In a landmark ruling, the Regional Court of Munich I set strict legal boundaries for generative music platforms. In proceedings involving German collecting society GEMA and AI platform Suno, the court determined that training generative AI models on copyrighted music without a licence constitutes copyright infringement whenever models memorise or reproduce protected elements in their outputs.
A cornerstone of the court's rationale addresses statutory exceptions for text and data mining (TDM). The judges clarified that TDM provisions do not apply when a generative model is designed to absorb protected works and recreate them upon receiving user prompts. The court firmly dismissed arguments that training generative audio architectures qualifies merely as non-infringing statistical data analysis.
Furthermore, the ruling underscored that European technology regulations do not provide a safe harbour against traditional copyright claims. Complying with transparency obligations or adhering to Codes of Practice under the EU AI Act does not exempt AI providers from securing appropriate licences. Traditional copyright law maintains its independent enforceability across creative content.
The court also rejected intermediary liability protections for generative AI operators under the Digital Services Act (DSA). The judges ruled that Article 6 DSA safe harbour provisions do not protect Suno, because the company actively directs and controls the memorised output generation through its deliberate curation and processing of training datasets.
The decision by the Munich Regional Court represents a pivotal development in European case law concerning generative artificial intelligence. AI audio developers will likely face intensified pressure to negotiate comprehensive licensing deals with rights holders, substantially bolstering the leverage of artists and music publishers across the European Union.