The Regional Court Munich ruled that OpenAI had infringed GEMA’s copyright by using the songs in question (in the following called “data”).
The court ordered OpenAI to refrain from reproducing the data and storing it on any infrastructure located in Germany. In particular, the court was convinced that the disputed song lyrics are reproducibly contained within the OpenAI models (ChatGPT versions 4 and 4o) and that such memorization constitutes a reproduction under copyright law. The court found that “reproduction” needs to be interpreted broadly in accordance with applicable EU copyright directives and that the fact that the model stores works within its parameters as probability values (and not as plain copy) is irrelevant when assessing whether a reproduction has taken place. The court argues that the data can be retrieved from the parameters.
The court also held that OpenAI can not rely on the TDM exceptions of Sections 44b and 60d Germany Copyright Act (Art. 3 and 4 EU DSM Directive, (EU) 2019/790). The court takes the opinion that these exceptions allow the reproduction of the data only for analytical purposes and only as long as they don’t interfere with the exploitation interests of rights holders. The reproductions within the model, however don’t serve such analytical purposes so the rules don’t apply. With this view, the court explicitly rejected broader interpretations of the TDM exception which are more technology and innovation friendly.
The court also held OpenAI liable in terms of damages, including information about the use and the profits generated, but left the amount to be determined in subsequent proceedings.
The court did not refer the case to the European Court of justice to decide on the underlaying EU copyright law as suggested by both OpenAI and Gema.
The court only dismissed the claim of a personality rights infringement which claimed that the erroneous attribution of altered song lyrics infringed the authors’ rights.