The Copyright Office concludes in a pre-published report that the fair use exception should generally not apply when training generative AI tools with copyrighted works. The subject of the relationship between artificial intelligence and copyright continues to dominate the headlines in the USA, with over thirty lawsuits against Google, OpenAI, Perplexity and Meta: we follow them in this series of articles. In our previous three episodes, we talked about the arguments of the platforms, the early ruling that decided in favor of the rightsholders, and the arguments of law professors in the Kadrey v Meta case. In this 4th episode, we move somewhat out of trial territory, with a report from the Copyright Office that could nonetheless weigh in on the various ongoing cases involving generative AI training with copyrighted works. It is interesting to note that, in addition to assessing American law, the Copyright Office (C.O.) analyzes the current provisions of European law on AI training and copyright in a rather dubious light. Pre-publication and departure of Shira Perlmutter A little background, to start with. The report is entitled « Copyright and Artificial Intelligence – Part 3: Training Artificial Generative Intelligence ». The Copyright Office states on its site that this is a « pre-publication version ». Unlike the other two reports that preceded it (one on digital twins, and the other on whether…
