Suno copyright ruling rejects fair use defense in German music case

Munich Regional Court I has ruled against AI music company Suno in a copyright case brought by German collecting society GEMA, making the Suno copyright ruling a notable setback for generative music services. The court found infringement in both the company’s model training and the music its system produced, and it rejected Suno’s fair use argument under U.S. law.

GEMA had sought an injunction, disclosure and damages. According to the court and GEMA’s case, the dispute covered six well-known songs, including Kristina Bach’s “Atemlos durch die Nacht” and “Rasputin” by Frank Farian, Fred Jay and George Reyam. The case concerned musical compositions rather than lyrics.

What did the Suno copyright ruling decide?

The court found that Suno’s versions 3.5 and 4 models reproducibly contained all six works at issue. In the court’s view, the models had retained protected musical material closely enough that prompts could cause outputs containing original elements from the songs.

Suno argued that its systems did not store songs, and instead learned mathematical patterns and broader musical features. The company also argued that any overlap between an output and a protected work came from user prompts and statistical association, rather than from the company itself reproducing a song.

The court was not persuaded. GEMA tested the system by entering the original lyrics, the musical style and the title for each song, without specifying melody, harmony, rhythm or arrangement. The court said the resulting outputs still contained recognizable original musical elements, and it ruled out chance as an explanation because of the length and complexity of the works.

Why the court put liability on Suno

Suno had argued that users’ prompts interrupted the chain of responsibility between the model and the generated music. Munich Regional Court I instead placed responsibility on Suno, citing the company’s control over the model, its training data choices and its system design.

The court described the prompts as simple and open-ended, and said the models substantially determined the outputs. It also said that offering the generator for music creation was itself legally relevant. If that reasoning survives appeal, it could become a problem for other AI music products whose systems can be induced to recreate protected works.

The court also rejected Suno’s reliance on Germany’s text and data mining exception. That exception permits certain automated analysis of protected material, but the court said it did not cover the memorization it found in Suno’s models.

How U.S. fair use figured into a German case

Under rules applicable to collecting societies, the Munich court also addressed training activity said to have occurred in the United States. Applying U.S. law, it concluded that fair use did not shield Suno.

The court distinguished Suno’s case from the Bartz and Kadrey proceedings in the U.S., where courts had treated AI training as transformative. According to the Munich court, the difference was that those cases did not involve training material being made available to users in outputs in the same substantial way. In Suno’s case, the court said simple inputs produced outputs substantially similar to protected works. The ruling is not final.

The court’s press release also said Suno used stream-ripping techniques to extract music from YouTube and bypassed YouTube’s Rolling Cipher, a technical measure intended to prevent downloading audio and video. That allegation raises a separate issue from output similarity: whether the way training data was obtained can defeat an AI company’s legal defenses even before the model’s behavior is considered.

This story draws on original reporting from The Decoder.