Munich court to rule on AI music training in landmark GEMA v. Suno case

On July 31, a Munich court will hand down a decision that could reshape how artificial intelligence companies train their systems on creative work. The case pits GEMA, Germany's music collecting society, against Suno AI, a US-based startup that generates songs from text prompts in seconds. What began as a lawsuit filed in January 2025 has evolved into one of the first major tests of copyright law in the age of generative AI—and the stakes extend far beyond Germany.

GEMA's argument is straightforward but consequential: Suno trained its AI models on copyrighted music without licenses or compensation to the songwriters, composers, and publishers those works represent. The company now generates roughly 7 million songs per day across its platform globally, with around 75,000 of them ending up on streaming services daily. GEMA contends this is not merely copyright infringement but unfair competition that threatens the economic survival of human musicians. Suno has countered that the German court lacks jurisdiction over AI training that occurred entirely in the United States, and that the company's practices fall within "fair use" protections and the EU's "text and data mining exception."

The case gained particular attention because it tests something courts have rarely examined: whether AI systems that produce outputs closely resembling existing songs constitute infringement, even if the training itself occurred in another country. During proceedings, GEMA demonstrated this by prompting Suno to generate tracks using the original lyrics of well-known songs—"Forever Young" by Alphaville, "Mambo No. 5" by Lou Bega, "Daddy Cool" by Boney M., and others from GEMA's repertoire. Without specifying melody, rhythm, harmony, or arrangement, the AI produced songs strikingly similar to the originals. Suno acknowledged that these songs were indeed part of its training data.

What makes this lawsuit distinctive is that it was brought not by a record label or individual artist but by a collecting society representing over 95,000 composers, songwriters, and music publishers in Germany, plus more than 2 million rights holders worldwide. This breadth matters. Unlike a lawsuit by a major label protecting its commercial interests, GEMA's action potentially extends protection to less famous creators who might otherwise lack resources to fight for their rights. As Martin Senftleben, a professor of intellectual property law at Amsterdam Law School, noted, such collective action is "particularly important because, by definition, you could say that they're more inclusive."

Tobias Holzmüller, GEMA's CEO, rejected the comparison between how AI systems learn and how human musicians learn from existing work. "These are concepts that come from the human brain and have been designed for the human brain, and not for a machine that is able to digest millions of sound files in split seconds," he said. The sheer volume of data consumed by AI systems—what GEMA calls music but what the industry calls data—bears no resemblance to human learning. Holzmüller's goal is not to eliminate Suno but to force it to the negotiating table as an equal, something he says has been impossible until now.

The economic argument cuts deeper. Until recently, songwriters could earn money composing for film scores and social media content. If AI-generated music remains free or nearly free while human composition requires payment, the market will inevitably shift. Holzmüller's proposed solution is licensing: if AI companies must pay for the right to use copyrighted music in training, and if that cost is passed to users, then AI-generated music would cost roughly as much as hiring a human composer. At that price point, he believes people would choose human-created work for its intangible qualities.

This case arrives as courts worldwide grapple with similar questions. Last year, the same Munich court ruled that OpenAI unlawfully trained its models on copyrighted song lyrics without authorization, ordering the company to cease reproduction and pay damages—though that decision is under appeal. The current lawsuit differs by focusing on entire musical compositions rather than lyrics alone, and by using evidence of output similarity to demonstrate market harm at the point of use rather than at the training stage.

The verdict on July 31 will not settle these questions globally, but it will signal how European courts view the intersection of AI capability and copyright protection. Whether Suno must obtain licenses before training, whether it must compensate artists whose work shaped its models, and whether the economic displacement of human creators constitutes actionable harm—these answers could determine whether AI music companies operate under the same rules as other industries, or whether they remain in a legal gray zone where the scale of their operations outpaces the protections designed for them.