024, n° 310 O 227/23
7 Paris Court of First Instance of 7 May 2025, RG 25/02294
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
