![\"Copyright](\"https:\/\/www.europesays.com\/ai\/wp-content\/uploads\/2026\/08\/1785962532_778_0x0.jpg\")
<\/p>\n<\/p>\n
31 July 2026, Bavaria, Munich: Tobias Holzm\u00fcller, CEO of Gema, speaks to reporters at the state court following the announcement of the verdict in a copyright lawsuit. The German organization Gema had filed a lawsuit against the U.S. company Suno, alleging that it had infringed on its copyrights to songs. Suno offers users the ability to have songs generated by AI. Photo: Karl-Josef Hildenbrand\/dpa (Photo by Karl-Josef Hildenbrand\/picture alliance via Getty Images)<\/p>\n
dpa\/picture alliance via Getty Images<\/p>\n
The GEMA Suno ruling<\/a> landed on July 31 when a Munich court found that Suno, the AI music platform that lets anyone generate a full song from a text prompt, infringed the copyrights represented by GEMA, Germany\u2019s music collecting society, by training its models on songs like “Rasputin” by Boney M. and “Forever Young” by Alphaville and reproducing them on demand. <\/p>\n GEMA represents more than 100,000 composers, lyricists and publishers and collects royalties on their behalf, the German equivalent of ASCAP or BMI in the United States. <\/p>\n GEMA filed the suit in January 2025, after asking Suno to license its repertoire and getting no response. The case centered on six well-known compositions, including “Atemlos” (Helene Fischer), “Daddy Cool” (Boney M.), “Big in Japan” (Alphaville) and “Mambo No. 5” (Lou Bega), which GEMA\u2019s lawyers were able to reproduce almost note for note<\/a> just by prompting Suno with the lyrics, a style and a title.<\/p>\n It\u2019s GEMA\u2019s second win against an AI company in nine months, following a November 2025 ruling against OpenAI over memorized song lyrics in ChatGPT, which OpenAI has appealed.<\/p>\n Suno Already Admitted It Trained On GEMA\u2019s Music Without Paying For It<\/p>\n Much of the ruling focused on what the court called memorization \u2013 evidence that Suno\u2019s models had stored near-complete copies of specific songs during training and could reproduce them almost note for note from a simple prompt, rather than generating something genuinely new. <\/p>\n GEMA\u2019s lawyers demonstrated this by typing lyrics, a style and a title into Suno and getting back audio the court found reproducibly close to the originals. <\/p>\n But the court didn\u2019t actually need that memorization analysis to decide the case. Suno conceded, before the verdict, that it trained on GEMA\u2019s repertoire without paying for it. Its actual defense rested entirely on the argument that training doesn’t require a license. The chamber rejected that argument, and the outcome followed from there. <\/p>\n Judge Elke Schwager also noted it was undisputed that Suno obtained the training data through stream-ripping, downloading audio off the open web rather than acquiring it through any licensed channel. That admission and that finding did the real work in this case. The memorization findings, the side-by-side listening comparisons, the language about randomness being ruled out given the complexity of the works, all of that reads as corroborating detail layered onto a case an admission had already decided. <\/p>\n Why A German Court Can Rule On Training Done In The U.S.<\/p>\n The finding with the widest reach is jurisdiction. This is the first time a European court has ruled on AI training conducted entirely outside its borders. The court held that German law reaches it anyway, because the training data was stored in the model and the outputs get served to users in Germany. <\/p>\n Under Section 131 of the VGG, the German statute governing collecting societies, the court held that was enough to put a Massachusetts company\u2019s Massachusetts training runs in front of a Munich judge, and further, that the company would have owed GEMA a license even under the U.S. copyright law governing where the training physically happened. <\/p>\n If that holding survives appeal, every AI company with EU users is exposed to EU collecting-society litigation regardless of where the model was built. According to this court, the relevant question for jurisdiction is where the outputs land.<\/p>\n Is The Warner Music Deal A Useful Precedent Now That A Court Has Actually Ruled?<\/p>\n Warner Music sued Suno in 2024 along with Universal Music and Sony Music<\/a>, all three majors accusing it of training on unlicensed catalogs. In November 2025, Warner Music and Suno announced they had settled that lawsuit<\/a> and signed a licensing deal at the same time, with Suno agreeing to build new models on Warner\u2019s catalog and acquiring Warner\u2019s Songkick platform as part of the package. That deal was a settlement. <\/p>\n No judge ever ruled that Suno had infringed anything, and Suno never had to admit liability to get to a paid partnership. <\/p>\n GEMA\u2019s position is different. It has an actual judgment, from an actual court, finding actual infringement, plus a disclosure order Warner never extracted because Warner never needed one. That raises an uncomfortable question for Warner: Did it settle too early? <\/p>\n Warner negotiated its licensing fee in the dark, without a court ruling establishing that Suno had infringed anything and without any compelled look at Suno\u2019s actual revenue. GEMA now has both. If GEMA\u2019s number ends up higher per stream or per work than whatever Warner accepted, Warner priced its own catalog against a company that was never forced to open its books, and every other rights holder still negotiating with Suno should be watching what GEMA extracts before signing anything.<\/p>\n There\u2019s a harder question underneath that one. A court has now found, on the record, that Suno trained on stream-ripped audio and stored entire protected works inside its models without permission. Warner Music licensed its catalog to that company anyway, eight month before this ruling made the underlying conduct legally undeniable. <\/p>\n Whatever the commercial logic, that\u2019s a major rights holder signing a revenue-sharing deal with a company a court has since found built its product on unlicensed use of other people\u2019s work. <\/p>\n What The Disclosure Order Actually Does<\/p>\n Alongside the injunction, the Munich court ordered Suno to disclose its revenue tied to the six infringing works, meaning Suno has to hand GEMA the actual financial records connecting those songs to what Suno earned from them, before any damages figure gets calculated.<\/p>\n Suno\u2019s headline numbers are already public<\/a>: a $5.4 billion valuation off a $400 million Series D, and CEO Mikey Shulman\u2019s own disclosure of 2 million paid subscribers generating roughly $300 million in annual recurring revenue. What isn’t public is which slice of that revenue traces back to GEMA’s repertoire specifically, and that’s what the disclosure order forces into the open. <\/p>\n The number the court eventually sets will be calculated off whatever Suno now has to hand over. For scale, Universal and Sony\u2019s parallel U.S. suits are seeking damages up to $150,000 per infringed work under statutory copyright law, and some reporting on the Boston litigation puts the majors\u2019 combined claims north of $9 billion. <\/p>\n GEMA\u2019s German case works off actual revenue and license-fee equivalents rather than U.S. statutory damages, so the eventual figure will land differently, but the German court now has the financial detail to price it, which is the part of this ruling that had been missing.<\/p>\n What Comes Next<\/p>\n