# German Court Rules Suno's AI Music Models Illegally Memorized Songs

Page: https://hirantha.me/blog/german-court-rules-sunos-ai-music-models-illegally-memorized-songs
Published: 2026-08-03
Written by: hirantha.me Newsroom (AI-generated). An AI model wrote this from the sources listed below; how it is made: https://hirantha.me/about/newsroom
Topics: suno, copyright, generative-ai, music-industry, germany

> A Munich court found Suno's models stored and reproduced copyrighted songs from training, rejecting the AI firm's fair-use defense.

## The ruling

A German court has told an artificial intelligence (AI) music company, for the first time in Europe, that training a model on copyrighted recordings without a license is copyright infringement, not fair use. On July 31, 2026, the Munich Regional Court's 42nd Civil Chamber, presided over by Judge Elke Schwager, ruled against Suno in a lawsuit brought by GEMA, roughly translated from German as the Society for Musical Performance and Mechanical Reproduction Rights, Germany's music rights collecting society, which functions much like the American Society of Composers, Authors and Publishers (ASCAP) or Broadcast Music, Inc. (BMI) in the United States, representing composers, lyricists and publishers.

The court ordered Suno to stop reproducing and training on GEMA's repertoire, and to disclose the revenue it generated from the infringing use so damages can be calculated in a follow-up proceeding. No dollar or euro figure has been set yet.

## Why the court sided with GEMA

The case, filed by GEMA in January 2025 and heard in March 2026, turned on a specific technical finding: that Suno's models, versions 3.5 and 4, did not merely learn statistical patterns from the songs they trained on but actually retained and could reproduce them. According to the court's own account of the case, GEMA showed that Suno had obtained the training audio through stream-ripping, extracting recordings directly from YouTube by circumventing the platform's technical protection measures, without ever licensing the underlying rights.

Judge Schwager's written reasoning, as quoted in coverage of the decision, held that the disputed songs "are reproducibly contained in the defendant's models" and that "randomness as the cause of their playback can be ruled out" given how consistently the outputs matched the originals when users prompted the system with matching lyrics, titles and style descriptions. On that basis, the court found both an unauthorized reproduction and an unlawful communication to the public under German copyright law, covering three separate acts: the extraction and reproduction that happened during training in the United States, the copies retained inside the model's weights on servers located in Germany, and the reproductions generated in the outputs themselves.

Six compositions anchored the case: "Rasputin" and "Daddy Cool" by Boney M., "Forever Young" and "Big in Japan" by Alphaville, Helene Fischer's "Atemlos durch die Nacht," and Lou Bega's "Mambo No. 5."

## Why "fair use" didn't save Suno

Suno, like most US-based AI developers sued over training data, argued that ingesting copyrighted material to train a model falls under fair use. Munich's court rejected that defense for a reason worth noting: it distinguished this case from earlier US rulings in disputes brought by book authors, where courts found in the AI companies' favor partly because the training data was not reproduced verbatim in what the models actually output. Here, the court found the opposite: the songs came back out, largely intact, when prompted. That combination, memorization plus reproducible playback, is what the court said took Suno's training outside any fair-use-style exception under German law, regardless of where the training itself took place.

That reasoning matters beyond this one case. It signals that German courts intend to independently scrutinize whether US fair-use arguments actually hold up on the facts, rather than deferring to them simply because the training occurred on US servers, so long as the resulting model or its outputs reach German users. Legal commentary following the ruling summarized the implication bluntly: offshore training is not a safe harbor if the output lands in a jurisdiction with stricter reproduction rights.

## Not the first, and not final

This is GEMA's second win of this kind at the same Munich court. In November 2025, the same court ruled against OpenAI in a separate GEMA suit over ChatGPT reproducing song lyrics. That case addressed text, not audio, and is itself still under appellate review. The Suno decision extends the same logic to musical recordings and adds the training-location argument that wasn't as central to the OpenAI case.

The judgment against Suno is not final. Suno said it disagrees with the ruling and is "evaluating all available options, including an appeal," which would go to the Munich Higher Regional Court, the same appellate court still weighing the OpenAI case. Until any appeal is resolved, the injunction and disclosure order stand under German law.

## What it means in practice

For AI companies training generative models on copyrighted audio, video or text, the practical takeaway is that where a model demonstrably memorizes and can reproduce specific protected works, courts in at least one major European jurisdiction are willing to find infringement regardless of where the training happened, and regardless of a fair-use defense that might succeed in a US court. For rights holders and collecting societies elsewhere in Europe, the ruling hands them a second precedent to point to when pursuing similar claims. For Suno specifically, the immediate effect is an injunction against using GEMA's catalog and a disclosure obligation, with the size of the eventual damages bill still to be determined and an appeal still possible.

## Questions

**What did the Munich court actually rule?**
The Munich Regional Court found that Suno's AI models memorized and could reproduce copyrighted recordings from GEMA's catalogue, both during training in the US and in outputs generated in Germany, and ordered Suno to stop and disclose infringement-related revenue.

**Who is GEMA?**
GEMA is Germany's music rights collecting society, similar to ASCAP or BMI in the United States, representing composers, lyricists and publishers.

**Does this mean Suno has to pay damages now?**
Not yet. The court ordered Suno to disclose revenue tied to the infringing use so damages can be calculated in a later proceeding; the amount has not been set.

**Is this ruling final?**
No. Suno has said it disagrees with the decision and is evaluating an appeal to the Munich Higher Regional Court, where a related GEMA ruling against OpenAI is also under review.

**How is this different from the US book-author AI copyright cases?**
US rulings in author cases turned partly on training data not being reproduced verbatim in outputs; the Munich court found the opposite for Suno, that songs were reproducibly stored in the model and could be played back, which it said ruled out a fair-use defense.

## Sources

- [German collecting society GEMA wins its copyright-infringement lawsuit against Suno](https://musically.com/2026/07/31/german-collecting-society-gema-wins-its-copyright-infringement-lawsuit-against-suno/), Music Ally
- [GEMA notches a second transatlantic AI copyright win in Germany](https://www.reedsmith.com/our-insights/blogs/viewpoints/102nfis/gema-notches-a-second-transatlantic-ai-copyright-win-in-germany/), Reed Smith
- [Munich Regional Court stops Suno using GEMA-protected music](https://www.juve-patent.com/cases/munich-regional-court-stops-suno-using-gema-protected-music/), JUVE Patent