The Munich Regional Court ruled against Suno, the AI music generator, in a case brought by GEMA, Germany's state-mandated licensing body. The decision requires payment for the systematic use of GEMA's repertoire.
The reach is what separates this from earlier cases. The court addressed two distinct stages: the use of the works to train the model, and the material the model produces afterwards. On the argument the court accepted, simple open-ended prompts produced outputs substantially similar to the original works, which distinguishes it from cases where the training material never reappeared recognisably in the output.
It is the same chamber's second consecutive finding against an AI developer on training-data grounds. In November 2025, the same judge held that OpenAI had unlawfully used the lyrics of nine well-known German songs.
Suno said it disagrees with the ruling and is appealing.
In the same period, on the other side of the Atlantic, five publishers sued Meta over the data used to train Llama, in a case where the judge found the plaintiffs hadn't shown sufficient market harm. He was careful to note the decision was narrow and did not establish that training on copyrighted material is lawful in every circumstance.
The two rulings arrive at the same place from opposite directions: the claim that training is always fair use no longer passes without examination. What decides the outcome is how much of the original work reappears in what the tool hands the user.