On May 9, 2025, the U.S. Copyright Office (USCO) released a pre-publication version of the third and final installment in its multi-part study on copyright issues raised by artificial intelligence. This final report addresses perhaps the most contentious topic in the AI copyright debate: the use of copyrighted works to train generative AI (GenAI) models, and whether such use constitutes “fair use” under Section 107 of the Copyright Act.
Background on the Copyright Office’s AI Initiative
In early 2023, the Copyright Office launched a comprehensive inquiry into copyright law and policy issues related to AI, focusing on the scope of protection for AI-generated works and the legal implications of using copyrighted material in the training of AI models. The final report follows earlier segments addressing digital replicas and copyrightability. Together, the reports provide a broad legal framework that will inform future litigation, legislation, and policymaking in this rapidly evolving area.
Fair Use and GenAI: A Case-by-Case Analysis
The 107-page report provides detailed legal and technical analysis, stressing that the application of fair use in the context of GenAI must be assessed on a case-by-case basis. The Copyright Office reiterates that a fact-specific inquiry into the four statutory fair use factors is essential. These include:
-
The purpose and character of the use, particularly whether it is commercial or nonprofit and whether it is transformative;
-
The nature of the copyrighted work;
-
The amount and substantiality of the portion used; and
-
The effect of the use on the potential market for or value of the copyrighted work.
Regarding the first factor, the Office acknowledges that AI training using large and diverse datasets “will often be transformative.” However, it cautions that “transformativeness is a matter of degree of the model and how it is deployed.” For example, the report notes that training a model for research or non-substitutive use in a closed system is more likely to be viewed as transformative. By contrast, “where the AI output closely tracks the creative intent of the input,” such as in generating music or artwork in the style of a known artist, the case for fair use becomes weaker.
Commerciality and Attribution Complexities
On the second factor—whether the use is commercial—the report observes the complexity of GenAI development, often involving a mix of commercial and noncommercial actors. “Even if an entity is for-profit, that does not necessarily mean the modeling use will be considered ‘commercial,’” the report states. It cites examples such as academic researchers publishing noncommercial studies or nonprofits developing models for later licensing as illustrative of the nuanced analysis required.
Use of Entire Works and Technical Safeguards
As to the third factor, the Office acknowledges that “machine learning processes often require ingestion of entire works,” but cautions that “the wholesale taking of entire works ordinarily weighs against fair use.” Still, the report introduces a nuanced test, asking:
-
Is there a transformative purpose?
-
How much of the work is made publicly available?
The report also underscores that fair use is more likely to apply where models are designed to prevent infringing outputs.
Market Harm and Industry Impact
The fourth factor—market harm—is where the report is most cautious. The Copyright Office acknowledges that fair use analysis in the GenAI context remains “uncharted territory,” but it urges a broad market-effects analysis, not just harm to specific works. The report warns of potential harms from GenAI output that could “displace, dilute, and erode” markets for human-created works, leading to a future where “fewer human-authored works” are sold. The report echoes concerns voiced by creators across sectors, from musicians to illustrators, about GenAI’s ability to undercut licensing opportunities and reduce demand for original content.
“In fields such as illustration, voice acting, or journalism,” the Office writes, “the fourth factor is likely to weigh strongly against fair use” if AI-generated content substitutes for licensed human-created work.
Practice Implications for Developers
Practice Note: The Copyright Office’s position signals heightened legal risk for GenAI developers relying on unlicensed copyrighted content. The report casts doubt on broad assumptions that AI training practices are per se protected by fair use. Companies are advised to consider proactive licensing strategies and technical safeguards to avoid infringing use. As fair use law evolves through court rulings and possible legislative changes, developers should be prepared to adjust their business models accordingly.
Leadership Turmoil in the Wake of the Report
Just one day after the report’s release, the White House abruptly terminated U.S. Register of Copyrights Shira Perlmutter, effective May 10, 2025. On May 12, Deputy Attorney General Todd Blanche—known for representing Donald Trump during his 2024 criminal trial—was named acting Register. The move has sparked controversy and legal questions, as the Copyright Office maintains that “only Congress has the power to fire the registrar or appoint a new one.”
The sudden leadership change raises broader questions about the future direction of copyright policy and the independence of the Copyright Office in the face of shifting executive priorities.