What the GEMA v. Suno Ruling Holds for AI Music Training

Record status, last verified August 25, 2026: LG München I, 42nd Civil Chamber, file no. 42 O 763/25, presiding judge Elke Schwager, judgment dated July 31, 2026. The defendant was Suno; the claimant was GEMA; the asserted repertoire in the court’s public release consisted of six musical works. The ruling is confirmed by the official court press release, but it is a first-instance judgment and not final.[1]

The useful way to read the GEMA v. Suno AI music training copyright ruling is not as a global answer to every training dispute. It is a concrete provider-liability record. A German court treated the training location as legally reachable, found infringement through multiple acts in the product chain, and tied the result to facts Suno was reported to have conceded or failed to neutralize.

The four infringing acts the court identified

For diligence purposes, the holding is best kept in four buckets. Law-firm summaries with paragraph references to the German judgment describe the prohibited acts as covering training copies, model memorization, offering the model, and infringing outputs. Because the reasoned judgment is not independently available in English, those paragraph-level mappings should be treated as attributed summaries of the judgment rather than as direct English-language verification from the court file.[2]

That sequence is the reason the ruling is more than a routine output-similarity dispute. A provider assessing EU-facing risk now has to separate at least four questions: how training material was acquired, whether training made copies under the applicable law, whether the model retained protected works in a legally meaningful way, and whether the commercial offering or outputs created additional acts of exploitation.

The second bucket—memorization—is the one that will travel most easily into technical review. It asks for evidence about what the model can reproduce, not only what the training set once contained. In a licensing review, acquisition, or insurance underwriting file, that creates a different review task: prompt testing, output matching, model-version tracking, and server-location documentation all become part of the legal-risk record.

Why Suno’s own factual record mattered

The court’s result did not rest on a generalized assumption that AI training is always infringing. The decisive facts, as reported in analyses of the judgment, were narrower: Suno was said to have conceded that it trained on the relevant repertoire without paying, and GEMA tied the six works to a stream-ripping path from YouTube involving the Rolling Cipher, which was treated as an effective technical protection measure under Sec. 95a UrhG.[2][3]

That is the diligence point most likely to be underweighted in vendor answers. A statement that training occurred outside the EU does not answer how the files were obtained. Nor does a statement that the model is probabilistic answer whether particular works can be retrieved, reproduced, or functionally reconstructed. The Munich record makes the access path and the retention behavior legally material.

The Rolling Cipher detail is not decorative. If training material was obtained by circumventing a technical protection measure, the legal risk is no longer limited to an abstract copyright exception debate. It also becomes a lawful-access and anti-circumvention problem. For counsel reviewing a model provider, the relevant questions are therefore factual before they are doctrinal: who sourced the files, from where, under what permissions, with what scraping or ripping tools, and with what records preserved?

Prompt evidence also appears to have mattered. Source accounts differ on the exact number of prompts used to demonstrate output similarity or reproduction—Bird & Bird reports 338 total prompts, while other summaries describe the testing differently—so the safe conclusion is not a universal prompt threshold. The narrower conclusion is that the court accepted a record in which targeted prompting helped connect outputs and memorization to protected works.[2][3][4]

The jurisdictional move: Sec. 131 VGG, Schutzlandprinzip, and US fair use

The main turn in the case is jurisdictional. Section 131 of the German Collecting Societies Act gives German collecting societies a gateway for certain claims involving foreign rights or foreign uses, including provisions addressing the assertion of rights abroad by collecting societies.[5] In this case, that gateway mattered because GEMA was not only challenging German-facing outputs. It was reaching training acts that occurred in the United States.

The court then applied US law to the US training acts under the Schutzlandprinzip, as summarized by Bird & Bird and other analyses. On that route, the Munich court did not purport to apply German copyright law to every foreign training copy. It used the collecting-society gateway and conflicts analysis to ask whether the relevant US acts were lawful under US law.[2][6]

The answer, in the Munich court’s view, was no. The court rejected Suno’s fair-use defense and, according to Bird & Bird’s paragraph-level account, assessed the four fair-use factors by reference to the US Supreme Court’s Warhol framework while distinguishing Bartz v. Anthropic and Kadrey v. Meta.[2]

That point has to be kept in its procedural lane. The Munich court’s US-law analysis is not binding on the District of Massachusetts or any other US court. It is still citable, and it will be read by parties who want a European court’s treatment of AI music training under US fair use. But it is not the same thing as a US appellate holding on model training.

