Munich Court Halts AI Music Training: GEMA Wins Copyright Infringement Case Against Suno
Munich, August 10, 2026 – A Munich court has ruled that the AI music generator Suno infringed on copyrighted music by reproducing protected works during its AI development. The decision, announced today, states that Suno recognized copyrighted material within unlicensed catalogs, affecting both the training data and the output of its models, even from simple prompts. This landmark ruling places significant pressure on providers of AI music models in Europe to secure contractual licenses for their catalogs.
GEMA’s Claims Upheld: Infringement on Specific Works
The 42nd Civil Chamber of the Munich I Regional Court largely granted the applications filed by GEMA, the German collecting society for musical performing and mechanical reproduction rights. These applications sought injunctive relief, disclosure, and claims for damages. The court accused Suno of infringing on rights to musical works, citing that the model utilized protected songs during its training and produced outputs containing recognizable elements from these works.
The court’s argumentation moved beyond the traditional defense of AI models merely learning statistical patterns. According to court documents, the training dataset contained six specific works from GEMA’s repertoire, which were reproducible in Suno models of generations v3.5 and v4. This shifts the legal assessment from a general sense of sound to a verifiable connection between the original and the output. The court also narrowly defined the distinction from a random hit, stating that due to the complexity and length of the affected musical pieces, it was impossible for matching excerpts to arise purely by chance from the model’s selection process.
Identified Copyrighted Works and Stream-Ripping Techniques
The court identified well-known titles as comparative anchors. These included “Atemlos durch die Nacht” by Kristina Bach, “Rasputin” and “Daddy Cool” by Frank Farian, Fred Jay, and George Reyam, respectively, “Big in Japan” and “Forever Young” by Marian Gold, Bernhard Lloyd, and Frank Mertens, and the refrain “Mambo No. 5 A little bit of” by David Lubega and Christian Pletschacher. Notably, the case focused on the protected musical works as a whole, rather than solely on the appropriation of lyrics. The court deemed the output not a mere newly composed paraphrase but a reproduction of recognizable elements.
Of particular concern for AI music model operators is the court’s description of how content allegedly entered the training process. The judgment indicates that Suno employed stream-ripping techniques to extract songs from YouTube, bypassing protection mechanisms such as rolling cipher handling. Furthermore, the court found that GEMA prompts served as the starting point for the outputs, containing original lyrics, desired musical style, and title, without specifying melody, harmony, rhythm, or arrangement. Despite this, original elements were recognizable in the outputs, a key point that weakens the narrative of “prompt as creative re-creation.”
Rejection of Fair Use Defense and Broader Implications for Europe
The court also rejected the defense commonly seen in US-based AI cases. Suno had argued that copyright protection did not apply in the asserted form, that recognizability was not present, and that the model stored mathematical patterns rather than training data. Additionally, Suno claimed that US training fell under Fair Use considerations, and that user interaction broke the chain of responsibility, as outputs were generated by targeted user actions. The court dismissed the “useful part” of this defense, extending the scope of protection to multiple reproduction paths: reproduction during US training, reproduction in models stored on German servers, and communication and reproduction of outputs in Germany.
This ruling highlights a crucial legal “turning point”: even if training steps are debatable in other jurisdictions, the assessment shifts when the model later provides outputs in Germany that recognizably contain protected works. The court deemed Fair Use insufficient in this context because the original works appeared in the outputs from simple, openly accessible prompts. The chamber also distinguished this case from US cases where training material was not made visible in comparably similar, directly retrievable outputs for users. Practically, this means the debate “Is training generally okay?” does not end with training but is decided at the point where something concretely extractable from a model is too close to the original.
Economic Impact and Precedent for Future Cases
The judgment holds significant market importance as it is not merely theoretical. The decision, dated August 10, 2026, follows a series of successful claims by rights holders in Munich. On November 11, 2025, the same court confirmed large parts of similar GEMA claims against two OpenAI companies in a case involving nine German song lyrics, issuing orders for injunction, disclosure, and damages in principle; an appeal is still pending. These two first-instance decisions do not resolve the entire legal question but provide a repeatable procedural logic: identify specific works, compare outputs with originals, and demonstrate memorization as a verifiable effect.
The economic context explains why this judgment cannot be easily dismissed. Suno reportedly raised over 400 million US dollars in a Series D round on June 3, 2026, valuing the company at 5.4 billion US dollars post-money, more than double its Series C valuation in November. Investors continue to bet on the ability to generate complete songs from text prompts in seconds. However, if courts in Germany deliver different outcomes for parts of the model chain, the cost logic shifts: instead of primarily funding computing power and growth, the question of licenses and catalog rights becomes a prominent operational factor. GEMA represents over 100,000 members in Germany and more than two million rights holders worldwide, increasing the negotiating leverage for European operators. While the decision can still be appealed and US proceedings continue separately, GEMA has provided a concrete blueprint in Munich that AI music model providers must take seriously, both organizationally and legally.
