The GEMA Suno ruling, handed down by the Munich Regional Court I on 31 July 2026, holds that the AI music generator Suno infringed copyright in six well-known songs: by copying them to train its models in the United States, by memorising them in models stored in Germany and by reproducing them in outputs. The court rejected both the EU text and data mining exception and the US fair use defence. It matters to music publishers, labels, collecting societies and AI providers serving European users.
Key takeaways
- The 42nd Civil Chamber largely granted GEMA’s claims for an injunction, information and damages (case 42 O 763/25). The judgment is not final.
- The songs were memorised in Suno’s models v3.5 and v4, a reproduction not covered by the text and data mining (TDM) exception.
- Applying US law to the training carried out in the US, the court held it was not fair use, distinguishing the Bartz and Kadrey decisions.
- International jurisdiction over the US acts rested on a forum reserved for collecting societies, not on a general rule open to every rights holder.
What did the Munich court decide in GEMA v Suno?
According to the court’s press release of 31 July 2026, the case covered six compositions, including “Atemlos durch die Nacht”, “Rasputin”, “Daddy Cool” and “Forever Young”. The claim concerned the music, not the lyrics. GEMA generated the disputed outputs by typing the original lyrics, a musical style and the title into Suno’s prompt box, with no instructions on melody, harmony, rhythm or arrangement.
| Issue | Suno’s argument | What the court held |
|---|---|---|
| Memorisation | Model weights hold learned patterns, not the training data | The songs are reproducibly contained in models v3.5 and v4; chance is ruled out given their length and complexity |
| TDM exception | Any use is covered by text and data mining | Section 44b does not cover reproduction in the model |
| Outputs | Users’ iterative prompts break the chain of responsibility | Simple, open prompts; Suno chose the data and the architecture, so it is responsible |
| Training in the US | Fair use, and the German court lacks jurisdiction to decide it | Jurisdiction under section 131 of the Collecting Societies Act; US law applies and fair use fails on all factors |
The court also recorded that Suno used stream-ripping tools to extract the tracks from YouTube, circumventing a technical protection measure designed to prevent downloads.
Why did training in the US not protect Suno?
Copyright is territorial. Under the “country of protection” principle, also set out in Article 8(1) of the Rome II Regulation, the court applied US law to copies made in the US and German law to acts in Germany. Because the songs reappeared in the outputs, it held that the training copies were not fair use under 17 U.S.C. § 107. Unlike Bartz and Kadrey, where two US courts accepted fair use because training data did not substantially reach the outputs, here the outputs were substantially similar to the originals, and every factor from the Supreme Court’s Warhol decision weighed against Suno.
The jurisdiction point needs care. The chamber founded its international jurisdiction over the US acts on section 131 of the German Collecting Societies Act (VGG), which gives collecting societies such as GEMA a special forum for connected claims. A label or author suing alone would need another basis, although acts in Germany, such as models stored on German servers and outputs generated there, can be pursued under the ordinary rules.
How does the GEMA Suno ruling fit with GEMA v OpenAI?
The same chamber reached a similar result on 11 November 2025 in GEMA v OpenAI (case 42 O 14139/24). As LTO reported that day, it held that nine song lyrics memorised in ChatGPT’s models were reproduced without authorisation and that section 44b covers analysis, not the storage of whole works. OpenAI appealed to the Higher Regional Court of Munich (case 6 U 3662/25 e), according to the Hogan Lovells AI litigation tracker.
What this means for your business
- Rights holders: test whether a model reproduces your works with simple prompts, and preserve dated, reproducible evidence of prompts and outputs.
- Check the representation chain. A collecting society has procedural options that individual owners lack, so decide early who sues and where.
- Licensors in Spain, Portugal and Latin America: works exploited in Germany are protected under German law, wherever the author is based.
- AI providers: training outside the EU does not remove exposure in the EU, and the EU AI Act separately requires providers of general-purpose AI models to have a policy to comply with EU copyright law (Article 53(1)(c)).
If you need a catalogue audit or a licensing strategy for AI uses across several countries, our team for AI and digital assets IP protection can coordinate it with your music and publishing advisers.
Where rights holders and AI companies get this wrong
- Assuming one judgment is the law everywhere. This is a first-instance decision in Germany. It can be appealed, and other courts may disagree.
- Weak evidence. The court relied on comparing training data with outputs; screenshots without the exact prompt rarely suffice.
- Choosing the wrong claimant or forum. The US-acts finding depended on section 131 VGG, which only collecting societies can use.
- Ignoring how data was obtained. Circumventing technical protection measures adds facts that weigh against a defendant.
- Fragmented action. A coordinated plan across IP enforcement and litigation avoids parallel actions that contradict each other.
Frequently asked questions
Is the GEMA v Suno judgment final?
No. The Munich Regional Court I stated on 31 July 2026 that the judgment is not final, and Suno told Reuters it would review all options, including an appeal. Until a higher court rules, the decision is persuasive but binds only the parties. It is a first-instance ruling limited to six specific works.
Can I sue in Germany if an AI model was trained in the US?
Possibly, but the basis matters. GEMA relied on a special forum for collecting societies to bring claims over US training. Individual owners can usually sue in Germany for acts there, such as outputs generated in Germany or models stored on German servers, while claims over acts abroad need their own jurisdictional ground.
Can IP Global Guard handle an AI infringement matter in Germany for a Spanish or Latin American rights holder?
Yes. We assess the works, the evidence and the best forum, and we coordinate qualified German litigation counsel and other local correspondents from a single point of contact. Where your catalogue is also exploited in Spain, Portugal or Latin America, we align the strategy across those markets.
How IP Global Guard can help protect your catalogue
The Suno decision shows that where and how a model was trained is only part of the picture: what it memorises and outputs in Europe can be enforced there. IP Global Guard, the IP services line of META Channel Corporation Limited, handles copyright, licensing and enforcement for creative and technology companies across more than 25 jurisdictions in Europe, Latin America and Africa, and the wider group covers AI Act compliance.
Tell us which works or catalogues concern you and where you have seen them reproduced by an AI tool. We will review the evidence, map the forums and propose a licensing or enforcement route. Contact our AI and copyright team.
This article is general information, not legal advice, and reflects the position on the date of publication.
Sources
- Landgericht München I, Press release 16, GEMA v SUNO, 42 O 763/25 (31 July 2026)
- Reuters via The Star, German court rules AI music firm Suno broke copyright rules (31 July 2026)
- LTO, LG München I: GEMA v OpenAI, 42 O 14139/24 (11 November 2025)
- Hogan Lovells, AI litigation tracker, GEMA v OpenAI (appeal 6 U 3662/25 e)
- EUR-Lex, Regulation (EC) No 864/2007 (Rome II), Article 8 (11 July 2007)
- EU AI Act, Article 53 (Regulation (EU) 2024/1689, 12 July 2024)
