The ruling by the 42nd Civil Chamber of the Regional Court of Munich I (judgment of November 11, 2025, case no. 42 O 14139/24) was a landmark decision. GEMA brought an action against two companies in the Open AI Group for injunctive relief, damages and information. The central question was whether the defendant language models were illegally reproducing and reproducing copyrighted song lyrics.
The Munich judges ruled in favor of the plaintiff and thus handed down a groundbreaking judgment that explicitly deals with the copyright liability of AI models.
Legal basis of the claims asserted
At the heart of the case was the question of whether the AI models memorized and thus reproduced copyrighted texts. According to the court, reproduction within the meaning of Section 16 UrhG and Art. 2 InfoSoc Directive already exists if copyright-protected works are reproducibly contained in the so-called model parameters of an AI model. These parameters consist of values in which the knowledge acquired by the AI through training is defined. The decisive factor here is not that the information is only stored as probabilities, but that it also remains technically perceptible.
The defendant OpenAI invoked the limitation for text and data mining provided for in Section 44b UrhG and argued that the use of corresponding training data was exclusively for analysis purposes. This did not convince the court.
No permitted text and data mining
In the opinion of the court, this limitation of copyright only applies if mere information is extracted when training the models, but not complete works are reproduced. The court also denied an analogous application of Section 44b UrhG due to a lack of comparable interests.
The non-essential accessory under Section 57 UrhG is also not relevant, as the song lyrics represent central content and cannot be considered incidental components of the training data set.
The issue of responsibility was another key point in the Munich ruling. The judges agree: the operators of the language models are responsible for copyright-infringing reproductions, as the results of the AI are the direct result of the model training that took place beforehand. The end user is therefore unable to recognize whether an AI model produces copyrighted texts.
Possible consequences of the procedure for rights holders and AI providers
The ruling is likely to have a significant impact on the AI sector even before it becomes legally binding: It makes it clear that copyright law is also applicable to the new challenges posed by the AI revolution and that technical innovation must not be at the expense of creative people.
