A German Court Rules Against Suno Over Using Protected Songs to Train AI

A German court has ruled against AI music generator Suno in a case testing whether protected songs can be used and reproduced through a generative system without permission. The judgment is important because it addresses both training and recognizable output, but it applies to six works in one case and remains open to appeal.

KEY NUMBER
Six songs were tested in the case, including “Atemlos,” “Mambo No. 5,” “Big in Japan” and “Forever Young.”

What the Munich court decided

On 31 July 2026, the court largely granted GEMA’s claims and ordered Suno not to reproduce, or cause the reproduction of, the specified musical works without permission. According to Reuters reporting on the judgment, Suno must also disclose information about revenue linked to unauthorized use, while the amount of any damages has not yet been fixed.

The dispute concerned “Atemlos,” “Daddy Cool,” “Mambo No. 5,” “Big in Japan,” “Forever Young” and “Rasputin.” The Munich court’s case notice identifies the proceeding as 42 O 763/25 and describes GEMA’s allegation as unauthorized use of protected musical works.

A German news report based on the ruling states that the restriction also covers reproductions used for AI training. The judgment is not legally final, so Suno can challenge the reasoning and scope through appeal.

Why generated output became central

GEMA did not rely only on a claim that songs entered a training dataset. It prompted Suno’s system and presented outputs that it said were recognizably close to the six protected compositions. Its argument was that the model had memorized protected expression rather than merely learning abstract musical patterns.

At the March hearing, Suno disputed that interpretation. Reporting on the proceeding says the company challenged whether the outputs were sufficiently recognizable, whether protected songs were stored in the model, whether users rather than Suno should bear responsibility, and whether a Munich court had jurisdiction when training occurred in the United States.

This distinction matters. An AI model can learn common traits such as rhythm, instrumentation or genre without necessarily copying a protected work. The legal concern becomes sharper when a simple prompt produces a melody or structure that a court considers a reproduction of identifiable expression.

Why this matters beyond six songs

The immediate order concerns a small, defined set of works, yet the reasoning could influence how AI music companies operating in Europe assess training data, model memorization and output filters. GEMA represents more than 100,000 members and rights from over two million creators worldwide, according to its official account of the March hearing.

NewTqnia’s reading is that the most consequential part is not a blanket declaration that training on any protected music is always illegal. It is the court’s apparent willingness to connect recognizable output with reproduction inside the system. If that reasoning survives appeal, developers may need stronger licensed datasets, anti-memorization testing and records showing where training material came from.

The commercial direction is already mixed rather than purely confrontational. Some music companies have settled disputes and pursued licensed AI products, while GEMA itself has launched a rights-cleared training dataset. The industry may be moving toward negotiated access, but the price, scope and attribution rules remain unsettled.

What happens next

Suno must decide whether to appeal and how to respond operationally in Germany. Rights holders will study the written reasoning, particularly its treatment of model memorization, training copies and responsibility for generated outputs.

For musicians and listeners, the ruling does not prohibit AI-generated music. It raises a more specific requirement: a service cannot assume that transforming music into model parameters makes recognizable protected expression legally disappear. The durable question is now whether licensing and technical safeguards can let AI music develop without treating the training catalogue as free raw material.

Further factual confirmation comes from dpa’s report on the non-final ruling.