Representative Litigation Settlement Agreements in Artificial Intelligence Copyright Infringement Disputes: A Comparative Reflection Based on the U.S

Abstract

The high-density, decentralized copyright conflicts triggered by generative AI training require more than ad hoc solutions; they demand structural governance tools. This article argues that representative litigation settlement agreements offer a distinct institutional advantage. Beyond reducing the transaction costs associated with the "tragedy of the anticommons," these agreements generate market-visible evidence, specifically pricing signals and licensing practices, that validate the "potential market" under the fourth factor of fair use. This phenomenon constitutes procedural market-making. Through a comparative analysis of the U.S. Bartz class action settlement, this study reveals a dual motivation: a surface-level drive for risk aversion and remedy locking, and a deeper logic of constructing a training-licensing market. In the context of Chinese law, the feasibility of such agreements depends not on replicating foreign models, but on establishing three interpretive mechanisms: expanding the functional definition of "same category" claims; adopting a hybrid registration/confirmation system for indeterminate class membership; and converting the "consent" requirement under Article 57, Paragraph 3 of the Civil Procedure Law into a workable opt-out right subject to judicial scrutiny.