Companies·5 min read·JUVE Patent

Suno Loses to GEMA: Europe's First AI Music Ruling

The Munich Regional Court held that Suno trained on protected songs without a licence, ordering it to stop, disclose the revenue and pay damages. The reasoning is the story: the EU text-and-data-mining exception does not cover a model that memorises what it was shown.

LANDGERICHT MÜNCHEN I Case 42 O 763/25 · judgment of 31 July 2026 ORIGINAL WORK GENERATED OUTPUT MEMORISED AND REPRODUCED BITSMINDS.COM GEMA v SUNO · SIX WORKS IN SUIT
Share:

The Munich Regional Court has ruled that Suno infringed copyright by training its music model on protected works without a licence — the first decision in Europe to say so about a generative music company. The 42nd Civil Chamber, presiding judge Elke Schwager, handed down judgment in case 42 O 763/25 on 31 July, in favour of the German collecting society GEMA.

Suno must stop reproducing the works, stop using them to train the model, disclose the revenue connected to the infringement, and pay damages once that disclosure allows a figure to be calculated. The company told JUVE Patent it intends to explore all its options, including an appeal to the Munich Higher Regional Court.

The finding that matters is technical, not moral

GEMA sued over six works on behalf of their composers, including Frank Farian's "Daddy Cool" and "Rasputin," Alphaville's "Forever Young," and Lou Bega's "Mambo No. 5." In court, GEMA played the originals against Suno outputs side by side. The court found recognisable melody, harmony and rhythm carried across.

That evidence did the doctrinal work. Germany implements the EU text-and-data-mining exception, which is the legal foundation the entire training-data industry stands on: analysing lawfully accessible works to extract patterns is permitted. Suno argued exactly that, and added that no training data is stored in its parameters at all — only, in its phrasing, mathematically learnt patterns.

The court disagreed on the facts. It held that the works were not merely analysed during training but retained in the model in a reproducible form — memorisation — and that a model which can be prompted into substantially similar output is not covered by an exception written for analysis. It also declined to import US fair use, distinguishing the Bartz and Kadrey decisions on the ground that fair use does not stretch to cover simple prompts yielding near-copies.

Read carefully, that turns a question of law into a question of measurement. The TDM exception has not been narrowed in principle. What the ruling establishes is that its protection depends on an empirical property of the weights, testable by anyone with a prompt box. Every model that memorises its training set is exposed; every model that genuinely generalises is not. No AI company gets to answer that question about itself by assertion any more.

And then there is the lock

The court also found that Suno obtained the works by stream-ripping them from YouTube, circumventing what it called the rolling cipher — the technical measure YouTube uses to prevent downloading. That is a separate wrong from the copying, and it is the same structural move that decided Reddit's win over Perplexity two days ago, where the operative claim was anti-circumvention rather than infringement. Two courts on two continents in the same week, both routing around the hard copyright question by asking a much easier one: did you break the lock to get in?

Rights holders have noticed. Proving that a model memorised your catalogue is expensive and contestable. Proving that a scraper defeated a technical protection measure is a matter of logs.

Where this leaves Suno

The immediate exposure is European. JUVE reports the judgment is not yet enforceable; German procedure generally allows provisional enforcement of a first-instance judgment while an appeal is pending, typically against security, so the practical question is how quickly GEMA moves rather than whether it can wait out the appellate calendar. Suno fought the case hard — it went as far as seeking the presiding judge's removal — and was represented by Latham & Watkins, with GEMA instructing the Berlin firm Raue.

This is also the second time GEMA has won at this courthouse against an American AI company. The same court ruled in its favour against OpenAI in November 2025 over song lyrics. Munich's 42nd Civil Chamber is quietly becoming the venue where the European position on training data gets written.

The commercial context cuts against Suno's litigation posture. Both Suno and Udio have settled with Warner Music, and Udio has an arrangement with Universal — the industry has largely decided that licensing is the way this ends. Suno itself raised $400 million at a $5.4 billion valuation in June on a story about pivoting toward licensed music. A judgment ordering it to disclose infringement revenue is an awkward companion to that pitch, and the disclosure order may prove more consequential than the damages: it puts numbers on the table that every other collecting society in Europe will read.

Want AI news before everyone else?

The morning's most important AI stories, straight to your inbox. No fluff.

Related Articles