Suno Ruling: What the GEMA Decision Means for AI Music in Advertising

Court

Landgericht München I (Munich Regional Court I)

Date

July 31, 2026, Case No. 42 O 763/25

Subject matter

Six compositions from the GEMA repertoire, Suno models v3.5 and v4

Core issue

Training, storage and output infringe copyright; claim largely upheld

Status

Not final; appeal under review (as of October 2026)

What did the court decide?

On July 31, 2026 the Landgericht München I largely upheld GEMA's claim: injunction, disclosure of revenues and damages. The court objected to the training, storage and output of six works from the GEMA repertoire in the Suno models examined.

Suno operates a service that generates complete pieces of music from text prompts. GEMA had accused the company of using countless protected works without a licence to train its models. The Landgericht München I upheld the claim on July 31, 2026 (Case No. 42 O 763/25).

The court examined models v3.5 and v4 and objected to three steps: training on the protected compositions in the United States, storing the works in reproducible form in models hosted on servers in Germany, and making them available through the generated outputs. Copyright infringement was claimed under US and German law.

The proceedings concerned six compositions from the GEMA repertoire: "Atemlos durch die Nacht", "Daddy Cool", "Rasputin", "Big in Japan", "Forever Young" and the chorus of "Mambo No. 5". The case was about the protected musical works, not about the performers' recordings, their neighbouring rights or the lyrics; lyrics were not part of the proceedings.

How did GEMA prove it?

GEMA gave Suno only the title, original lyrics and style of the six works, no melody. The generated results still resembled the originals so closely that the court assumed protected components had been absorbed into the model.

For the proceedings the collecting society generated new versions of the six titles using Suno. The prompts contained original lyrics, title and style, but no instructions regarding melody, harmony or arrangement. The results nevertheless clearly resembled the works.

The court saw this as evidence that protected components of the works had been absorbed into the model during training and stored there in reproducible form. For the decision, the trained model itself mattered, not only the user interface.

According to the court's announcement, Suno used stream ripping to extract the works from YouTube, circumventing the Rolling Cipher, a technical protection measure intended to prevent downloads.

Why does the text and data mining exception not apply?

The court rejected the exception under § 44b UrhG because the model does not merely analyse works but stores them in reproducible form. It also dismissed the US fair use defence.

Suno had relied, among other things, on the text and data mining exception: § 44b UrhG permits reproductions of lawfully accessible works for analysis under certain conditions. The reproducible storage of the works found here was not covered by this exception, the court held.

The argument that US law permits the training as fair use also failed. The court emphasised that the protected works are recognisable in the generated outputs.

The legal bases of the decision are reproduction under § 16 UrhG and communication to the public under § 15 (2) UrhG. The court based its international jurisdiction, including over the US law claims, on § 131 (1) and (2) VGG.

What does this mean for AI music in advertising?

The ruling targets the provider Suno, not advertisers. For campaigns what matters is which rights exist in the specific piece of music and which uses are permitted; the ruling does not decide on users' liability in general.

For agencies and brands the ruling mainly means this: the origin of training data is no longer a niche topic but a liability risk. Anyone using music from a generator in a campaign whose model may have been trained without a licence is moving in a grey zone that courts are now beginning to map.

In practice, three steps are advisable. First: check the provider's training data basis and model version; a contractual indemnity does not replace a rights clearance, it only shifts the financial risk. Second: document provider, model version, creation date and licence terms for every campaign track and have similarity to well-known works checked. Third: settle usage rights, channels, territories and duration contractually before the campaign goes on air.

For client projects we document the music source and the agreed usage rights before production starts. An overview of current models and their licensing situation is in our article on AI music models.

Is everything settled now?

No. The ruling is not final; Suno is considering an appeal. The amount of damages has not yet been decided.

Suno announced on the day of the verdict that it would examine all options, including an appeal. The company considers the ruling wrong and says its models are trained to create new songs, not to reproduce existing ones. Whether an appeal has since been filed cannot be determined from the sources available (as of October 2026).

Independently of the proceedings, the offering has moved on. Warner Music ended its own dispute with Suno on 25 November 2025 and signed a licensing agreement. On 9 September 2026 Suno introduced model generation v6, according to its own statements developed with Warner Music Group, BMG and Believe. A vendor's statement about its training basis does not replace checking the specific campaign use. Universal's and Sony's lawsuits continue, according to Tagesschau of 31 July 2026; talks with the major labels were reported as blocked.

In November 2025 GEMA had already largely succeeded against OpenAI before the same court. For the industry the question remains whether the Suno decision will stand. Until then the rule is: clarify usage rights before publication.

Frequently asked questions

Can I keep using Suno songs I have already generated?

The ruling does not answer this question categorically. What matters is the specific track, possible third-party rights and the provider's terms. For running campaigns we recommend having the music legally reviewed and keeping alternatives ready.

Does the ruling apply to all AI music services?

It concerns Suno and the six named works. The legal situation for other providers depends on their training data and contracts. The fundamental question of whether training without a licence is permissible remains unsettled.

Which costs and rights should I clarify before a campaign starts?

Provider tariffs differ by model version, scope of use, channels, territories and duration. Clarify before production whether the tariff covers advertising, and have the usage rights put in writing. Licensing agreements such as the one between Warner and Suno show that the industry is betting on contracts.

Is the ruling relevant outside Germany?

The ruling does not bind other courts generally; its res judicata effect is between the parties. Its reasoning may however influence other proceedings, since the court also examined US law.

This article is not legal advice. For binding guidance consult a lawyer specialising in copyright and media law.

Sources: Landgericht München I (official court announcement, July 31, 2026) · GEMA (press release by the party GEMA, July 31, 2026) · Tagesschau (July 31, 2026) · LTO (July 31, 2026) · Suno (v6 announcement, September 9, 2026) · Warner Music Group (November 25, 2025)