A German court found Suno liable for copyright infringement in a lawsuit filed by GEMA. What the court decided, which songs were involved, and what the ruling could mean for the AI industry.
The Munich Regional Court I has sided with the German collecting society GEMA in its dispute with the American company Suno. The court found that the developer of the music generator infringed authors’ rights both when training its AI models in the United States and when storing and reproducing protected musical works through a service available to European users.
The ruling was issued on July 31, 2026, by the specialized 42nd Civil Chamber in case No. 42 O 763/25. GEMA’s claims for an injunction, disclosure of information, and compensation were largely upheld. The exact amount of damages has not yet been determined.
What Exactly the Court Ruled
Suno is prohibited from reproducing and using without authorization six musical works that were the subject of the proceedings. The restriction applies not only to finished compositions generated by the service, but also to the storage of reproducible elements of the original works within the AI models.
The company must also provide the information required to calculate the damage caused and pay compensation. The amount will be determined later, after disclosure of revenue connected with the established infringements. The ruling is not yet final and may be appealed before a higher court. Suno has already stated that it is considering an appeal.
Importantly, the court did not merely find that individual generated outputs were too similar to the original songs. The chamber concluded that protected elements of the compositions had been reproducibly stored in Suno’s models and could later be extracted in the form of new audio recordings.
Which Songs Were Involved in the Case
The official ruling concerns six musical works:
Atemlos durch die Nacht, written by Kristina Bach; Rasputin by Frank Farian, Fred Jay, and George Reyam; Big in Japan and Forever Young by Alphaville members Marian Gold, Bernhard Lloyd, and Frank Mertens; the chorus of Mambo No. 5 (A Little Bit of...) by David Lubega and Christian Pletschacher; and Daddy Cool by Frank Farian.
Some reports about the dispute also mention Cheri Cheri Lady by Modern Talking. GEMA did indeed cite the song among its early examples of Suno’s output, but it is not included in the official list of six works covered by the July ruling.
The proceedings primarily concerned musical compositions — melody, harmony, rhythm, and structural organization, rather than rights in specific sound recordings or artists’ performances. Song lyrics were also not an independent subject of the claims, although GEMA entered the original words into prompts submitted to the generator.
How GEMA Proved the Infringement
GEMA filed the lawsuit on January 21, 2025, after Suno, according to the organization, failed to respond to a proposal to regularize its use of the repertoire through licensing.
To test the system, GEMA specialists entered the title of a work, its original lyrics, and a description of the musical style into Suno. The prompts did not contain sheet music or descriptions of the melody, harmony, rhythm, or arrangement. Despite the absence of such instructions, the generator produced compositions in which, according to the claimant and the court, recognizable elements of the original music could be identified.
The court considered the prompts sufficiently simple and open-ended. This became a crucial point: Suno failed to convince the chamber that the similarities resulted exclusively from deliberate actions by a user attempting to force the system to copy a well-known song.
According to the court materials, the parties did not dispute that the six works had been included in Suno’s training dataset. To obtain the audio material, the company used stream-ripping technology, extracting music from YouTube and bypassing Rolling Cipher, a technical mechanism used by the platform to prevent the downloading of audio and video.
The Court Found That the Works Had Been “Memorized”
The central concept in the proceedings was memorization — a situation in which a model does not merely analyze statistical patterns in training data, but retains the content of individual works in a form that allows their substantial elements to be reproduced later.
The court found that the disputed compositions were reproducibly contained in Suno’s v3.5 and v4 models, including models hosted on servers in Germany. The chamber compared works from the training dataset with generated outputs and concluded that the prolonged and structurally complex musical similarities could not be explained by coincidence.
The court classified such storage as an infringement of the reproduction right under Section 16 of the German Copyright Act. Additional infringements occurred when the model generated audio for users containing recognizable original elements and made that audio available in Germany.
Why Suno, Rather Than Its Users, Was Held Responsible
Suno argued that outputs are created at the initiative of users and that their prompts therefore determine the content of the finished music. The company also insisted that the model does not store copies of songs, but only numerical parameters reflecting general patterns, styles, and relationships between musical elements.
The court rejected these arguments. The chamber stated that Suno had independently assembled the training dataset, selected the model architecture, conducted the training process, and bore responsibility for any memorization of the data. Because GEMA’s prompts did not contain musical instructions sufficient to reconstruct the melody and arrangement independently, it was the operation of the model that determined the key content of the outputs.
The court therefore refused to treat the platform as a neutral intermediary that merely executes user commands. Under the logic of the ruling, the operator of an AI service may bear direct liability when the design of the model enables it to reproduce protected elements of works even after comparatively simple prompts.
Why the Data-Mining Exception Did Not Apply
Suno relied on European rules governing text and data mining — the automated analysis of large volumes of text and data. Under certain conditions, these provisions allow technical copies of works to be created for machine analysis.
