GEMA v. Suno: what the Munich AI ruling means for producers

On July 31, 2026, the Munich I Regional Court ruled in GEMA v. Suno: using protected musical works to train an AI without a license violates German copyright law. GEMA, Germany's collecting society for music, won on nearly every point against the AI music generator (case 42 O 763/25). The verdict came from the court's specialized copyright chamber, presided over by Judge Elke Schwager.

Here's what the ruling says, how GEMA proved its case, and why it matters for you.

What the court decided

Suno can no longer reproduce the six musical works at issue without a license, and the court spelled out that this covers using them as training material for its AI models. Suno also has to disclose how the works were used, so GEMA can put a number on the damage. And the court established that Suno owes damages in principle. The amount is still open.

The works in question: "Atemlos", "Daddy Cool", "Mambo No. 5", "Big in Japan", "Forever Young" and "Rasputin". Six songs almost anyone can hum. The case was about the compositions, not the lyrics.

One thing to keep in view: this is a first-instance ruling and not yet final.

How GEMA proved it

The evidence is the part I find most interesting as a producer. GEMA simply used Suno. Song titles, lines of lyrics and style prompts went into the interface, the outputs came back, and they were compared against the originals. According to the court, the results landed so close to the source material that coincidence was off the table.

From there the chamber drew the decisive conclusion. In the judge's words, the court is convinced that the musical pieces in question are "reproducibly contained" in the defendant's models (the case looked at versions v3.5 and v4). Not stored as a file in the classic sense, but retrievable enough that the model can serve them back up on request. Lawyers call this memorization.

Analysis is fine, memorizing is not

Which brings us to the core of it. German copyright law has a text-and-data-mining exception (Section 44b of the Copyright Act) that AI companies like to lean on: if you merely analyze publicly available works to learn patterns, you don't need a license. The Munich court reads that exception narrowly. Analysis is covered. A model that ends up carrying the works inside itself and can play them back is not.

In the court's view, that produces two separate infringements: storing the works in the model violates the reproduction right, and delivering them to users violates the right of communication to the public.

This line of reasoning isn't new. The same chamber argued the same way in November 2025 in GEMA v. OpenAI, a case about song lyrics that ChatGPT would recite on demand. That ruling is under appeal. What's new is that the reasoning has now been applied to music itself. Several observers, including the trade outlet MusikWoche and the law firm WBS Legal, read this as the first decision in Europe to establish a licensing requirement for training music AI.

A side note worth knowing: according to press reports, Suno pulled its training data via stream-ripping from freely accessible sources such as YouTube, and had previously turned down a paid licensing deal with GEMA. That license is exactly what the court has now made mandatory.

What Suno says

Suno, unsurprisingly, sees it differently. The company said it disagrees with the ruling, that the decision rests on a mischaracterization of how its technology works and is used, and that it is "evaluating all available options, including an appeal". Suno also stresses that its platform is built to create new songs rather than reproduce existing ones, and points to safeguards built into the product.

GEMA's leadership, CEO Tobias Holzmüller and general counsel Kai Welp, called the outcome a complete success.

What it means for you

First, the sober part. Six songs are not a catalog, and a non-final ruling from a court of first instance is not the end of the road. But if the decision holds, it changes the ground AI music companies stand on in Europe. Training data stops being a free buffet and becomes something that has to be licensed and paid for.

For anyone who writes and releases music, that's the bigger story than any single line of the verdict: your catalog has value as training material, and there is now a German judgment backing that value up. Whether this turns into a working licensing market where money actually reaches songwriters will be decided over the next few years, in appeal courts and in negotiation rooms.

And if you use AI tools yourself? The ruling doesn't ban AI music. It says the companies behind the generators need licenses for their training material. What this means for the use of the outputs is a question the court did not decide. Suno tried to shift responsibility onto its users during the proceedings and didn't get anywhere with that.

Think what you want about AI generators. But the order of operations this ruling locks in is the right one for anyone who makes a living from their own catalog: license first, then train.