Remedies: injunction, disclosure, unquantified damages

The court granted injunctive relief and ordered Suno to disclose revenue information. Damages have not yet been quantified; the revenue-disclosure order is the mechanism that may allow GEMA to calculate or pursue a monetary claim later.[1][2]

One remedy deserves separate attention because it is unusual in practical effect. The injunction has been reported to include a limb requiring Suno to stop copying the works within the territory of the United States of America for the purpose of training an AI model to generate music.[2][8] That is the clearest expression of the provider-side, extraterritorial risk signal: the order did not stop at German outputs or German servers.

Revenue disclosure also explains why settlement comparisons should be handled carefully. Warner’s reported deal with Suno may matter commercially because settlement can avoid the compelled accounting that litigation may produce, but it does not tell the court what damages are owed in GEMA v. Suno. The operative fact here is simpler: the Munich court ordered disclosure, and damages remain unquantified.

What the ruling does not settle

Several boundaries should stay attached to this record. First, the fully reasoned judgment has not been published in English. The official court release verifies the existence, posture, parties, date, and broad result, while the more granular act-by-act and paragraph-numbered discussion comes from law-firm summaries and should be cited as such.[1][2]

Second, the case does not establish that every AI training copy is infringing in Europe, or that every model output creates provider liability. The ruling turned on a record that included unlicensed repertoire, alleged stream-ripping around a technical protection measure, memorization, and outputs tied to the asserted works. Remove or change those facts and the analysis may change.

Third, the EU AI Act should not be smuggled in as a copyright defense. Compliance duties for general-purpose AI models, including transparency obligations, may affect documentation and regulatory exposure, but they do not themselves answer whether a particular training act, retained work, or output infringes copyright. That point deserves its own regulation note; it should not be used to soften the holding here.

Parallel US docket to watch: UMG Recordings v. Suno

The main US watch item is UMG Recordings, Inc. v. Suno, Inc., No. 1:24-cv-11611, pending in the District of Massachusetts before Judge Indira Talwani. The case was filed on June 24, 2024, and the tracker record lists summary-judgment proceedings as pending. Reported statutory damages exposure is up to $150,000 per work, with reported combined claims north of $9 billion.[9]

The Massachusetts case is not a prediction engine for the German appeal, and the German ruling is not binding on the Massachusetts court. The useful connection is evidentiary and argumentative: both disputes force the same provider-side questions about training sources, copying, output similarity, and fair use. A future record should treat UMG v. Suno on its own docket, judge, motions, and evidence rather than folding it into the Munich result.

Verified risk assessment

GEMA v. Suno is a confirmed first-instance EU risk signal for AI music training. It makes non-EU training location a weak standalone comfort point where a provider offers a generative tool into Europe, where a collecting society can use the relevant statutory gateway, and where the factual record includes unlicensed training, disputed lawful access, memorization, and outputs linked to protected works.

It is not final appellate law. The items to track are appeal filing and forum, enforceability pending appeal, the damages phase after revenue disclosure, any publication or translation of the reasoned judgment, and the District of Massachusetts summary-judgment record in UMG v. Suno. Related future records should separately cover GEMA v. OpenAI, UMG v. Suno, AI Act Article 53 as a compliance obligation rather than a copyright defense, memorization evaluation screens, and lawful-access documentation records.

References

1. Pressemitteilung 16 vom 31.07.2026, Landgericht München I, July 31, 2026
2. Munich District Court rules on AI-generated music: GEMA v Suno, Bird & Bird
3. GEMA v Suno: Munich Court Finds AI Music Training and Outputs Infringe Copyright, Bristows
4. GEMA also wins against Suno, kpw.law
5. Collecting Societies Act, Gesetze im Internet
6. GEMA notches a second transatlantic AI copyright win in Germany, Reed Smith
7. Suno Lost To GEMA. Why The Ruling Should Worry AI Music Companies, Forbes, August 5, 2026
8. Suno infringed copyright in GEMA case, German court rules, Music Business Worldwide
9. UMG Recordings v. Suno, AI Lawsuit Tracker