Copyrightability of AI-generated materials and U.S. Copyright Law

The United States Copyright Office (Copyright Office) has released its latest report on the copyrightability of outputs generated by artificial intelligence (AI) systems. This report is the second in a three-part series examining legal and policy issues related to AI and copyright law, and aims to clarify the extent of human contribution required for copyright protection of AI-generated outputs. The first report, published in July 2024, discussed whether current legal frameworks adequately protect against the unauthorized distribution of digital replicas. The third part, expected later this year, will address the legal implications of training AI models on copyrighted works and licensing considerations.

The Copyright Office reiterated its longstanding principle that human authorship is essential for claiming copyright protection; materials generated solely by non-humans cannot be copyrighted. The Copyright Office also took the view that it does not see the need for new legislation to resolve questions of copyrightability.

The report affirmed that using assistive AI systems to enhance or augment human expression preserves the underlying copyright protection. However, when an AI system makes expressive choices or entirely generates the work, the resulting product is not eligible for copyright protection.

When discussing whether the prompts given to generative AI systems are copyrightable, the Copyright Office explicitly rejected this on the premise that the prompts are instructions that convey unprotectable ideas and, at least based on the technology available today, do not provide a user with sufficient human control over the conversion of the ideas into fixed expression. This conclusion remained consistent even if a user repeatedly revised prompts on the basis that “the time, expense, or effort involved in creating a work by revising prompts is irrelevant, as copyright protects original authorship, not hard work or “sweat of the brow”.” However, this analysis changes where a user inputs its own copyrightable work into the prompt and that work is perceptible in the output. In such instances, the user can claim authorship of at least that portion of the output which was perceptible. 

The Copyright Office also analyzed whether modifying the AI-generated materials would allow for the work to be copyright protected. It concluded that copyright protection may be afforded to the user’s “selection, coordination, and arrangement of text and visual elements” if the resulting work constitutes a protectable work. The determination of whether the human contribution meets the test of authorship required under copyright law will continue to be determined on a case-by-case basis. Similarly, where AI-generated content was incorporated in or included within a larger human-authored work, the copyrightability of the larger human-authored work would not be impacted, and only the AI-generated elements would not be protectable on their own.

In summary, the key takeaways from the Copyright Office’s latest report are:

• Prompts used for generative AI systems are not copyrightable, even if users are revising prompts extensively and there is effort, time, or cost involved.
• Inputting existing copyrightable work into a prompt that makes such works perceptible in the output, would allow the user to claim authorship over that specific part of the output.
• Human contributions to AI-generated materials may be copyrightable, depending on a case-by-case determination of whether the underlying authorship tests are met.
• When AI-generated content is part of a larger, human-authored work, the overall work’s copyrightability remains intact, but the AI-generated elements themselves may not be protected.