A German Court Ruled Suno's AI Music Generator Violated Copyright
Munich's regional court found six GEMA-represented songs memorized inside Suno's models. Suno must stop, and must disclose the revenue tied to it.
A German court says Suno's AI models memorized real songs. Suno must now open its books.
Munich Regional Court I ruled on July 31, 2026 that Suno, the AI music generator, infringed copyright in six songs represented by GEMA, Germany's music licensing agency, and ordered Suno to disclose the revenue tied to that infringement.
What the court actually found
The six songs, including Alphaville's 'Forever Young,' Lou Bega's 'Mambo No. 5,' and Boney M's 'Rasputin,' were found reproducibly memorized inside Suno's v3.5 and v4 models, stored on servers in Germany. Under German copyright law, storing a work that way breaches the reproduction right on its own, and serving it back out breaches the making-available right. The ruling is not final. Suno disputes it and is weighing an appeal.
What Suno actually has to do
The court enjoined Suno from further unauthorized reproduction or training use of the six works, and ordered it to disclose revenue information tied to the infringement, a step German law uses to set up a later damages claim rather than award one directly. GEMA's general counsel, Kai Welp, framed the underlying position: providers of generative AI have to respect copyright and compensate creators, whether or not the system was trained outside the EU, if it is offered to EU users.
Part of a bigger pattern
The same Munich chamber ruled against OpenAI in a related GEMA suit over song lyrics in November 2025. This is a second win for the same argument, now applied to a different AI company and a different kind of output. It is also a separate matter from the Suno data breach covered on this site in June, which exposed details of Suno's training pipeline through leaked source code rather than a court finding.
Why a build studio cares
This is the same open question every AI product with training data eventually has to answer: what happens when a model does not just learn a style, but reproduces the underlying work closely enough that a court calls it memorized. Worth knowing where that line has actually been drawn in a real ruling, not just where a company's terms of service says it is.
Next step: read MusicAlly's coverage or JUVE Patent's legal analysis. If training data provenance is a live question in your own AI build, write to us at hello@gattyworks.com.