German Court Delivers Landmark Blow to OpenAI in Song Lyrics Copyright Battle

Munich, Germany – In a pivotal ruling that reverberates across the burgeoning artificial intelligence industry, a German regional court has found OpenAI liable for copyright infringement, determining that its popular ChatGPT chatbot utilized copyrighted song lyrics to train its sophisticated language models without proper authorization. The decision, handed down by the Munich Regional Court on November 11, 2025, marks a significant victory for creative artists and rights holders, potentially reshaping the legal landscape for AI development in Europe and beyond.

The Core of the Dispute: AI Training and Creative Works

The lawsuit was brought by GEMA, Germany's prominent music rights society, which represents over 100,000 composers, lyricists, and music publishers. GEMA alleged that OpenAI had systematically ingested and reproduced lyrics from a protected repertoire of songs, using them to train ChatGPT to understand and generate human-like text. Central to the case were nine widely recognized German hits, including Herbert Grönemeyer's "Männer" and "Bochum," and Helene Fischer's "Atemlos Durch die Nacht," which GEMA demonstrated could be reproduced by ChatGPT through simple prompts.

OpenAI countered these claims by arguing that its language models operate by absorbing vast datasets to identify patterns and relationships in language, rather than storing or directly copying specific content. The company asserted that its models merely reflect learned patterns in their settings and that any infringing output generated by ChatGPT was the responsibility of the end-user who provided the prompt. This defense aimed to position the training process as an analytical act that did not constitute reproduction and to shift liability for generated content.

Court Rejects OpenAI's Defense, Cites "Memorization"

However, the Munich court unequivocally rejected OpenAI's arguments. The presiding judge ruled that both the "memorization" of song lyrics within the language models and their subsequent "reproduction" in the chatbot's outputs constituted clear infringements of German copyright law. The court emphasized that copyrighted works can indeed "subsist within AI models" through the "taking over into the LLMs parameters" of the training data.

Crucially, the court also addressed Germany's text and data mining (TDM) exception, a provision in copyright law designed to facilitate automated analysis of works. It determined that this exception did not apply to OpenAI's actions because the models permanently memorized and reproduced substantial parts of the protected works, exceeding the scope of transient analysis or preparatory measures for training datasets. The court further clarified that responsibility for these infringements lay with OpenAI as the operator, not with the users, due to the company's decisions regarding model architecture and dataset selection.

The ruling explicitly cited §§ 15, 16, 19a, 44b UrhG (German Copyright Act) and Articles 2, 3 InfoSoc Directive, and Article 4 DSM Directive, reinforcing the legal basis for its findings. Damages, the amount of which remains undisclosed, have been ordered, and OpenAI must also cease storing unlicensed German lyrics on its infrastructure within Germany. A symbolic aspect of the judgment includes a mandate for OpenAI to publish the decision in a local newspaper.

Broader Implications for AI and Creative Industries

This German court decision is being hailed as Europe's "first landmark AI ruling" in the creative domain and is expected to have far-reaching consequences beyond Germany's borders. Lawyers and industry experts suggest it sets a powerful precedent, challenging the "fair use" rationale often invoked by tech companies in Silicon Valley for training AI models on vast quantities of data without explicit licensing.

The ruling underscores a growing international divergence in how courts are interpreting AI and copyright. For instance, the Munich court's findings contrast with a recent UK High Court ruling in a case involving Getty Images and Stability AI, where the UK court found no infringement, determining that the AI model did not physically store or reproduce copyrighted images. This highlights the varying legal interpretations of what constitutes "reproduction" in the context of AI models across different jurisdictions.

For the music industry, this verdict represents a significant vindication. GEMA's CEO, Tobias Holzmüller, emphasized that the internet is "not a self-service store" and that "human creative achievements are not free templates" for AI systems. The ruling is seen as a "milestone victory for copyright law" and empowers creators and rights organizations to demand fair compensation when their works are used to train AI. The German Journalists' Association also noted that the ruling sends a clear signal extending to journalistic texts, strengthening the legal position of journalists against AI developers.

OpenAI's Next Steps and the Path Forward

OpenAI has expressed disagreement with the ruling and stated it is "considering next steps," which is widely interpreted as an intent to appeal the decision. A company spokesperson noted that the decision applies to a "limited set of lyrics" and does not impact the millions of people, businesses, and developers using its technology daily in Germany. However, the legal community sees the precedent set as far more impactful than the scope of the specific songs involved.

This case arrives as European regulators are finalizing the AI Act, a comprehensive legislative framework that stresses accountability, data source transparency, and respect for intellectual property in AI development. The Munich court's decision aligns with this regulatory direction, signaling a rigorous approach to IP enforcement against generative AI. While the specifics of compensation and implementation will require further negotiation and potentially more litigation, the ruling clearly establishes that AI developers cannot sidestep licensing obligations when their models can reproduce protected works. The outcome could compel AI companies to pivot towards licensing deals and partnerships, rather than a "scrape first, ask questions later" approach, ultimately reshaping the economics of AI development and the relationship between technology and the creative industries.