In its ruling of November 11, 2025 (Ref. 42 O 14139/24), the Munich I Regional Court handed down a landmark decision in the legal dispute between the collecting society GEMA and the AI provider OpenAI. The central issue is whether and to what extent copyright-protected song lyrics may be used in language models such as ChatGPT – in particular with regard to training, storage, and subsequent reproduction by AI. The ruling is likely to have significant implications for the creative industry and for the increasing use of generative AI systems.
The background to the proceedings was GEMA's claim that OpenAI had used entire copyrighted song lyrics without a license when training ChatGPT. Specifically, this concerned nine well-known German songs, including titles such as “Atemlos” (Kristina Bach), “Männer” (Herbert Grönemeyer), “Über den Wolken” (Reinhard Mey), and “Wie schön, dass du geboren bist” (Rolf Zuckowski). According to GEMA, these lyrics were not only included in the training data, but could also be reproduced in large parts by ChatGPT in response to simple user inputs (so-called prompts) (so-called outputs).
OpenAI, on the other hand, denied having stored the song lyrics in the form described by GEMA. The company argued that language models do not “store” entire works, but rather learn statistical patterns. The outputs generated by users were the result of probabilistic calculations, not the blind retrieval of stored text files. In addition, OpenAI invoked several copyright exceptions, in particular the text and data mining barrier (Section 44b UrhG) and the provision on insignificant accessories (Section 57 UrhG).
However, the Munich I Regional Court largely followed GEMA's line of argument. In the opinion of the 42nd Civil Chamber, the decisive factor was the so-called “memorization” of the song lyrics by the AI model. The judges found that the lyrics were embedded in the parameters of the models (such as GPT-4 and GPT-4o) in such a way that they could be retrieved when users entered certain prompts. Given the length, complexity, and characteristic structure of the song lyrics, random reproduction was practically impossible.
From this memorization, the court deduced an inadmissible reproduction within the meaning of copyright law (Section 16 UrhG and the European Directive on intellectual property rights, Art. 2 InfoSoc-RL). But that's not all: the court also considered the output of these texts by ChatGPT to be an independent infringement of copyright. According to the chamber, if users can retrieve certain song lyrics via prompt, this constitutes a form of making them publicly available – and is therefore an act relevant to copyright for which OpenAI is responsible. The court rejected OpenAI's objection that users are responsible because the prompts come from them. Rather, OpenAI, as the operator of the architecture, the training data set, and the model structure, is largely responsible for what the system outputs.
The court also rejected the exceptions put forward by OpenAI. The text and data mining restriction provision (§ 44b UrhG) does, in principle, allow the searching and collection of data and the evaluation of copyright-protected works for scientific or data analysis purposes. However, in the opinion of the Regional Court, what OpenAI does is not just evaluation, but goes far beyond that to include the storage and permanent copying of entire works, which can be accessed at any time. This means that the use seriously infringes the exploitation interests of the authors. Even invoking the concept of insignificant accessories (Section 57 UrhG) did not help: the court emphasized that the song lyrics were not incidental components of a larger work, but rather independent content worthy of protection with its own copyright value.
In its ruling, the court clarified that OpenAI has control over training, data, and architecture—and is therefore also liable for any legal violations. The judge drew a simple but concise analogy: if you want to build something and need components, you have to purchase them and not use the property of others.* This is therefore not a case of accidental or unintentional data retrieval, but rather the planned use of intellectual property without consent or licensing.
The Munich I Regional Court awarded GEMA extensive claims: OpenAI must refrain from reproducing the texts in question in the future and is obliged to provide information about the scope and use. In addition, there are claims for damages because the use took place without a license. However, the court rejected claims based on personal rights, such as incorrect attribution of modified song lyrics.
The implications of this ruling are far-reaching: it marks a clear breach of law for providers of generative AI who integrate copyright-protected works into their training data without a license and store them in their models. The ruling sends a strong signal to the tech industry: creative works are not free templates just because they are used for training artificial intelligence. For authors and collecting societies, this decision could be a breakthrough in negotiating fair remuneration when their works are to be used in an AI context.
Furthermore, the ruling also has implications for the legal classification of AI training processes: it shows that existing legal exceptions such as the text and data mining barrier do not permit all forms of use across the board – in particular, the permanent storage and retrieval of complete works. This means that action is needed - both at the legislative and contractual levels—to clearly regulate the use of creative content by AI providers.
For AI providers, the ruling results in a key obligation: before using copyright-protected texts in training, they should conclude license agreements with rights holders. The risks of not doing so are not only legal but also economically significant. For rights holders, on the other hand, the ruling creates an improved negotiating position: they can insist more strongly on remuneration and have a say in the terms of use.
As the ruling is not yet final, an appeal by OpenAI cannot be ruled out. Nevertheless, the case is likely to serve as a reference for future AI litigation – not only in the music industry, but also in other creative fields such as literature, visual arts, and photography. It dispels the myth that AI can use copyrighted texts without permission and sets a clear framework for balancing innovation and intellectual property protection.
