Wilt Toikka Kraft LLP

Generative AI Training Could Fall Outside Fair Use Protection

Copyright Office Releases Final AI Report, Raising Doubts About Fair Use Defense for Model Training

Last week, the U.S. Copyright Office published the third and final installment of its comprehensive report on copyright issues raised by artificial intelligence (AI). Unlike the first two parts, this installment was released as a “pre-publication” version—under unusual political circumstances. It was issued less than 24 hours after President Trump dismissed Dr. Carla Hayden, the Librarian of Congress, and just one day before he also fired Shira Perlmutter, the Register of Copyrights.

Building upon the findings of the earlier installments, the final report zeroes in on how copyright law—and particularly the fair use doctrine—should apply to companies that use copyrighted materials as part of AI model training.

The Copyright Office did not mince words in its conclusions. It found that “companies presumptively infringe the copyright protections of others when they copy materials to use in training data.” Moreover, it warned that infringement may extend beyond training inputs: “the numerical parameters of the model can also be infringing when it can reproduce the copyrighted work as a memorized example.”

While the fair use doctrine permits the use of copyrighted content under certain conditions, the Office raised serious concerns about its application in the generative AI context. Although it acknowledged that “many uses of copyrighted works by an AI model are likely transformative,” it emphasized that the commercial use of such materials to develop products that compete with the original works undermines a fair use defense. As the Office noted, “AI models rapidly create new works that imitate a creator’s style,” and concluded that “this market dilution weighs against the fair use argument for generative AI companies.”

Acknowledging the necessity of copyrighted works in training large-scale AI models, the report’s final section turned to licensing as a potential solution. It explored voluntary licensing frameworks that would allow companies to lawfully obtain training data. However, the Office stopped short of recommending immediate government action, stating instead that “the market should continue to develop” before any regulatory intervention.

Given the timing of the release and the political upheaval at the Library of Congress and Copyright Office, questions remain about whether the Trump administration may rescind or revise the report. Still, the current version carries significant legal weight and offers guidance that AI developers would be prudent to consider—regardless of any forthcoming administrative changes.

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top