GEMA case against OpenAI: copyright infringement in the training of ChatGPT

The Munich Regional Court I has upheld the claim presented by GEMA, the German copyright management organization, against OpenAI. The case focuses on the unauthorized reproduction of lyrics from nine protected songs during the training and response generation of ChatGPT.

Fundamental Principles

1. The court had to decide whether the memorisation of protected lyrics in the parameters of the AI model and their almost literal reproduction in responses constitutes unauthorized reproduction in the sense of the Copyright Act (UrhG) and the InfoSoc Directive. It also assessed whether these uses could be covered by the text and data mining exception (TDM), invoked by OpenAI to justify training as an analytical act.
2. The Court applied Articles 15, 16, and 19a of the UrhG, which regulate exploitation rights, and analysed the text and data mining exception (TDM) of Article 44b of the UrhG and Article 4 of Directive (EU) 2019/790 of April 17, 2019. These rules only allow preparatory and analytical reproductions, not the memorisation or permanent fixation of works within an AI model.
3. The Court found literal coincidences demonstrating the memorisation and reproduction of works, dismissing that the AI model 'only learns patterns'. It considered that the TDM exception must be applied restrictively, allowing only temporary and analytical copies, not permanent storage or subsequent reproduction. Since permanent memorisation exists, the TDM exception does not apply, and therefore OpenAI is directly liable.
4. The Court found an infringement due to the unauthorized fixation of works in the AI model and their subsequent reproduction in generated responses. The ruling, which is not yet final and is subject to appeal by OpenAI, sets a precedent requiring prior licenses to train AI with protected content, consolidating a strict and protective European approach, different from that maintained by US courts under the fair use doctrine.

Informative note prepared by ECIJA Madrid's TMT department.