A Munich court ruled Friday that the AI music company Suno infringed copyright by training its models on songs from the repertoire of GEMA, Germany’s music rights collective, and ordered the company to stop using the works. The decision, in the first European case against a generative AI music tool, is among the most consequential copyright rulings yet for the AI industry.
The Munich I Regional Court found that Suno reproduced six well-known songs, including “Atemlos,” “Daddy Cool,” “Rasputin,” “Big in Japan,” “Forever Young” and “Mambo No. 5,” in ways that went beyond permissible analysis of training data. The court said the works had been memorized, stored in the model’s parameters in reproducible form, and that Suno had obtained them by stream-ripping from YouTube, circumventing the platform’s technical protections.
The ruling applies two bodies of law. For the training that took place in the United States, the court applied U.S. copyright law directly and rejected Suno’s fair-use defense, finding that the model’s outputs reproduced recognizable elements of the original songs, a fact that distinguished the case from earlier U.S. decisions that found AI training to be fair use. For the model’s operation in Germany, the court found a reproduction and a communication to the public under German law.
The court also rejected the text-and-data-mining exception that AI companies have leaned on across Europe. That exception, the court said, covers analysis of data, not the retention of the works themselves in a model’s parameters. If memorization cannot be prevented, the ruling said, training on protected works falls outside the exception entirely, a sentence that extends well beyond music and reaches every generative model that can reproduce its training data.
The injunction carries teeth. Each future violation is subject to a fine of up to 250,000 euros, and the court ordered Suno to disclose information about its use of the works and to pay damages, with the amount to be determined. The judgment is not final; Suno said it disagrees with the ruling and is evaluating its options, including an appeal to the Munich Higher Regional Court.
The decision lands at a delicate moment for Suno. The company raised $400 million in June, and it signed a licensing partnership with Warner Music late last year while the other two major labels, Universal Music and Sony Music, continue to press a lawsuit against it in the United States. A European judgment that rejects the free use of music for training gives those plaintiffs new ammunition, even if the U.S. courts reach different conclusions.
For the AI industry, the ruling is the strongest signal yet that the training-data free lunch is ending. GEMA won a similar case against OpenAI in Munich last November over song lyrics, and the new decision extends that reasoning to music generation. The pattern emerging in European courts, that copyrighted works used to train models require licenses, is the one that rights holders have been seeking for years.
The ruling’s logic, if adopted elsewhere, would reach far beyond music. The court’s treatment of memorization, its rejection of the data-mining exception and its application of U.S. fair-use law all speak to questions at the center of the AI copyright battles: whether models store what they learn, and whether the industry’s practice of ingesting the world’s creative output can continue without payment.
For the music business, the stakes are existential in a different way. Streaming royalties are the industry’s economic engine, and an AI model that can reproduce hits on command threatens the value of the recordings themselves. GEMA, which represents about 100,000 members, has said its goal is a licensing framework in which AI companies pay for what they use, and Friday’s ruling moves that goal measurably closer.
The ruling’s immediate effect on consumers is limited: Suno’s service remains available while the appeal runs, and the company said it would keep operating. But the legal terrain beneath it has shifted. Every music-generation service in Europe now knows that training on the popular repertoire of the past carries a price, and the negotiations between AI companies and rights holders, which were already moving toward licensing, will move faster.
The broader question is what happens next. Suno’s appeal will test the judgment in a higher court, and the European Court of Justice may ultimately decide how the EU’s copyright rules apply to AI training. In the meantime, the message to every company building generative models is the one the Munich court delivered: the works used to train AI systems belong to someone, and the law is beginning to treat them that way. For Suno, the appeal is the next chapter; for the rest of the industry, the negotiations with rights holders that were already underway have just become more urgent, and the price of a license is likely to rise with them.
