Artificial Intelligence, Copyright and Authors' Rights: Kadrey vs. Meta Platforms, the US Vision

By Marine Vanhoucke et Philippe Ginestié, Ginestié Paley-Vincent, Lawyers.

In the context of AI using copyright-protected content, Marine Vanhoucke and Philippe Ginestié analyze the recent “Kadrey vs. Meta Platforms” decision.

The recent case of KADREY vs. META PLATFORMS (District Court of California, June 25, 2025) provides an interesting perspective on copyright issues in the era of artificial intelligence (AI). Thirteen American authors (including Richard Kadrey and Sarah Silverman) filed a lawsuit against Meta Platforms, Inc. They allege that their works were used without authorization to train Meta's Llama generative AI models, thereby constituting a violation of their copyright.

In this decision, the California judge has so far partially rejected the plaintiffs' claim, relying on the American doctrine of "fair use," which allows the use of protected content without prior authorization from the rights holder. This exception, codified in Section 107 of the Copyright Act of 1976, is based on four key criteria:

• The purpose and character of the use (educational rather than commercial and transformation rather than imitation);

• The nature of the protected work and in particular its degree of creativity (“creative work” vs. “work based on facts”);
• The quantity and importance of the part used in relation to the whole of the work;
• The effect of the use on the potential market or value of the copyrighted work.

The California Court first found that Meta's use of the plaintiffs' works to train its AI models was highly transformative because generative AI models create new content from their analyses of training data, which is by definition extremely large, and thus makes creative use of the original works.

Furthermore, although Meta's Llama models are freely available, they were developed for commercial purposes, which could have tilted the analysis towards a lack of Fair Use. This factor was not considered decisive due to the particularly transformative nature of the use.

Finally, the Court concluded that the plaintiffs failed to demonstrate that Meta's use would cause significant harm to the market for their works.

A partially favorable judgment was therefore granted to Meta Platforms, although the claims relating to possible moral damages have not yet been decided at this stage.

This decision highlights the need to adapt the Fair Use doctrine in the face of new technologies such as generative AI. The Court emphasized the importance of considering the potential effects of these technologies on the market for copyrighted works. It emphasized the specific nature of this decision, as each case may be open to discussion.

Other legal actions in similar contexts are underway in the United States and also in France.

This decision takes on particular significance in light of the proceedings initiated last March by the Syndicat national de l'édition, the Société des gens de lettres, and the Syndicat national des auteurs et des compositeurs against Meta Platforms Inc. before the Paris Judicial Court. It is alleged that Meta allegedly used the Books database, containing approximately 200 books, to train its generative AI model. Interestingly, an important piece of evidence in this case is Meta's admission, in the context of the KADREY vs META PLATFORMS action in the United States, of its use of the Books database.

The American decision cannot in any way prejudge the outcome of the proceedings initiated in France, where there is no Fair Use doctrine but rather a TDM (Text and Data Mining) exception. This exception to copyright introduced by the DAMUN Directive (Directive No. 2019/790) and codified in Article L. 122-5-3 of the Intellectual Property Code, authorizes any person to carry out searches, regardless of the purpose, provided that the author has not expressed his opposition in an appropriate manner. AI providers thus potentially have access to large volumes of protected works unless their authors have objected. The burden of proof is reversed in favor of AI providers, who would therefore no longer have to seek the authors' authorization.

In this context, a position taken by the French judge on the application of this TDM exception will certainly be as informative as monitoring the development of American case law relating to Fair Use.