Copyright Office Advises on Copyrightability of Generative AI Works

On January 29, the U.S. Copyright Office published the second part of a planned three-part report on copyright and artificial intelligence (AI), this time focused on the question of copyrightability for AI-generated creative works. The first part, published in July 2024, explored the legality of so-called digital replicas of individuals’ likenesses, or “deepfakes.” The report is the product of a sweeping new initiative on AI launched by the Copyright Office in 2023 in response to the first crop of copyright registrations for works containing AI-generated expressive elements.

The Copyright Clause vests in Congress the authority to “secur[e] for limited times to authors . . . the exclusive right to their . . . writings.” In Community for Creative Non-Violence v. Reid (D.C. Cir. 1988), the U.S. Supreme Court explained that an author is “the person who translates an idea into a fixed, tangible expression entitled to copyright protection” (emphasis added).

In a preliminary statement of policy preceding the report’s publication, the Copyright Office confirmed the requirement of human authorship for obtaining copyright protection; applicants intending to register a work containing more than a de minimis amount of AI-generated material must disclose that fact and to describe their own human contribution to the ultimate work. This comports with the requirement of human authorship for copyright protection (see the Ninth Circuit denying copyright registrations to a monkey in Naruto v. Slater, 888 F.3d 418 (9th Cir. 2018) and a non-human “spiritual being” in Urantia Found v. Kristen Maaherra, 114 F.3d 955 (9th Cir. 1997)).

In the new report, the Office clarifies the circumstances in which a human author is eligible for copyright in a work containing AI-generated expression, but ultimately refuses to endorse a bright-line test, relegating the issue to fact-specific, case-by-case determination by federal courts. The Office imagines a sliding scale of human control on which every work containing AI-generated elements may be placed. On one extreme are works that are wholly the expressive output of an AI, wherein the AI is responsible for the “spark of creativity” necessary for copyright to attach. On the other side are more “assistive uses” of AI, such as de-aging actors or digitally excising an object or person from a photograph, which merely enable a human author to create the final product they already have in mind.

The ultimate question is a familiar one: Is the work “basically one of human authorship, with the computer merely being an assisting instrument,” with the traditional elements of authorship — selection and arrangement of its component elements — conceived by a natural human, or is the computer responsible for the work’s conception?[1]

Today’s popular generative AI systems can create everything from text (ChatGPT, DeepSeek) to entire images (Midjourney, DALL-E) after receiving only a few words of natural language prompting. Users can describe their desired output with varying specificity, instructing the AI to create an image of a certain subject or topic, in a particular visual style or by applying a distinct visual technique. Once an output has been received, the user can iterate and reiterate on the work by revising the prompt, adding, removing, or clarifying instructions as needed until the user is satisfied with the end-product.

One troubling aspect of the prompting process, according to the report, is that prompting is an unpredictable process deficient in the necessary element of human control: A user could input identical prompts on two separate occasions and receive completely different outputs. The AI may choose to disregard certain instructions, or it may inexplicably add undesired elements that were never triggered by human prompting. If the AI is responsible for the final creative interpretation of the user’s text input, and indeed exercises some “creative” judgment in terms of arrangement and selection of an image’s elements, the question of authorship becomes a murky one.

The report provides at least one concrete conclusion: Prompts alone do not provide sufficient human control to make generative AI users the author of an output for copyright purposes, at least given today’s available technology. Whether an output is copyrightable depends on the nature and extent of a human’s contribution beyond mere prompting, and whether that contribution qualifies as authorship of the output’s expressive elements. The Office confirmed, however, that the actual text of the prompts remains copyrightable just like any other human-generated expression, provided it meets the requisite level of creativity.

The report also makes clear that human authors may claim copyright of a work that incorporates some wholly AI-generated expressive elements if the human author was personally responsible for selecting, coordinating, and (re)arranging the AI-generated material in a creative way, with the copyright extending to the creative selection and arrangement of those elements. Some AI programs, like Midjourney, actually allow users to select and regenerate regions of a generated image with a modified prompt; the Office believes that some works created in such a manner will meet the minimum standard of originality.

2025 is sure to be a whirlwind for the regulation and growing acceptance of AI. As the generative AI landscape continues to evolve, Troutman Pepper Locke is your resource for understanding the potential risks and opportunities associated with the new technology.

[1] U.S. Copyright Office, Sixty-Eighth Annual Report of the Register of Copyrights for the Fiscal Year Ending June 30, 1965 (1966), https://www.copyright.gov/​reports/​annual/​archive/​ar-1965.pdf.