Text and Data Mining (TDM) Exception and AI: First European Ruling Denying Its Benefit to ChatGPT (OpenAI)

To encourage innovation in technologies that require the analysis of large amounts of data, such as the training of AI models, the so-called text and data mining (TDM) exception of the 2019 Copyright Directive allows the use of protected content subject to certain conditions.

The Regional Court of Munich has just held that OpenAI’s use of nine songs from the GEMA repertoire (the German equivalent of SACEM) to train its ChatGPT IA does not comply with this conditions and therefore amounts to copyright infringement:

1️⃣ TDM exception not applicable: the works were not only used for training the ChatGPT, but were also reproduced in outputs, demonstrating that the model had “memorised” the works — whereas the TDM exception does not allow for the permanent reproduction of protected works.

2️⃣ A ruling with potential European significance: the TDM exception originates from Directive 2019/790 on copyright and related rights in the Digital Single Market, which has been transposed with comparable provisions across EU Member States. The German decision could therefore be followed by other jurisdictions in the European Union, such as France.

🔎 However, OpenAI has announced it will appeal, meaning the decision is not final… to be continued!