German court rules Suno violated music copyrights, rejects fair use defense

A Munich court ruled that AI music generator Suno violated copyrights through both its training process and generated outputs. The case was brought by German music rights organization GEMA and centered on six well-known songs, with only the musical compositions at issue, not the lyrics.

The ruling is not final, but it marks a significant legal test for AI music systems that train on copyrighted works and can produce similar material on request.

The court found that all six tracks were reproducibly contained in Suno’s version 3.5 and 4 models. It treated this as memorization: a model storing specific content from training data rather than only learning general musical patterns.

GEMA tested the system by entering each song’s original lyrics, musical style, and title, without specifying melody, harmony, rhythm, or arrangement. Suno still generated results in which the court recognized original elements, and the court ruled out coincidence given the complexity and length of the songs.

Suno argued that user prompts broke the causal chain between the model and any infringing output. The court disagreed, saying the prompts were simple and open-ended while Suno operated the models, selected training data, and controlled model architecture.

That reasoning matters for other AI music services: if upheld, merely offering a generator that can reproduce protected works could create legal exposure for the company behind it, not just the person entering prompts.

The court rejected Suno’s reliance on Germany’s text-and-data-mining exception, finding it did not cover the memorization at issue. It also applied US law to training activities that took place in the United States and concluded that fair use did not protect Suno.

The court distinguished Suno from US AI training cases where outputs did not make training data substantially accessible to users. Here, the court said simple inputs produced results substantially similar to the originals, while also noting that key questions remain unresolved.

The court’s press-release notes said Suno used stream-ripping techniques to extract music from YouTube and bypassed YouTube’s Rolling Cipher, a technical safeguard designed to prevent downloading audio and video content.

That shifts part of the debate from output similarity to whether copyrighted material was lawfully acquired for training in the first place. The takeaway for AI developers is direct: training-data sourcing, technical safeguards, and output behavior are all becoming central legal risks.