The various lawsuits initiated in the USA by copyright owners against generative AI developers for illegal use of their works in LLM training are in their early stages. It is not yet clear whether copyright will apply, and whether the « fair use » exception will be accepted or refuted. But the documents provided in the course of these trials, and the judges’ interim rulings, do provide a few elements, and above all the arguments of the various parties. The year 2025 should provide a clearer picture. The battle between tech and culture is an old story, one that goes on and on. Its latest episode concerns the unauthorized use of protected works to drive generative AI. In Europe, the matter has been partially resolved by the IA Act, although its practical application is already raising many questions. On the other hand, for EU law, there is no doubt that the use of works to feed or train technological tools in the broadest sense does fall within the scope of copyright, since Articles 3 and 4 of the 2019 Directive have erected an exception to copyright for this type of use in research. U.S. law is currently, if anything, less advanced. Unlike the EU, there are no specific regulations in this area, so we have to ask the judge to rule on the basis of existing law. Despite the almost 20 lawsuits launched since 2020 by rightsholders from all walks of life – books, music, press, software developers, visual arts…
