The Munich Regional Court struck a blow to the business models of generative AI companies, ruling that OpenAI’s ChatGPT infringed the copyrights of songwriters represented by the collecting society GEMA. The dispute centered on "memorization," where GEMA demonstrated that ChatGPT could reproduce exact lyrics of copyrighted songs when prompted. OpenAI defended its model by arguing that it does not store copies of works but rather "probability weights" and statistical associations, and that the model "learns" patterns rather than storing text.
The Court rejected OpenAI's "statistical defense" as a metaphysical distinction irrelevant to copyright law, holding that if a model can reproduce a work with near-perfect fidelity, it has effectively "reproduced" it. Furthermore, the Court denied OpenAI the protection of the Text and Data Mining (TDM) exception, reasoning that TDM allows for data analysis but not for the generation of competing content. By outputting lyrics, the AI acts as a market substitute for the original work, violating the principles of copyright exceptions and establishing that generative AI outputting training data constitutes infringement.
Source: Bird & Bird