German Court Rules AI Music Firm Suno Violated Copyright

A German court ruled Wednesday that U.S. artificial intelligence startup Suno violated copyright laws by training its music-generation models on protected songs without permission. The decision marks one of the first binding European rulings on the legality of training commercial AI systems on copyrighted media.

The Munich Regional Court found that Suno had no legal right to process the work of artists represented by GEMA, Germany’s state-mandated licensing body. The court concluded that Suno’s systems actively memorized and reproduced six specific GEMA-represented songs during the training of its AI models.

The ruling forces Suno to disclose detailed information regarding the revenue it earned from the infringement. Once that information is handed over, the court will calculate the financial damages owed to the rights holders.

Under German civil law, a first-instance judgment can be enforced even while an appeal is pending. This procedural rule means GEMA could pursue immediate injunctions to restrict or halt Suno’s operations across European markets without waiting for the appeals process to conclude.

Inside the Courtroom

The lawsuit, filed by GEMA earlier this year, targeted the fundamental methods used to build modern generative AI systems.

During hearings that began in March, GEMA lawyers presented visual and audio evidence demonstrating that Suno’s software could generate music that closely mirrored existing copyrighted works. By entering simple text prompts, users could prompt the AI to generate tracks that reproduced recognizable melodies, harmonies, and rhythms from the original songs.

The examples chosen by the plaintiff were highly recognizable. GEMA demonstrated that Suno’s outputs contained the distinct musical fingerprints of tracks including “Forever Young,” “Mambo No. 5,” “Daddy Cool,” “Atemlos,” “Rasputin,” and “Big in Japan”.

The 42nd Civil Chamber of the Munich Regional Court agreed with GEMA’s core argument: the presence of these exact melodies in the generated outputs proved that the original songs had been memorized by the system during the training phase, which constitutes an illegal reproduction under copyright law.

Suno mounted an aggressive defense leading up to the verdict. The company’s legal team questioned the copyright protection of the original works, challenged GEMA’s legal standing to bring the suit, and disputed the similarity of the AI outputs to the original tracks. In the weeks before the ruling, Suno filed fresh legal arguments and attempted to have the presiding judge removed from the case. The court rejected the maneuver.

The Technical Argument

The case required the court to examine the underlying architecture of AI music generators.

During the proceedings, the chamber analyzed how Suno operates its platform using a combination of transformer and diffusion models. To train these models, AI companies ingest massive volumes of audio files alongside associated metadata. This semantic, syntactic, and structural information is converted into mathematical vectors, which the system uses to determine the parameters—or weights—that allow it to generate new music from text prompts.

A central technical dispute involved how Suno acquired its training data. Both parties acknowledged that the six specific tracks in question were accessed via YouTube. GEMA accused Suno of actively circumventing YouTube’s “Rolling Cipher”—a technical protection mechanism that obscures the direct storage location of audio and video files to prevent automated downloading. The court noted this behavior but left open the question of whether circumventing the cipher constituted a separate violation of technical protection measures under the German Copyright Act.

The ruling clarifies a crucial technical threshold for AI developers. It establishes that when a machine learning model internalizes copyrighted material so deeply that it can reproduce substantial elements of it—a phenomenon data scientists call “overfitting”—it crosses the line from data analysis into unlawful reproduction.

Broader Industry Stakes

The decision lands at a precarious moment for the generative AI sector, which has built multi-billion-dollar valuations on the premise that scraping publicly available internet data falls under fair use or data mining exceptions.

Suno, based in Massachusetts, reached a valuation of approximately $5.4 billion in a funding round earlier this year. That valuation relies heavily on the capabilities of the very AI models that the Munich court has now deemed unlawful under German law.

The ruling signals a growing divide between how AI companies and traditional rights holders view the legal obligations of training models. For GEMA, the legal calculus is straightforward: commercial AI developers must secure explicit permission and pay licensing fees before using protected music to train their models. Collecting societies in Europe are legally obligated to grant usage rights on reasonable, non-discriminatory conditions to anyone who requests them, but GEMA noted that Suno never initiated settlement negotiations or offered payment.

The music industry is currently fighting AI copyright battles on multiple fronts, combining aggressive litigation with strategic licensing deals.

Suno is facing separate, massive copyright infringement lawsuits in the United States brought by major record labels. However, the industry is also showing a willingness to settle when terms are favorable. Recently, both Suno and its primary competitor, Udio, reached settlements with Warner Music, while Udio secured a separate agreement with Universal Music Group.

Simultaneously, major platforms are attempting to build a fully licensed ecosystem for generative music. Spotify recently signed a licensing deal with Universal to create a legal framework for AI-generated covers and remixes. These parallel tracks suggest the music industry is attempting to force AI companies to the negotiating table through court victories, rather than trying to ban the technology entirely.

The European Legal Environment

Munich is quietly establishing itself as a central venue for global AI copyright disputes.

The same regional court issued a related ruling late last year against OpenAI. In November 2025, the 42nd Civil Chamber found that OpenAI’s ChatGPT infringed copyright by reproducing almost verbatim excerpts of German song lyrics when prompted by users.

In both the OpenAI and Suno cases, the court systematically dismantled a common defense used by AI companies operating in Europe: the Text and Data Mining (TDM) exception. Under the German Copyright Act, researchers and companies are allowed limited rights to reproduce works for the purpose of analyzing text and data. However, the court ruled that Suno’s activities went far beyond mere analysis. Because the models essentially memorized the compositions well enough to reproduce them as direct outputs, the TDM exception did not apply.

The court also established that liability sits with the model providers, not just the users who enter the prompts. Because companies like Suno direct the architectural design, control the training data, and deploy the commercial service, they hold responsibility for the resulting copyright infringement.

What Happens Next

The immediate threat to Suno is operational. Because the ruling is enforceable immediately, GEMA has the legal leverage to seek injunctions that could force Suno to geoblock its services in Germany, or potentially alter its models for European users.

Suno has the option to appeal the decision to a higher German court. However, the company now faces a difficult strategic choice regarding its European operations. It can continue to fight the ruling through the appeals process, attempt to negotiate a retroactive licensing settlement with GEMA, or restrict its services in regions with strict copyright enforcement.

The Munich decision does not directly impact the ongoing litigation in the United States, which will be decided under American fair use doctrines rather than European copyright directives. But as one of the first binding judicial decisions anywhere in the world on AI music training, the German ruling provides a clear legal roadmap for rights holders. It demonstrates that courts are willing to look under the hood of AI models, examine their training data, and hold developers financially accountable for the media they ingest.

For the broader generative AI industry, the ruling underscores a mounting financial reality: the era of building commercial AI models on unlicensed data is facing serious, quantifiable legal barriers.