Copyright Office’s New Guidance on Fair Use in AI: A Delicate Balance

As the digital age continues to evolve, the intersection of technology and law becomes increasingly complex. In a groundbreaking move, the U.S. Copyright Office has released its third and final report in the “Copyright and Artificial Intelligence” series, offering nuanced guidance on fair use in the realm of generative AI (GenAI) training. This comprehensive analysis, dated May 16, 2025, sheds light on the intricate legal landscape surrounding the use of copyrighted materials in AI model training.

The report, titled Copyright and Artificial Intelligence: Part III – Generative AI Training, does not offer a blanket endorsement or rejection of fair use in AI training. Instead, it emphasizes the need for context-specific evaluations, adhering to the four factors outlined in Section 107 of the Copyright Act. These factors include the purpose and character of the use, the nature of the copyrighted work, the amount and substantiality of the portion used, and the effect of the use on the potential market.

Key Takeaways

1. Transformativeness: The report highlights that the concept of “transformativeness” must be meaningful rather than mechanical. AI training that is transformative often involves large, diverse datasets, but the degree of transformativeness depends on the model’s functionality and deployment.



2. Commerciality: The distinction between commercial and non-commercial use is not solely based on the entity type. Instead, it focuses on the activity itself. Even for-profit entities can engage in non-commercial uses, such as academic research.



3. Whole-Work Usage: The use of entire works in AI training can undermine a fair use defense, particularly when outputs are made public. However, if the model obscures outputs or results in non-expressive outputs, fair use is more likely.



4. Market Harm: Market harm is a central concern. The potential for AI-generated outputs to displace or dilute markets for copyrighted works is significant, especially in fields like illustration and journalism.



5. Licensing and Monitoring: The Copyright Office encourages the development of licensing frameworks and legislative monitoring. While it stops short of endorsing compulsory licensing, it calls for scalable mechanisms to obtain rights for AI training.

Implications for Stakeholders

• Developers and Technology Companies: AI developers should proceed cautiously when using copyrighted material. The report suggests that current training practices may not be broadly protected under fair use, urging developers to consider licensing content.



• Content Creators and Rights Holders: The report supports creators who are concerned about the use of their works without permission. It underscores the importance of exploring registration, monitoring, and enforcement strategies.



• Legal and Compliance Teams: Legal teams should treat GenAI training as a distinct area of copyright risk, ensuring visibility into the provenance of training data and the intended use of outputs.



• Policymakers and Industry Groups: While the report advises against immediate government intervention, it anticipates further congressional interest. Industry groups should prepare for continued dialogue on licensing standards and transparency obligations.

Looking Ahead

This report marks a significant step in clarifying the relationship between copyright law and GenAI development. As the debate continues, companies and creators alike must adapt to the evolving legal landscape. For more detailed insights, the original report is available on the Wiley Rein website.

Authored by legal experts David E. Weslow, Scott Nuzum, and Stephanie Rigizadeh, this report serves as a crucial resource for understanding the future of AI and copyright law. For further reading, visit their profiles on the Wiley Rein website.