Several courts have opened proceedings against the companies behind certain generative AI systems, accused of violating intellectual property rights. Amongst plaintiffs are Walt Disney , and Universal (a subsidiary of Comcast), which are taking action against Midjourney; Getty Images, which is in litigation with Stability AI; and the New York Times, which has filed a lawsuit against OpenAI and Microsoft. All these cases are based on two pillars: the accusation of misuse of works protected by the models, and a defense systematically based on "fair use." This principle allows, in certain cases, the limited use of protected works for "transformative" purposes.

But this is where the problem lies: the concept of transformation remains vague and open to broad interpretation. At the heart of the issue is the lack of a strict legal framework for generative AI. As a result, each judge has relative freedom in their assessment, and previous decisions serve as a legal compass. In a case between Reuters and the startup Ross Intelligence (now closed), the judge ruled that "fair use" did not apply because there was no proven transformation. He even added that this principle could not be invoked when the content used made it possible to create a "market substitute."

Given this lack of clarity, it remains difficult to predict the outcome of the ongoing proceedings. While the threat of fines hangs over the companies concerned, it is above all the fear of being forced to delete some of their training data that worries players in the sector. Losing a learning database risks creating a huge gap in the model's knowledge, and this loss could be far more serious than any financial penalty.