The court concluded that the identified memorization went beyond permissible analysis. In the chamber’s view, the works had not merely been used to identify general musical patterns: their content had been transferred into the model’s parameters in a manner that allowed it to reappear in finished songs. The exception under Section 44b of the German Copyright Act therefore could not be applied.
This distinction has major implications for the entire AI industry. The court effectively drew a line between analyzing a work to obtain abstract knowledge and retaining its recognizable creative elements inside a commercial generator.
The German Court Even Assessed Training Conducted in the United States
One of Suno’s main arguments was that a German court lacked authority to assess model training conducted in the United States.
The court nevertheless accepted international jurisdiction, relying on Section 131 of Germany’s Collecting Societies Act. This provision allows interconnected claims brought by organizations such as GEMA to be examined even when part of the alleged infringement occurred outside Germany.
At the same time, the court applied US law to conduct that took place in the United States and separately considered Suno’s argument based on fair use.
The chamber concluded that training on the disputed works did not qualify as fair use because substantial elements of the original music appeared in outputs created with simple prompts. The court distinguished the Suno case from the US proceedings Bartz and Kadrey, in which material from training datasets was not reproduced or made available to users in substantial form.
The international aspect of the ruling may prove especially important. It indicates that moving model training to another jurisdiction does not necessarily shield a developer from claims brought by European rightsholders if the model or service is operated in Europe.
Suno’s Position
Suno disagreed with the court’s conclusions. A company representative stated that its models were designed from the outset to generate new songs rather than reproduce existing works, and that protective safeguards are built into the service.
The company believes that the ruling is based on an incorrect understanding of Suno’s technology, the way it is used, and US law. The developer is examining all available options, including an appeal.
It is therefore too early to claim that the dispute has been finally resolved. A higher court may reconsider questions of international jurisdiction, the technical nature of memorization, the application of the data-analysis exception, and the interpretation of the US fair-use doctrine.
Why the Ruling Matters for the Music Industry
GEMA represents more than 100,000 composers, lyricists, and music publishers in Germany and, through international agreements, the interests of more than two million rightsholders worldwide. The dispute therefore extends far beyond the claims of several well-known authors.
GEMA’s victory strengthens the negotiating position of collecting societies, music publishers, and catalog owners. It is becoming more difficult for developers of generative models to build their defense solely on the claim that a neural network “learns like a human” and therefore does not require licenses.
At the same time, the ruling does not prohibit music-generation technology itself and does not require Suno to shut down immediately. It concerns specific works, specific models, and reproduction of original elements that the court found to have been proven. As a first-instance judgment, it also does not automatically establish a binding rule for every court in the European Union. Nevertheless, other rightsholders and courts may use its reasoning as an important reference point.
Economic developments are already pushing the market toward licensing. In November 2025, Warner Music settled its own dispute with Suno and entered into a licensing agreement providing for a transition to licensed models. At the same time, other major participants in the music market continue to pursue litigation concerning the training of generative AI.
Connection to GEMA’s Previous Case Against OpenAI
The Suno ruling became GEMA’s second major victory in Munich. In November 2025, the same court system upheld the organization’s claims against OpenAI in a dispute involving protected song lyrics that could be reproduced through ChatGPT.
OpenAI appealed that ruling, so this area of case law cannot yet be considered finally settled either. The difference in the new case is that it concerns not only lyrics, but also musical composition as a complete creative work.
Conclusion
The Munich court’s ruling has become one of the most substantial judicial assessments to date of how generative music AI operates. The court examined several stages of the technological chain at once: obtaining files for training, building the training dataset, retaining works in the model parameters, and the appearance of recognizable music in generated outputs.
The main conclusion is not that every use of protected music for AI training is automatically unlawful. The court linked Suno’s liability to specific circumstances: the unlicensed acquisition of works, their established memorization, and the ability to reproduce substantial elements of the originals through comparatively simple prompts.
If the ruling survives appeal, music-AI companies will have to pay far greater attention to the origin of training data, catalog licensing, technical safeguards against memorization, and transparent revenue sharing with authors.
Minatrix.FM Editorial Opinion
It is misleading to present the conflict surrounding Suno as a battle between progress and musicians. Generative AI has already become part of music production and is unlikely to disappear. The real dispute is different: can a commercial company build a multibillion-dollar product for free on music that it neither created nor licensed?
In our view, the answer should be no. AI can become a useful tool for composers, producers, and independent artists, but a sustainable market is impossible without clear rules: rightsholder consent, transparent data provenance, protection against the copying of works, and fair remuneration.
At the same time, the court’s decision should not be turned into a ban on experimentation or an attempt to entrench the monopoly of major catalog owners. Licensing must be accessible not only to global corporations, but also to small developers, research projects, and independent services. The optimal model is not the blocking of technology, but the creation of a legal market for training data in which authors understand in advance how their music will be used and what share of revenue they will receive.
The GEMA v. Suno case shows that the period in which developers first collected content on a massive scale and postponed licensing questions until lawsuits appeared is coming to an end. The future of music AI will depend not only on the quality of generation, but also on how fairly the technology interacts with the people whose creativity it is built upon.
