.png){kind=small}
Munich Court Rules Against Suno AI in Landmark Copyright Case
- Maximillian Wollenberg
- 21 hours ago
- 8 min read
A German court has handed the music industry one of its strongest legal wins yet against unlicensed AI training.
On July 31, 2026, the Munich Regional Court ruled that Suno AI breached copyright law by using protected musical works to train its AI music models without a license. The lawsuit was brought by GEMA, Germany’s music rights organization, which represents more than 100,000 composers, lyricists, and music publishers.
The ruling matters well beyond Germany. It marks Europe’s first major decision against unlicensed AI music training and gives rightsholders a concrete legal path for challenging AI systems that can reproduce recognizable parts of protected songs.
The court found that versions 3.5 and 4 of Suno’s models had “memorized” protected works. According to GEMA’s case, when prompted, those systems generated outputs that matched the melody, harmony, and rhythm of copyrighted songs in its repertoire.
This article is for informational purposes only and does not provide legal advice.
What the Munich court decided
The 42nd Civil Chamber of the Munich Regional Court sided with GEMA on the central points of the case. The court held that Suno had infringed copyright by using protected musical works in ways that required permission.
The most striking part of the decision is how the court treated the AI model itself. It ruled that storing copyrighted tracks in model weights can amount to reproduction under copyright law. That finding goes to the heart of how generative AI systems are built.
Suno argued that its training practices should be protected under U.S. “fair use” principles or under the European Union’s text and data mining exceptions. The court rejected both defenses.
The decision also ordered Suno to stop unauthorized reproductions and public distribution of GEMA’s repertoire within Germany. The company must disclose revenue connected to the infringements and pay damages, though the amount has not yet been set.
Issue | Court’s finding |
Use of protected music | Suno used works from GEMA’s repertoire without a license |
Model memorization | Versions 3.5 and 4 generated outputs that matched protected songs |
Legal defenses | U.S. fair use and EU text and data mining exceptions did not shield the conduct |
Model weights | Storing protected material in model weights was treated as reproduction |
Remedies | Suno must cease certain uses in Germany, disclose related revenue, and pay damages |
Suno publicly disagreed with the ruling and said it would evaluate an appeal.
GEMA’s strongest evidence was in the outputs
Many AI copyright cases focus on the training stage. That means arguments often center on whether scraping or ingesting copyrighted material was lawful.
GEMA took a more direct route. It focused on what Suno’s models produced.
The organization showed that Suno’s systems could generate music that closely matched protected works when given certain prompts. The court accepted that the outputs were not merely inspired by broad musical styles. They reflected specific protected elements.
GEMA played AI-generated music alongside well-known songs, including 1980s synth-pop examples such as Alphaville’s “Forever Young” and Lou Bega’s “Mambo No. 5.” The point was to demonstrate market harm in a way that judges could hear, not just read about in technical filings.
That evidence appears to have made a difference. By showing resemblance across melody, harmony, and rhythm, GEMA moved the case beyond a general complaint about AI training. It placed the alleged harm in the market for music itself.
The ruling signals that AI music cases may turn less on abstract training theory and more on whether a model can reproduce recognizable protected expression.
The concept of memorization has become a central issue in generative AI lawsuits. If a system produces only broad stylistic references, defendants may argue that it has learned general patterns. If it produces work that matches protected expression, rightsholders can argue that the model has stored and reproduced protected material.
The Munich court’s ruling gives weight to the second argument.
Why fair use and text and data mining arguments failed
Suno’s defense relied in part on the idea that AI training should be treated as a lawful technical process. It pointed to U.S. fair use and to EU rules that allow certain forms of text and data mining.
The Munich court did not accept that framing.
The court’s rejection of U.S. fair use was expected in one sense. Fair use is a U.S. doctrine, and German courts do not simply apply it as a blanket defense. Still, the court’s treatment matters because Suno is a U.S.-based company and has argued that its training activity took place in the United States.
The court also rejected the EU text and data mining defense. That exception can allow certain automated analysis of data and protected works, but it is not unlimited. In this case, the court found that Suno’s conduct went beyond what the exception permits.
The most important legal point was the court’s view of model weights.
AI companies often describe training as a process that extracts patterns from data rather than stores copies of works. Rightsholders argue that if those patterns allow the system to recreate protected expression, the model contains something legally meaningful.
The court sided with GEMA on that issue. It ruled that storing copyrighted tracks in model weights constitutes reproduction.
That finding could affect how AI companies explain their systems in future cases. A company may not be able to avoid copyright liability simply by saying the original files are no longer present in a database. If the model can reproduce protected works, courts may look at the system as a new place where protected expression has been stored.
Germany had jurisdiction despite U.S. training claims
Suno also challenged the German court’s authority to hear the case. The company argued that the training occurred in the United States, outside Germany.
