On May 9, 2025, the U.S. Copyright Office published the prepublication version of its third report in a series dedicated to the intersection of copyright and artificial intelligence. This installment, titled “Part 3: Generative AI Training”, takes an in-depth look at the use of copyrighted works to train generative AI systems and how it fits within the Fair Use doctrine under U.S. law.
The report is part of a process that began in 2023 with the publication of the Notice of Inquiry, which garnered more than 10,000 public comments on the use of copyrighted content in AI training.
Fair use in generative AI: a case-by-case analysis
The report’s main finding is that determining whether AI training constitutes fair use requires a fact-specific analysis, that is, on a case-by-case basis, taking into account the four factors established in the law:
Purpose of use: Distinguishes between commercial and research use, highlighting that commercial use may fall outside the scope of fair use.
Nature of the copied work: Highly expressive works (such as books, art, music) receive greater protection.
Quantity and substantiality: Mass copying of complete works makes fair use claims difficult.
Impact on markets: Warns of potential “cannibalization” of the creative market by models trained on protected works.
The report states that commercial training of generative AI may not be covered by fair use due to its scale and potential competition with original creator markets.
Risks for Creative Markets and the Need for Licensing
The report warns that training AI with copyrighted materials without authorization could pose significant risks to creative markets by weakening the economic incentives of authors and creators.
It recommends the development of voluntary licensing schemes, in which rights holders and AI developers can negotiate mutually beneficial agreements.
Models such as collective licensing, volume fees, or specific commercial clauses are discussed, although the report acknowledges that implementing these systems will be a complex and long-term process.
Institutional and Political Reactions
The publication of the Report has intensified the public and political debate on AI and copyright:
Creative sector stakeholders, such as the News Media Alliance, have praised the initiative for promoting a balanced approach between innovation and creators’ rights.
The document was seen by many experts as a clear warning to companies and courts about the limits of fair use in commercial AI.
Political Dimension of Perlmutter’s Removal
One day after the release of the preliminary report, the Trump administration removed Shira Perlmutter, the top copyright official, and Carla Hayden, the Librarian of Congress. Perlmutter called her dismissal “unlawful” and filed a lawsuit on May 22 seeking reinstatement, but a federal judge temporarily denied her request.
Analysts and Democrats interpreted this as a political backlash against the report’s content, calling it an “unprecedented capitulation” and linking the removal to Perlmutter’s reluctance to support massive data mining approaches for AI.
Global Legal Context and Future Implications
The report acknowledges that the current legal system is flexible and can be adapted to regulate the use of generative AI, although the resolution will depend on a case-by-case basis. It also mentions international comparative approaches:
The European Union, the United Kingdom, Japan, Israel, and Korea are discussing similar TDM (text and data mining) exemptions with specific conditions.
While no specific legislative reforms are proposed, the report recommends the creation of voluntary licenses as a medium-term solution.
