A Munich court has ruled that Suno, a popular AI music generator, violated copyright law both during training and when producing output. The court identified six songs that were reproducibly stored within Suno's models, establishing clear evidence of infringement rather than incidental similarity.
The decision carries weight because it rejected multiple defenses commonly cited by AI companies. Suno argued that Germany's text-and-data-mining exception should apply, allowing AI systems to learn from copyrighted material for research and development. The court disagreed. The company also invoked the US fair use doctrine, which permits certain uses of copyrighted work without permission. The Munich court rejected this defense as well, finding it incompatible with German and European copyright law.
The ruling marks one of the first major judicial determinations that AI training on copyrighted material without consent constitutes infringement, not just a technical gray area. It signals that European courts will take a stricter approach to AI copyright cases than US courts have historically done.
However, the decision remains preliminary. The court made clear that several key questions remain unresolved. Judges did not establish how many instances of reproducible content trigger liability or whether different standards apply depending on the type of work infringed. They also left open the broader question of whether companies can claim fair use protections under any circumstances when training generative models.
Suno has options to appeal. The company may argue that the reproducible songs represent statistical anomalies rather than intentional memorization, or that removing infringing training data should retroactively shield it from liability. Other AI music generators and broader generative AI developers will watch closely as this case develops.
The ruling establishes that European copyright protections extend directly to the training process itself. This differs from some AI discourse suggesting liability only attaches to final outputs. For companies developing generative models, the Munich judgment means training data provenance now