ABSTRACT
Generative AI systems can sometimes produce output that infringes copyrights. The question is: who is liable for the unauthorised acts of reproduction involved? The first candidate would be the user, as the person who directly initiates and steers the generation process. One could also argue, however, that in certain types of case, it is actually the AI provider who directly infringes. In other cases it may be appropriate to at least hold the AI provider secondarily liable. As far as communication to the public is concerned, the CJEU has already established a flexible liability regime covering such secondary liability cases. This regime also leads to compelling conclusions regarding the secondary liability for reproductions in AI output.
In the event that output produced by an AI system infringes copyrights, a nuanced, case-by-case assessment must be made. This article looked at three different scenarios: (1) copyright infringements produced as a result of user inputs, (2) copyright infringements due to system-based causes (in particular, training), and (3) copyright infringements inherent in the intended purpose.
Nordemann, Jan Bernd and Pukas, Jonathan and Baumann, Malte, Liability for copyright infringing output of generative AI systems – evolving the European reproduction right (November 7, 2025). English version of the German language article ‘Haftung für Urheberrechtsverletzungen im Output generativer KI-Systeme’, published in Gewerblicher Rechtsschutz und Urheberrecht (GRUR) 2025, 955. The English version has been prepared with Adam Ailsby, Belfast, https://www.ailsby.com/.
Leave a Reply