In a landmark decision with potential Europe-wide implications, the District Court of Munich I has ruled that ChatGPT infringes the copyrights of GEMA and its members. In a case brought by GEMA, the German collecting society representing songwriters, the court held that both the training of the ChatGPT model using GEMA’s repertoire and the generation of AI-created music based on that training violate German copyright law.

Although the fully reasoned judgment has not yet been released, the court’s official press release indicates that the ruling may become a significant reference point for licensing creative works to AI developers for both training and deployment.

The Judgment

The court found that OpenAI had clearly used GEMA’s repertoire in training ChatGPT and that resulting model outputs would be highly similar to the copyrighted works in that repertoire. It further concluded that no copyright exception applies. In particular, the statutory text and data mining (TDM) exception was deemed insufficient to cover AI training processes that involve not only information extraction but also the reproduction of protected works within a model. In the court’s view, such reproduction interferes with rights holders’ ability to exploit their works.

In the absence of a licence, the court held that both the training and resulting outputs infringe GEMA’s rights. It ordered OpenAI to cease activities involving GEMA’s repertoire, provide information on the scope of use and related revenues, and pay damages. A minor portion of GEMA’s claims relating to personality rights was dismissed. The court also declined to refer the case to the European Court of Justice, despite both parties requesting a preliminary ruling.

Significance

This is the first major European decision to hold an AI developer liable for training on copyright-protected works without a licence. While the ruling is not final and may be appealed, a final-instance win for GEMA would reinforce the position of rights holders and could solidify broad licensing requirements for generative AI development.

GEMA has already positioned itself as a leader in this space. In September 2024, it became the first collecting society to introduce a dedicated generative AI licensing model designed to balance innovation with meaningful participation for creators in the economic value generated by AI systems.

The judgment also reaches well beyond music, touching on the use of any copyrighted or protected material in AI development. It highlights the growing need for lawful data sourcing, transparency around training data, and rights-respecting deployment practices.

Broader Legal Context

The decision sits in tension with several recent judgments and underscores the territorial nature of copyright law. The District Court of Hamburg’s decision in Kneschke v. LAION applied the TDM exception in favour of a defendant providing datasets to AI developers. In the United States, decisions in Bartz v. Anthropic and Kadrey v. Meta suggest that fair use may apply to AI training. The ruling also arrives just one week after the UK High Court’s decision in Getty Images v. Stability AI, reflecting a rapidly evolving and increasingly fragmented global landscape.

Key Takeaways

The Munich ruling is likely to influence courts, regulators, and lawmakers beyond Germany. It may be welcomed by rights holders across creative sectors. For AI developers, it signals that reliance on the TDM exception may not be sufficient and that courts may scrutinize the underlying mechanics of AI training more closely.

OpenAI has indicated it will appeal. Until a higher court weighs in, more litigation and divergent outcomes across jurisdictions are likely as courts continue to grapple with the copyright implications of generative AI.