A German court has dealt another legal blow to AI companies training on copyrighted material, ruling on July 31 that AI music generator Suno violated copyright law by using tracks from German rights organization GEMA's catalog without a license. The Munich Regional Court found Suno both trained its models on protected works and reproduced them without authorization.

What the Munich Court Found

The case centered on six specific tracks, spanning compositions from the 1970s through the 1990s, that GEMA argued were used without permission both to train Suno's AI models and to generate outputs that echoed the originals. The court's decision now requires AI companies operating in Germany to secure licenses covering both stages of the process — training on a rights holder's repertoire and generating new material derived from it — rather than treating either step as exempt under fair-use-style reasoning.

Someone is operating a sound mixing console.
Photo by Rob Wingate on Unsplash

Leaked Training Data Numbers

Evidence submitted during the case included leaked source code detailing the scale of Suno's training pipeline: more than 113,000 hours of audio pulled from YouTube Music, another 62,000 hours from stock media platform Pond5, and 12,000 hours sourced from streaming service Deezer. The scale of that dataset became a central point in establishing that copyrighted material had been ingested at volume without the licensing the court says was required.

Suno's Response

Suno pushed back on the characterization of its technology, saying it "trained our models to create new songs, not reproduce existing ones." The company indicated it is now weighing "all available options, including an appeal," signaling the ruling is unlikely to be the final word in the dispute.

Suno said it "trained our models to create new songs, not reproduce existing ones," and is evaluating "all available options, including an appeal."

Part of a Bigger Pattern in Europe

The ruling follows a similar defeat GEMA secured against OpenAI roughly eight months earlier, when Munich courts found that ChatGPT had unlawfully reproduced copyrighted song lyrics. Together, the two decisions mark a consistent stance from German courts that AI companies cannot treat copyrighted creative works as free training material, and they add Germany to the list of jurisdictions where rights holders are successfully using litigation to force licensing negotiations with AI developers rather than waiting for legislation to catch up.

Why It Matters Beyond Germany

For AI companies operating across the EU, the ruling raises the stakes of expanding into new markets without first securing licensing deals with local rights organizations. GEMA's back-to-back wins against OpenAI and Suno give it a template other collecting societies across Europe are likely to follow, potentially reshaping how AI firms — including those building tools adjacent to Web3 and crypto media — approach content licensing before launching in new territories.