Executive Summary
This paper addresses how uncertainty over copyright infringement and fair use in the context of generative AI (GenAI) creates fundamental instability in AI development and across creative markets, imposing a de facto “uncertainty tax” that raises capital costs, protects incumbents, and stifles new entrants. Against this backdrop of massive capital requirements and precarious litigation risks, the paper makes four essential assertions: (1) GenAI is a general-purpose technology that aids expression; (2) GenAI systems can respond to end-user commands to produce an infinite variety of new, original expression but can also be used to produce substantially similar and potentially infringing content; (3) there is a complete disconnect between allegations aimed at model training and the elements that govern infringement and fair use for outputs; and (4) the proper defendants are not the trainers and developers of the AI systems, but the end-users who determine the function and purpose of particular uses of this general purpose expressive technology.
The paper explains that plaintiffs’ legal theory is misplaced: the class actions anthropomorphize a tool into an actor and sue the wrong defendants. GenAI is not a person, not an artist, not an author; it is a sophisticated tool. Training of AI systems draws data at web-scale from training samples to build a general-purpose expression machine, and this process has been held “exceedingly,” “spectacularly,” and “quintessentially” transformative in part because training of a model does not, by itself, republish protected expression. By contrast, infringement and fair use questions about outputs turn on what end-users cause the tool to generate and how they use those outputs under the transformative function-and-purpose test clarified in Warhol.
Realigning the correct claims with the correct defendants accords with decades of doctrine on dual-use and intermediate copying technologies from photocopiers and VCRs to search engines and software interfaces. The paper argues that courts should not accept the alleged hypothetical harms to participation in “markets that could have been,” realign the claims to address the actions of the true responsible parties to avoid doctrinal distortions, and apply the established copyright fair use framework to exonerate the creators of general-purpose tools for transformative training, and evaluate end-user outputs under ordinary infringement and fair-use analysis. This approach protects both creators’ rights and American innovation.
