Munich court orders Suno to license AI training data and output
A German court has ruled that AI music startup Suno must license both its training data and generated output, establishing a costly precedent for the generative AI sector.
The Munich Regional Court ruled against AI music startup Suno in a copyright lawsuit brought by GEMA, finding the company violated copyright law by using the rights group's catalog to train its models. The court mandated that AI companies must secure licenses for both the training phase and the subsequent generation of music using GEMA's repertoire.
The lawsuit focused on six specific tracks, including "Daddy Cool," “Rasputin," "Forever Young," and "Mambo No. 5." GEMA had accused Suno of ingesting these copyrighted songs without permission to build its core product.
For investors and AI developers, the ruling represents a significant escalation in the legal costs associated with generative AI. It builds on a previous Munich court decision roughly eight months ago, where GEMA secured a similar victory against OpenAI over ChatGPT reproducing song lyrics. Together, these decisions signal a consistent and expanding legal standard in Germany that treats both AI inputs and outputs as licensable commercial activities.
Suno pushed back against the decision, stating it plans to evaluate its legal options, including a potential appeal. “From the beginning, we trained our models to create new songs, not reproduce existing ones, and built protections into our platform,” Suno said in a statement. “We disagree with today’s ruling—which rests on a fundamental mischaracterization of how Suno’s technology works, how it is used and how U.S. law applies—and are evaluating all available options, including an appeal.”
The court's decision arrives amid separate, mounting legal troubles for Suno regarding its training methods. Leaked source code earlier this month revealed the startup's training data contained over 113,000 hours of YouTube Music, alongside 62,000 hours from Pond5 and 12,000 hours from Deezer.
Those figures underscore the sheer volume of copyrighted material required to build competitive AI models and the massive potential liability if rights holders successfully demand compensation. If the dual-licensing requirement established in Germany gains traction in other jurisdictions, the capital required to train and operate foundation models could increase substantially. This would directly pressure the unit economics and profit margins of AI startups across the sector.
Suno’s defense relied partly on the application of U.S. law, highlighting the jurisdictional friction that global AI developers face. European courts are increasingly asserting local copyright standards over foreign-trained models operating within their borders. For multinational AI firms, this legal fragmentation means navigating a complex patchwork of licensing requirements rather than relying on a single, unified legal shield.
