Legal limitations and justifications in the training and deployment of generative AI models under copyright law

Abstract The article investigates the legal limitations and exceptions relevant to the training and deployment of generative AI models under EU copyright law. It analyzes whether existing copyright exceptions, such as those for temporary reproduction, scientific text and data mining (TDM) and commercial TDM, can be applied to AI training processes. These exceptions generally do not apply due to fundamental technological and conceptual diff erences between traditional TDM and AI model training. The authors also evaluate the applicability of other exceptions (e.g., quotation, parody or private use) to the public use and output of generative AI. Ultimately, the article argues that current legal frameworks inadequately address the unique challenges posed by generative AI technologies.