A Study on the Normative Framework and Policy Dynamics of the Fair Use Doctrine under U.S. Copyright Law

Abstract

Large models such as ChatGPT and Stable Diffusion rely heavily on automated scraping and machine learning of massive amounts of text, images, and code during pre-training. Whether this process, in its legal nature, constitutes an infringement of the copyright holder’s exclusive rights or can be exempted through the fair use doctrine has become one of the most controversial frontier propositions in current intellectual property law. This article analyzes whether generative AI’s scraping of training data constitutes fair use under U.S. copyright law. It systematically examines the application boundaries of the four-factor framework in AI scenarios, integrates the Copyright Office report and congressional legislative dynamics, reveals the tripartite game pattern among the judiciary, the administration, and the legislature, and provides methodological enlightenment for China’s relevant institutional design.