What did the Munich court actually find?
On 31 July, the 42nd civil chamber of Munich I Regional Court, led by Judge Elke Schwager, ruled that Suno infringed copyright by training its v3.5 and v4 models on songs represented by GEMA, Germany's music-royalties collecting society, without a license. The case centered on six compositions, among them Boney M.'s "Rasputin" and "Daddy Cool" and Lou Bega's "Mambo No. 5" (the melodies only, not the lyrics). GEMA showed the court that Suno's outputs matched the originals in melody, harmony and rhythm closely enough that the judges concluded the works were reproducibly contained in the models, stored on German servers. That's the crux of the ruling: the court treated this as memorization, not the kind of statistical learning AI companies usually lean on to argue their models don't copy anything. Suno argued the opposite, that its models generate new songs rather than reproduce existing ones. The chamber didn't buy it.
What does Suno actually have to do now?
The court issued three orders. An injunction: Suno has to stop reproducing the six works. A disclosure order: Suno must open its books and reveal the revenue tied to the infringing outputs, the figure that will set the damages bill. And a damages order: Suno owes GEMA payment, with the exact amount still to be calculated once the revenue disclosure lands. GEMA's chief executive, Dr Tobias Holzmuller, put it bluntly.
"AI providers must acquire licenses and may not use the works of our members free of charge. AI models based on the theft of intellectual property are not protected by law."
Suno's response was just as direct: the company says it "disagrees with today's ruling," calls the decision a "fundamental mischaracterization" of its technology, and is "evaluating all available options, including an appeal."
Why does this ruling matter beyond six songs?
This is the case Time To House flagged as pending back on 23 July, and the verdict lands as one of the first rulings anywhere in the EU to treat training an AI model on copyrighted music without a license as the infringement itself, independent of what the model later outputs. GEMA has other AI companies in its sights, and this judgment is the template it will point to next. For labels, publishers and every other European collecting society, from SACEM in France to SIAE in Italy, it's the first hard evidence that a court will side with rightsholders over the "the model just learns patterns" defense. For Suno and every AI music generator built on scraped catalogs, it's the opposite: a real cost, attached to a real number, coming after a real disclosure order. The appeal will take time, likely years, and the ruling isn't final until a higher court weighs in. But the legal theory that got Suno here, that memorization inside a model equals infringement, doesn't need Suno to lose the appeal to already be doing its job on the rest of the AI music industry.