The court rejected that argument.
According to the ruling described by GEMA, Suno hosted its models on German servers and made infringing outputs publicly available in Germany. That was enough for the Munich court to claim jurisdiction over the German part of the dispute.
This part of the decision may prove important for other AI companies that operate across borders. Training, hosting, user access, and output generation often happen in different places. That makes jurisdiction a key issue in AI litigation.
The court’s approach suggests that companies cannot rely only on the location of training to avoid legal claims in other countries. If the model runs on local infrastructure or produces infringing outputs for local users, national courts may find a legal connection.
That matters for a global service. A U.S.-based AI company can face European copyright claims if its tools are available in Europe and if rightsholders can show harm there.
The remedies could hit Suno’s product and its finances
The ruling gives GEMA several forms of relief.
Suno must stop unauthorized reproductions and public distribution of GEMA’s repertoire within Germany. That could require technical changes to how the service operates there, depending on how Suno responds and whether the ruling stands after any appeal.
The court also ordered Suno to disclose all revenue generated in connection with the infringements. That disclosure will help determine damages.
The damages are not yet quantified. Still, the disclosure requirement matters. It may give GEMA a clearer picture of how Suno monetized the features or outputs connected to protected works.
For AI companies, that part of the ruling is a warning. Copyright exposure may not stop at removing content or filtering outputs. Courts can require financial transparency tied to alleged infringement.
For rightsholders, it offers a route to calculate harm. If a company earned revenue from subscriptions, access fees, or other product uses connected to infringing outputs, those numbers could shape damages talks or later court orders.
The ruling raises pressure on U.S. lawsuits against Suno and Udio
The Munich decision comes as major record labels pursue parallel cases in the United States against Suno and Udio. Those lawsuits include claims from major industry players such as Universal Music Group and Sony Music.
The German ruling does not decide the U.S. cases. U.S. courts will apply U.S. copyright law, including fair use, under their own standards.
Still, the decision adds pressure.
The reason is practical as much as legal. A European court has now accepted the argument that an AI music model can memorize protected songs and generate outputs that harm the market for those works. That finding gives rightsholders a persuasive example to cite in public debate and, where allowed, in related legal arguments.
The U.S. cases will likely pay close attention to several issues that also mattered in Munich:
Whether training on copyrighted music requires a license
Whether model outputs can reproduce protected expression
Whether the model stores protected works in a legally relevant form
Whether AI-generated substitutes cause market harm
Whether licensing markets for AI training already exist or should exist
The fair use question will be especially important in the United States. U.S. courts weigh factors such as purpose, amount used, and market effect. GEMA’s strategy in Munich, which emphasized output similarity and market harm, speaks directly to one of those concerns.
If U.S. plaintiffs can show that an AI music model produces close copies or substitutes for protected songs, the fair use defense becomes harder. That does not guarantee the same result, but it changes the pressure around settlement, licensing, and product design.
GEMA wants licensing to become the standard model
GEMA has pushed for a licensed dataset approach for AI music platforms. Its position is that AI companies should negotiate with rightsholders before using protected catalogs for training.
That approach would create a paid pathway for AI development rather than a fight over unlicensed scraping. It could also give creators and publishers more control over how their works feed into generative systems.
For AI companies, licensing raises cost and complexity. Music rights are layered. Compositions, lyrics, sound recordings, performance rights, and publishing rights can involve different owners. A license for one use may not cover another.
Yet the Munich ruling gives licensing advocates a stronger hand. If courts treat unlicensed AI training as infringement, companies may decide that licensing is safer than litigating market by market.
The decision may also encourage more technical safeguards. AI companies could invest in stronger filters, prompt controls, content matching, and audit systems to reduce the risk of generating protected material. Those tools would not solve the training question by themselves, but they could become part of compliance plans.
What happens next
Suno has said it disagrees with the verdict and plans to evaluate an appeal. If it appeals, a higher court could revisit key findings, including the treatment of model weights, the rejection of exceptions, and the scope of German jurisdiction.
For now, the ruling stands as a major win for GEMA and a warning to AI music platforms operating in Europe.
The case also gives the wider music industry a clearer legal strategy. GEMA did not rely only on the claim that copyrighted songs had been scraped for training. It showed outputs that, according to the court, matched protected works in specific musical ways. That made the harm easier to understand and harder to dismiss as theoretical.
The broader takeaway is simple: AI music tools are entering a legal system built to protect human-created works. Courts are now starting to decide where learning ends and copying begins.
After the Munich ruling, companies that build music models without licenses face a much harder question. If a model can reproduce the songs it trained on, courts may treat the technology not as a neutral tool, but as a system that stores and distributes protected music.
Comments