Last updated: July 31, 2026
On July 31, 2026, a court in Munich handed the AI music industry its first real defeat in Europe. The Munich Regional Court ruled that Suno, one of the biggest AI song generators in the world, infringed copyright by training on and reproducing music without a licence. Germany's authors' rights society GEMA brought the case, and it won on almost every point.
This is not a warning letter or a settlement offer. It is a binding judgment from a European court, and GEMA's CEO called it "a verdict of global significance." Here is exactly what the court decided, the evidence that sank Suno's defence, and what it means for anyone making or selling music in 2026.
What the Munich court actually ruled
The case (file number 42 O 763/25) was heard by the 42nd Civil Chamber of the Munich Regional Court under presiding judge Elke Schwager. GEMA had filed the lawsuit back in January 2025, accusing Suno of training its models on protected songs and then generating outputs that copied them.
The court agreed. It found that Suno's models (versions 3.5 and 4) did not simply "analyse" music to learn general patterns. Instead, the court held that the actual works were memorised and stored inside the model in a reproducible form, and that users could pull them back out with simple prompts.
"The court is convinced that the musical pieces in question are reproducibly contained in the defendant's models," the ruling stated, adding that "randomness as the cause of their playback can be ruled out."
In plain terms: this was not a coincidence, and it was not the AI being "inspired." The court concluded the songs were inside the system and coming back out.
The evidence: six songs Suno's AI reproduced
GEMA built its case by generating outputs from Suno and comparing them, side by side, with the originals. The court examined six protected works, including three instantly recognisable European hits:
- Rasputin (Boney M.)
- Daddy Cool (Boney M.)
- Mambo No. 5 (Lou Bega)
By feeding in lyrics, styles, and titles, GEMA showed that Suno produced tracks whose melodies and harmonies were, in the court's words, "too extensive and specific to be coincidental." That reproducibility was the heart of the ruling.
Why "we just analysed the data" did not work
Suno's core legal defence rested on two ideas: Europe's text-and-data-mining (TDM) exception, and US fair use. The court rejected both.
On the TDM exception, the judges were blunt: it "does not apply, because the works were not only analysed during training but retained in the models in a reproducible form." The exception is meant to cover learning from data, not storing and re-emitting the data itself.
On US fair use, the court distinguished Suno's situation from cases where "the original training materials were not reproduced for users in the outputs." Here, the opposite happened: "simple prompts led to outputs that were substantially similar to the original musical works." Once the output copies the input, the fair-use argument collapses.
What Suno has to do now
The court ordered Suno to:
- Stop reproducing the six protected works.
- Stop using those works to train its AI model.
- Disclose its revenue linked to the infringement, so damages can be calculated.
- Pay damages, in an amount still to be determined.
Under German law, the injunction is enforceable immediately, even while an appeal is pending. GEMA was represented by the firm Raue (partner Robert Heine); Suno was represented by Latham & Watkins (partner Anne Löhner).
GEMA's position: licence first, not after
"AI providers must acquire licenses and may not use the works of our members free of charge," said GEMA CEO Tobias Holzmüller.
That is the whole point of the case. GEMA is not trying to ban AI music. It is trying to force AI companies to pay for the catalogue they train on, the same way a radio station or a streaming service already does. The ruling gives that argument real legal weight in Europe.
Suno's response and the appeal
Suno said it disagreed with the decision, arguing that the ruling "rests on a fundamental mischaracterization" of how its technology works. The company said it would evaluate "all available options, including an appeal." So this is very likely not the final word, and a higher court could revisit the reasoning.
Still, for now, the precedent stands: in Germany, training a generative model on protected music and letting it reproduce that music is copyright infringement.
Why this is a global turning point
This ruling does not sit in isolation. It follows GEMA's earlier win in November 2025 against OpenAI, where a German court found that ChatGPT had reproduced protected song lyrics. Two AI copyright victories in under a year is a pattern, not a fluke.
The stakes are enormous. Suno was valued at around 5.4 billion dollars earlier in 2026 and, according to reporting on the case, was trained on more than two million songs scraped from the web. Across the Atlantic, the major labels have been pushing hard too: Sony has sued rival generator Udio, and both Suno and Udio have reportedly moved toward licensing settlements with Warner Music rather than fighting every battle in court.
The direction of travel is clear. The "train on everything, ask permission never" era of AI music is closing. Licensing is becoming the price of doing business.
What this means for producers and artists
If you make music, this ruling matters more than it might look. It reinforces something that has been true all year: music generated entirely by AI sits on shaky legal ground, both for the platforms that make it and, potentially, for the people who release it. We covered the money side of that in why your AI-generated songs won't make money in 2026.
The safest way to build tracks you actually own is still the oldest one: start from real, licensed human vocals with a clear paper trail. That is exactly why producers compare AI vocals vs human vocals before committing a release, and why cleared acapellas and vocal stems keep winning.
Want vocals you can release without wondering who really owns them? Every acapella and cover vocal on The Vocal Market is performed by a real vocalist and licensed to you, so your next track starts on solid legal ground, not inside a lawsuit.
