Author

Zhang Jiyu, Professor, Law School; Research Fellow, Law and Technology Institute, Renmin University of China

Li Xinmeng, Research Fellow, Law and Technology Institute

Abstract

The use of copyrighted works in training large AI models has sparked numerous lawsuits globally. This Article examines China’s evolving regulatory landscape, and analyzes two academic proposals for China’s Artificial Intelligence Law, identifying key areas of divergence and consensus regarding the fair use of copyrighted works in AI training. By comparing three different legal approaches to characterizing AI model training, this Article argues that this process qualifies as fair use. This is because machine learning leverages vast corpora to internalize underlying linguistic and creative patterns, rather than storing or directly reproducing the protected works. As a result, the use of copyrighted material in the training phase qualifies as incidental reproduction and transformative use, which, according to our empirical study, does not unreasonably harm the legitimate rights and interests of copyright holders. Furthermore, given themarket failure in AI model training licensing, this Article contends that recognizing AI model training as fair use better aligns with China’s legal framework and the practical needs of technological development. To ensure legal certainty, this Article proposes introducing a machine learning exception within either the ongoing revision of the Regulations for the Implementation of the Copyright Law, or future AI legislation in China.

Keywords:Generative AI; copyright law; fair use; model training; artificial intelligence law