The Suno lawsuits, explained: Warner, UMG, GEMA and what changes for you
One settlement, one loss in Munich, two majors still fighting. What the July 2026 GEMA ruling actually found, and why none of it touches your ownership of what you generate.
On 31 July 2026 a court in Munich ruled largely against Suno, finding that protected works were reproducibly stored inside its models and that outputs generated from simple prompts infringed. In November 2025 Warner Music had gone the other way entirely, settling its case and signing a licensing deal. Universal and Sony are still litigating in Boston. Suno raised another $400 million in June 2026 while all of this was live.
That is the shape of it: one settlement, one loss abroad, two majors still fighting, and a company that keeps growing through the middle. Here is what it means for someone whose actual question is whether their own songs are safe.
This is a summary of reported facts, not legal advice.
The short version, by party
Warner Music Group. Settled with Suno in November 2025 and struck a licensing deal at the same time, the first major label to do so. Reporting from Reuters, Music Business Worldwide and TechCrunch describes it as more than a settlement of past claims: it includes a go-forward licence for Suno to use Warner recordings as training data for new models. That combination is why the agreement became the most fought-over document in the case.
Universal Music Group and Sony Music. Still suing, in the US District Court for the District of Massachusetts. Suno's defence rests on fair use. Through spring 2026 the parties fought over discovery, with UMG and Sony trying to obtain the terms of the Warner agreement. A magistrate judge refused, and the plaintiffs filed an objection on 20 April 2026, arguing the Warner deal is a forward-looking commercial arrangement rather than a simple settlement. In July 2026 UMG and Sony moved to expand the case to roughly 61,000 recordings, against 560 works in the original complaint. Suno opposed.
GEMA, in Germany. On 31 July 2026 the Munich Regional Court I ruled largely for GEMA in case 42 O 763/25. It is the more consequential ruling of the two jurisdictions so far, and the details matter.
The American Federation of Musicians. In June 2026 the union sued UMG and Warner, arguing that the labels licensed recordings featuring their members to AI companies without compensating or informing the performers who played on them. The lawsuit targets the labels, not Suno, which tells you something about where the money questions are moving.
Why the Munich ruling is the interesting one
Most coverage of AI music litigation is about whether training on copyrighted material is allowed. The Munich court answered several harder questions at once.
It found memorisation. The court held that the works, six well-known compositions including Rasputin, Daddy Cool and Forever Young, were reproducibly contained inside versions of Suno's models. Storage itself was treated as infringing the reproduction right, and the text-and-data-mining exception was held not to cover full retention of works in model parameters.
It rejected the prompting defence. Suno argued that similarity came from GEMA's own complex, iterative prompting. The court disagreed, noting the prompts were simple and open-ended and that the architecture plus the memorised data, not the prompt, determined the output.
It ran a US fair use analysis itself. Rather than deferring, the Munich court applied 17 U.S.C. § 107 to the American training conduct and found fair use inapplicable, expressly distinguishing the Anthropic and Meta cases on the grounds that outputs there did not closely reproduce training works.
It reached across borders. The injunctions include an order to stop copying the works within the United States for the purpose of training a music AI model. Extraterritorial reach of that kind is unusual, and it is why lawyers on both sides of the Atlantic wrote it up within days.
The judgment is not final and can be appealed. Treat it as a strong signal, not a settled rule.
What none of this changes for you
Here is the part most articles bury. Every one of these cases is about training data: what went into the model. None of them is about your ownership of what comes out.
Your rights as a user come from Suno's own terms, and the rule there has not moved. Commercial rights attach to songs made while you hold an active Pro or Premier subscription. Songs made on the free plan cannot be monetised, and subscribing afterwards does not backdate a licence. That is a contract question between you and Suno, entirely separate from what UMG argues in Boston. The full commercial-use picture is the practical article.
Two risks that genuinely are yours, though, and neither depends on the litigation:
Output similarity. If a generation comes back sounding conspicuously like a specific existing record, that is your problem when you release it, regardless of who wins any case. The Munich evidence was built precisely on outputs that resembled originals from naming a title and style.
Named artists in prompts. Asking for a specific artist's voice or style by name is the single fastest route to an output that resembles a real record, and to a rights claim that has nothing to do with training data. This is the entire reason our style catalogue describes sound through instruments, era and production rather than through artist names. A prompt that says the sound instead of the person gets you the texture without the exposure.
What to actually do about it
Subscribe before you make the song that matters. Retroactive licences do not exist here, and it is the most common expensive mistake.
Do not prompt with artist names if you intend to publish. Describe the sound: instruments, decade, production character, vocal treatment. Writing a style prompt that way is more precise anyway, so you lose nothing.
Listen for resemblance before release. If a hook feels familiar, it may be. Regenerate rather than argue with a rights holder later.
Keep your own uploads clean. Suno matches uploads against known recordings, and material it recognises is rejected. That check exists because of everything described above, which is also why the workaround people ask for is not one. If your own material trips it, the fix is support and proof of rights.
Watch the licensing trend, not the headlines. Warner licensed. GEMA launched its own licensed dataset product for AI music tools shortly before the Munich ruling. Udio settled with majors at the end of 2025. The direction of travel across the industry is toward licensed training, and a licensed model is a safer model to build a habit on.
FAQ
Is Suno legal to use? Suno operates and sells subscriptions, and users hold commercial rights under its terms when subscribed. The disputes are between Suno and rights holders over training, not between you and anyone over using the product.
Did Suno lose the lawsuit? It lost in Munich, at first instance, on 31 July 2026, and that judgment can be appealed. The US case brought by UMG and Sony was still active as of summer 2026.
Who owns Suno? It is a private US company, still raising capital, including a $400 million round reported in June 2026.
Will Suno get shut down? Nothing in the current record points that way. Settlements and licensing deals, including Warner's, suggest a negotiated future rather than a switched-off one.
Can I still sell songs I made? Under Suno's terms, yes, if the song was made while subscribed and you hold all rights to everything in it, including lyrics.
Does the German ruling affect me? Only indirectly. It binds Suno, not users, and its practical effect is on what Suno can train on and offer in Germany.
Should I avoid AI music because of the lawsuits? That is a judgment call, not a legal one. The pragmatic position is to use it while keeping your outputs distinct from existing records, which is good creative practice regardless of any court.