Germany Ruled Against Suno. What It Means for the American Cases.

This article explains a major European court ruling against the AI music generator Suno, how its reasoning compares to United States law, and what it may signal for the American cases now moving toward decision. It is informational only. It is not legal advice, and it creates no attorney client relationship.

What the German court actually decided

On July 31, 2026, the Regional Court of Munich I, the Landgericht München I, ruled in case 42 O 763/25 that Suno infringed copyrights administered by GEMA, the German collecting society, by training its models on protected songs and by storing and reproducing them. According to Variety and analysis from the law firm Bristows, the court found that six well known works, among them Atemlos durch die Nacht, Daddy Cool, Rasputin, Big in Japan, Forever Young, and the refrain of Mambo No. 5, were memorized inside Suno's version 3.5 and version 4 models and could be pulled back out as recognizable reproductions. In plain terms, the machine had not merely studied these songs. It had swallowed them whole and could cough them back up on request.

Why Suno's main defense failed

Europe, unlike the United States, has a specific statutory carve out for this kind of activity. Section 44b of the German Copyright Act, which implements Article 4 of the European Union's 2019 Digital Single Market Directive, permits text and data mining, the automated analysis of large bodies of work, for commercial purposes, but only where the rights holder has not opted out and only where the miner had lawful access to the material to begin with. Suno leaned on that exception. The court agreed it can apply to generative AI training in principle, then found it did not apply here, for two reasons worth understanding.

First, the court drew a line between analyzing a dataset and retaining the works inside it. Memorizing entire compositions in the model's parameters, it reasoned, goes well beyond the temporary analysis the exception was built to allow. Second, and more damaging, the court found Suno never had lawful access at all. The evidence showed Suno pulled its training audio by stream ripping from YouTube, defeating the rolling cipher that YouTube uses to prevent downloading. Under German law that cipher is a protected technical measure, and breaking it is itself unlawful. A defense that depends on lawful access collapses the instant the access turns out to be theft.

The part that should get American lawyers' attention

The most striking move in the ruling was jurisdictional. Suno's training happened in the United States, on American servers, and Suno argued that German courts had no business judging it. The Munich court disagreed. Applying the territoriality principle, it held that a company serving AI generated music to European users answers to European law, and it reached the American conduct anyway. Then it went one step further. It analyzed the training under United States copyright law, the very fair use test Suno would raise at home, and concluded that even under Section 107 of the U.S. Copyright Act the use did not qualify, pointing to the substantial similarity of the outputs and the way they substitute for the originals in the market.

The remedies matched the findings. The court ordered Suno to stop training on and reproducing the works, to disclose what it had earned, and set penalties of up to 250,000 euros for each future violation, with damages to be calculated later. Two cautions belong here. This is a first instance decision, and it can be appealed, so it is a powerful signal rather than a settled continental rule. And it rests heavily on European machinery, the opt out and the technical measure, that has no exact twin in American law.

Would an American court reach the same result

Here the honest answer is that the doorway is different, even if the room on the other side might look similar. The United States has no text and data mining exception to fail. Instead the whole fight runs through fair use, the four factor balancing test in Section 107. American courts have split in the early rounds. In the music cases themselves the question is still open, but in adjacent disputes the signals point both ways. In Thomson Reuters Enterprise Centre GmbH v. Ross Intelligence Inc., 765 F. Supp. 3d 382 (D. Del. 2025), a Delaware federal court held that copying protected material to train a tool was not fair use, leaning on market harm. In the Anthropic books litigation, Bartz v. Anthropic, a California federal court found that training on lawfully purchased copies could be transformative fair use, while separately holding the company liable for building a library out of pirated copies, a dispute that later settled for a reported 1.5 billion dollars.

Read against that backdrop, two of the German findings travel well across the Atlantic. Verbatim, memorized outputs are poison for a fair use defense anywhere, because a model that reproduces a recognizable song is both non transformative and a direct market substitute, which turns the first and fourth fair use factors hard against the defendant. And obtaining a training corpus through stream ripping or piracy echoes the exact conduct that sank Anthropic on its pirated library, and in the United States defeating a technical protection can trigger separate liability under the Digital Millennium Copyright Act. What does not cross the ocean is the automatic quality of the German result. An American court could still find the training step itself transformative and lawful while holding a company liable only for specific infringing outputs, or only for how it acquired the data. The likeliest American outcome is therefore not a clean win for either side but a split verdict, and the pieces most dangerous to the AI companies are the ones the German court already spotlighted.

Which brings us to Sony and Udio

The American reckoning is arriving on its own schedule. The major labels first sued Suno and Udio together in June 2024. Since then the wall has cracked. Universal Music Group settled with Udio in October 2025, and Warner Music Group settled with Suno in late November 2025, both trading their lawsuits for licensing deals that require the AI companies to rebuild their models on cleared data. That leaves Sony Music standing alone, and Sony has shown no appetite for a quiet deal.

In fact Sony escalated. As Billboard and Variety reported, on July 20, 2026 Sony filed a fresh suit against Udio naming 30,117 sound recordings it says were copied to train the model without a license, a filing that pushes the potential damages from tens of millions toward as much as 4.5 billion dollars. Sony's case against Suno continues in the federal court in Massachusetts, and its case against Udio in the Southern District of New York. The strategy is not hard to read. A private settlement pays Sony alone. A court ruling that training on unlicensed recordings is infringement would force every AI music company on earth to license before it trains. Sony is spending legal fees now to try to set the price of the whole market later.

Where this leaves things

The German ruling is the first time a serious court in Europe has told an AI music company, in detail, that its training broke the law, and it did so while reaching across the ocean to judge American conduct under American law. It is not final, and it binds no United States court. But it lands at the exact moment the American cases are ripening, and it hands the labels a persuasive template built on the two facts that hurt the AI defendants most, memorized outputs and stolen access. Whether an American judge follows that logic through the different machinery of fair use is the question that Sony, Suno, and Udio are now spending fortunes to answer.

As always, this is general information and not legal advice, and this area of law is moving quickly and across several countries at once. For any specific situation, consult a licensed attorney.