The U.S. Copyright Office’s AI Initiative and Its Legal Framework
Copyright Office launched its Artificial Intelligence Initiative in March 2023 to systematically examine how copyright law applies to AI-generated works and the use of copyrighted materials in AI training. These sessions aimed to gather stakeholder perspectives on emerging challenges, such as the copyrightability of AI outputs and the implications of AI training on existing copyright protections. The Office’s efforts culminated in a Notice of Inquiry published in the Federal Register on August 30, 2023, which solicited public comments. Over 10,000 responses were submitted by December 2023, reflecting broad industry and public engagement with the issues.
The initiative’s findings are being published in a multi-part report titled Copyright and Artificial Intelligence, with each part addressing distinct legal and policy questions. Part 1, released on July 31, 2024, focused on digital replicas and recommended federal legislation to address gaps in copyright protection for high-fidelity reproductions of existing works. Part 2, published on January 29, 2025, centered on the copyrightability of outputs generated by AI systems, reinforcing the Office’s position that human authorship remains a prerequisite for copyright protection. Part 3, released in a pre-publication version on May 9, 2025, responds to congressional inquiries and stakeholder feedback, though no substantive changes to its analysis are expected in the final version.
The Office’s guidance on AI-generated works is voluntary, meaning it does not carry the force of law but instead provides interpretive direction to applicants, registrants, and the public. The guidance reflects the Office’s stance that AI outputs lack copyright protection unless a human author contributes creative expression to the final work. This position is currently being tested in ongoing litigation, including Bartz v. Stable Diffusion LLC, and The New York Times Co. v.
Key Legal Positions: Human Authorship and Fair Use in AI Training
The Copyright Office’s report and registration guidance emphasize two critical legal positions: the requirement for human authorship in AI-generated works and the limited scope of fair use for AI training datasets. On human authorship, the Office has consistently ruled that copyright protection does not extend to works produced solely by AI, regardless of their technical sophistication or creative output. For example, the Office’s Review Board decisions in cases such as Zarya of the Dawn (February 21, 2023) and A Recent Entrance to Paradise (February 14, 2022) have denied registration for AI-generated works lacking human involvement, reinforcing this principle.
Regarding fair use, the Office acknowledges that the use of copyrighted materials in AI training may qualify as fair use under Section 107 of the Copyright Act, but only under specific circumstances. The Office’s guidance suggests that transformative uses—such as training AI models to generate novel creative outputs—may be more likely to qualify as fair use, whereas wholesale replication of copyrighted works for training purposes is less defensible. However, the Office has not issued definitive rulings on fair use in AI contexts, leaving this area subject to judicial interpretation in ongoing litigation.
Registration Guidance and Practical Implications for AI Developers
The Copyright Office’s Registration Guidance for Works Containing AI-Generated Content, published in the Federal Register on March 16, 2023, provides practical instructions for applicants seeking to register works that incorporate AI-generated elements. The guidance clarifies that registrants must disclose any AI involvement in the creation process and demonstrate that human authorship played a determinative role in the work’s final form. Conversely, works generated entirely by AI—without meaningful human contribution—are ineligible for protection.
The Office’s approach reflects a pragmatic balance between fostering innovation and upholding copyright principles, though it remains to be seen how courts will apply these standards in disputes involving AI-generated content.
Beyond registration, the Office’s reports and guidance have broader implications for AI developers, copyright holders, and platform providers. For developers, the emphasis on human authorship may incentivize the creation of tools that facilitate collaborative human-AI workflows, where AI serves as an assistant rather than an autonomous creator. For copyright holders, the fair use analysis in AI training contexts could influence licensing negotiations and the development of opt-out mechanisms for copyrighted materials.
The Office’s multi-part report and registration guidance also highlight the need for further legislative or judicial clarification in areas such as digital replicas and the scope of fair use in AI training. While the Office has recommended federal legislation to address digital replicas, no such legislation has been enacted as of May 2025. Similarly, the fair use doctrine’s application to AI remains unresolved, with courts likely to shape its boundaries through litigation.
Required actions
- Disclose AI-generated material in registration applications; purely AI-generated content is not registrable, prompts alone do not confer authorship - Part 3 concludes that some training uses are fair use and others, especially those producing market substitutes, are not