Music generators like Suno can create complete songs in just a few seconds. But are AI providers allowed to use copyrighted music to train their systems without permission? On July 31, 2026, the Munich I Regional Court issued a landmark ruling on this matter (Case No.: 42 O 763/25), which has significant implications for creators, platform operators, and companies.
GEMA Wins Case Against AI Music Service Suno
In the case of GEMA v. Suno, the Munich I Regional Court ruled largely in favor of the collecting society. Suno is prohibited from reproducing certain copyrighted musical works without permission. According to the ruling, this prohibition also applies to the use of the works for training the AI model. However, the judgment is not yet final.
The subject of the proceedings was six well-known compositions, including “Atemlos,” “Daddy Cool,” “Rasputin,” “Big in Japan,” “Forever Young,” and “Mambo No. 5.” The focus was not on the specific original recordings or the rights of the performers, but rather on the copyright-protected melodies and compositions.
Why AI spending was problematic
GEMA had asked Suno to generate new songs based on the lyrics, title, and a specified musical style. The prompts did not include any specific details regarding melody, harmony, or arrangement. Nevertheless, the generated musical pieces clearly resembled the original works.
The court viewed this as strong evidence that the works had not merely been analyzed abstractly during training, but had been “memorized” by the AI model in such a way that they could later be reproduced. This was established for model versions v3.5 and v4. The court deemed the reproducible incorporation into the model and the recognizable reproduction of protected elements in the AI output to be reproductions relevant under copyright law.
Text and Data Mining Is Not a Free Pass
Among other things, Suno invoked the statutory exception for text and data mining. Section 44b of the German Copyright Act (UrhG) permits, under certain conditions, the automated analysis of works that are lawfully accessible, for example, to identify patterns, trends, and correlations. However, rights holders may reserve the right to prohibit such use.
However, in the opinion of the Regional Court, the Suno case was no longer merely about such an analysis. If protected works or clearly identifiable components are reproducibly included in the AI model and are later output again, the provider cannot automatically invoke the exception for text and data mining.
Fair use didn’t help Suno either
Since the AI model had been trained in the United States, the court also had to address the fair use doctrine there. This doctrine may permit the use of copyrighted works in certain cases even without permission. According to the ruling, however, it did not justify the use in this instance. A key factor was that the copyrighted compositions were recognizable in the generated musical pieces.
What does the ruling mean for businesses and creative professionals?
The ruling initially strengthens the position of composers and music publishers. AI providers cannot use copyrighted musical works for free simply because they are being used for technical training. Rather, licenses and appropriate compensation for rights holders may be required. This is the first ruling to establish such a licensing requirement for training audio AI.
Users of AI music tools should also carefully review the results before publishing them. Paying for access to a platform does not automatically mean that the generated song is free of third-party rights. Particularly in advertising, corporate videos, podcasts, or social media campaigns, recognizable similarities to existing works can give rise to legal risks. To learn about other obligations that apply when using AI in communications, read our article on AI in advertising and marketing.
An Important Signal for How We Approach AI Music
The ruling shows that copyright law must also be observed when using generative AI. New creative tools remain possible, but they must take into account the rights of those whose music serves as the basis for the systems. Our article on the AI editing of photos before the Frankfurt Regional Court illustrates how courts typically assess the scope of protection in cases involving AI-generated works. Since the ruling is not yet final, it remains to be seen whether and how a higher court will rule on the issues of AI training and memorization.
In our podcast episode “AI and Music Production,” we discuss how AI is changing music production in practical terms and what legal issues arise in the process.
What is Suno?
Suno is an AI music generator. Users can create complete musical pieces—including melody, vocals, harmony, and arrangement—by entering text.
Did the court ban AI-generated music in general?
No. The ruling does not ban AI-generated music or the use of music generators in general. It concerns the unlicensed use of certain copyrighted compositions during training and in the resulting output.
Can copyrighted songs be used for AI training?
There is no general answer to this question. Key factors include the source of the training data, any restrictions on use, existing licenses, and whether copyrighted works are reproduced in the model.
Can users of AI-generated music also infringe on copyrights?
Yes. If an AI-generated song contains recognizable, copyrighted elements of an existing work and is published or used commercially, copyright claims may generally arise.
Is the judgment already final?
No. The judgment of the Munich I Regional Court dated July 31, 2026 (Case No.: 42 O 763/25) is not yet final.
Are you a law firm outside Germany? We act as German and European counsel for foreign firms in trade mark, copyright and unfair competition matters, and the client relationship stays on your side. How we work with foreign law firms.
