AI and counterfeiting: lessons learned from the first legal cases
In Europe and the United States, a number of court rulings have been handed down on counterfeiting following the use of artificial intelligence. AI and counterfeiting can therefore coexist. That’s why Numeum has brought together experts and professionals to discuss how acts of counterfeiting can be characterized and the exceptions that can be raised.
How can we deal with cases where artificial intelligence is used to generate counterfeit products? As the technology becomes more and more widely used, companies need to guard against the ripple effect of AI’s popularity. They need to protect their intellectual property assets. Numeum has brought together professionals and experts on the subject, in the light of the latest court rulings in Europe and the United States.
Despite the complexity of the subject, it is accepted that there is no universal definition of AI. The notion is therefore relatively vague and general. However, Article 3§63 of theAI Act defines“an AI model, including where such AI model is trained using large-scale self-supervision, that exhibits significant generality and is capable of competently performing a wide range of distinct tasks, regardless of how the model is brought to market, and which can be integrated into a variety of downstream systems or applications, with the exception of AI models used for research, development or prototyping activities prior to market release.”
In this context, generative AI training systems can use protected works contained in large databases. These works are often exploited without the knowledge or consent of the rights holders.
“To counter this practice, detection tools could be used to detect the illicit content generated. Specific recourse mechanisms could also be put in place “explains Naomi Meynle-Hamza, a CIFRE lawyer at Cabinet Racine.
AI and counterfeiting: acts that can be considered as such
The challenge lies in the fact that a training data set systematically involves acts of reproduction. Either by downloading works or by making copies. “Even if this data is deleted after the systems have been trained, there is still an infringement of copyright “, explains Maître Charles Bouffier, associate lawyer at the Racine law firm. The spectrum is therefore wide when it comes to characterizing an infringement following the use of artificial intelligence.
In the same vein, the use of RAG (Retrieval-augmented generation), a popular technique that combines information retrieval and content generation, can be an infringement. A developer can copy content from a dedicated database, or query conventional websites to integrate it into his or her response.
Maître Charles Bouffier explains:“This customization of the tool is likely to implement the reproduction rights of an author on the elements that can be integrated “. In short, RAG is not legally neutral, as it raises questions of liability for developers, hosts and professional users. In France, an unprecedented decision by the Paris judicial court on May 7, 2025 marked the first victory for publishers against an AI-generated site. The court condemned companies that had used press articles without authorization. In the United States, some forty proceedings are also underway on infringement grounds (Disney and Universal Studios against Midjourney, for example).
Exceptions to conviction and individual responsibilities
Certain uses can justify the use of a protected work. The Law for the Digital Republic, much of which has been taken up in the Digital Services Regulation (DSA), prohibits authors from objecting to a reproduction of their work for the purposes of obtaining information or improving research tools. This exception could be taken up in the 2019 European Copyright Directive when it comes to scientific uses. Also, if the reproduction is temporary and has no economic value of its own, copyright cannot be invoked.
The fact remains that, in the event of proven infringement, the user is more likely to be held liable than the developer or the system itself. This is particularly true if the user expressly asks to reproduce a protected work or uploads such elements.
For this reason, the experts make a number of recommendations. As a preventive measure, developers ofartificial intelligence tools should obtain database licenses in advance. They must also be transparent about the data sources they use. Finally, developers must take into consideration the potential legal disputes that may arise from the use of such tools.
To find out more, here’s support and a replay of the conference: https: //preprod25.numeum.fr/focus-sur-la-contrefacon-au-moyen-de-systemes-dia-generatives-analyse-comparee-des-premieres-decisions-europeennes-et-americaines/
In order to understand the issues mixing AI and counterfeiting, Numeum offers its guide to implementing the AI Act proposed by Numeum. It is available at this address : https://preprod25.numeum.fr/europe/actu-informatique-guide-de-mise-en-oeuvre-de-lai-act/
