Generative AI has increasingly destabilized copyright standards, and its widespread use has prompted the U.S. Copyright Office to clarify how copyright law applies to works containing AI-generated material. The Office’s 2023 updated statement created a standard in which applicants would not receive copyright protection for “AI-generated content that is more than de minimis,” raising the question of whether the applicant’s contribution is sufficiently original to warrant copyright protection [1]. Although this framework may be relatively straightforward in genres where human and AI-generated contributions are readily separable, poetry presents a difficult case. Seemingly small writing decisions can alter a work’s effect. This characteristic poses an issue because Copyright Office examiners are not legally permitted to evaluate artistic merit, yet they must still judge whether the piece contains the required minimal originality [2]. Courts must therefore clarify the de minimis approach to minimize the exploitation of loopholes in copyright claims.
The Office’s current practices for evaluating works involving generative AI are relatively uncomplicated for creative genres other than poetry. As a foundational principle in copyright protection, a human being must author copyright-eligible work. Thaler v. Perlmutter (2025) reinforced this: Stephen Thaler, a computer scientist, argued that he deserved copyright protection for a visual artwork because he owned the AI system that created it. However, the U.S. Court of Appeals for the District of Columbia Circuit affirmed the rejection of Stephen Thaler’s application by stating that “the Copyright Act of 1976 requires all eligible work to be authored in the first instance by a human being” [3]. Thaler’s work-made-for-hire argument failed because the AI system had “no copyright to transfer” in the first place [4]. This reasoning also applies to works involving a mix of human and AI contributions. For instance, the Office granted Kristina Kashtanova limited copyright protection for her comic book Zarya of the Dawn, as Kashtanova received protection “covering only” the book’s human-authored text and arrangement of visual elements but not the AI-generated images [5]. Zarya shows that copyright can apply to human contributions in a creation that mixes AI and human work, even when other elements remain unprotected. In these instances, a clear distinction between human-created and AI-generated contributions lets Copyright Office examiners decide without considering artistic value. However, cases involving poetry are more challenging.
Poetry’s distinctive nature as a genre in which small, specific creative decisions can drastically change a piece complicates copyright evaluation. “The Red Wheelbarrow” by William Carlos Williams is a straightforward example—the poem consists of only 16 words, but its lineation, spacing, and punctuation choices are among its most expressive features [6]. For instance, Williams uses line breaks to split words such as “wheel / barrow” and “rain / water,” which enables readers to slow down and appreciate simplicity, and the lack of punctuation marks invites readers to savor each word and read at an intimate pace of their choosing [7]. What should a copyright reviewer do, then, when AI partially generates a poem's text while a human decides on line breaks and punctuation? Similarly, a single repeated line in a poem can have a non-trivial impact on its effectiveness. For instance, in “Stopping by Woods on a Snowy Evening,” by Robert Frost, the concluding line “And miles to go before I sleep” appears twice, emphasizing a heavy feeling of resignation and weariness that would not have been as compelling had Frost not included the repetition [8]. This demonstrates the difficulty of evaluating minor poetic modifications. The contributions in Zarya of the Dawn were not difficult to evaluate in that the images and text within a graphic novel could be easily separated, but for a poem, the issue becomes whether the few human-authored words, considered collectively, make the resulting expression original. This raises a dilemma for the Office: how can Copyright Office examiners determine whether such small modifications constitute sufficient original expression, all while remaining neutral on aesthetic merit?
The doctrine of aesthetic nondiscrimination creates an undesirable application loophole. As established in Bleistein v. Donaldson Lithographing Co. (1903), the doctrine rests on the principle that it would be a “dangerous undertaking for persons trained only to the law to constitute themselves final judges of the worth of pictorial illustrations” [9]. This ensures that reviewers do not deny copyright protection based on subjective taste. However, in Feist Publications, Inc. v. Rural Tel. Serv. Co. (1991), the Supreme Court emphasized that the originality prerequisite for copyright requires a “modicum of creativity” rather than “sweat of the brow” effort [10]. Potential issues arise when both Bleistein and Feist apply to scenarios involving a mix of AI and human work. Under Feist, an applicant modifying AI-generated text can argue that their personal minor edits satisfy the requisite “modicum of creativity,” thus making the piece sufficiently original, but under Bleistein, copyright reviewers face the challenge of refraining from judging aesthetic quality when deciding whether it meets that requirement. Applicants could exploit this tension by claiming copyright protection for predominantly AI-generated works on the basis of minimal human edits.
Closing this loophole requires careful deliberation of potential approaches. One tempting resolution to this dilemma is amending the Copyright Act of 1976 to include poetry-specific standards. However, the Act’s broad and flexible phrasing is its strength, as it allows for versatility amid unpredictable societal and technological shifts; courts can adapt to changes, and interpretations can evolve. Additionally, updating the de minimis standard with numerical metrics, such as requiring a fixed percentage of human-authored words, is not practical because methods of human expression are impossible to quantify.
Instead, the most feasible solution is for courts to clarify how Feist would be applied to AI-assisted poetry within the boundaries of Bleistein. In an appropriate case, a court could establish precedent for copyright reviewers by basing its decision solely on whether the modifications constitute independently created expression that satisfies Feist’s minimal creativity requirement, rather than on whether a human’s modifications to an AI-generated poem enhance its aesthetic quality. This would avoid violating Bleistein’s principle of aesthetic nondiscrimination. To encourage courts to qualitatively clarify the boundaries of copyright for poetry in this manner, interested organizations can express their concerns in amicus curiae briefs—filings submitted by someone not a party to the case but who is concerned with its outcome—to courts taking on matters related to AI and copyright [11]. Moreover, just as Thaler tried to push courts toward a pro-AI route through a lawsuit, organizations can encourage experienced litigators to identify and take on promising cases that might move the law in a positive direction.
Because poetry tends to rely on intentional yet subtle creative decisions such as enjambment and punctuation, it complicates the Office’s de minimis standard for AI contributions in copyrightable material when taking into account the originality and aesthetic nondiscrimination doctrines. The Office should recognize the possibility that applicants may take advantage of the paradox to exploit the copyright application process. To address poetry’s copyright dilemma, courts must therefore clarify the application of Feist and Bleistein by directing copyright reviewers to assess whether human contributions reflect independently written creative expression. Long-term judicial engagement, strategic litigation, and amicus filings are methods that organizations can use to guide courts toward establishing relevant precedents. The resolution of this issue will ensure copyright law can better fulfill its duty to incentivize human creativity and revitalize poets amid a time of serious uncertainty.
Footnotes
[1] U.S. Copyright Office, “Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence,” 88 Fed. Reg. 16190, 16194 (March 16, 2023): 16190-94, https://www.federalregister.gov/documents/2023/03/16/2023-05321/copyright-registration-guidance-works-containing-material-generated-by-artificial-intelligence.
[2] David Fagundes, “Who Is the Bad Art Judge? Aesthetic Nondiscrimination in Andy Warhol Foundation v. Goldsmith,” NYU Law Review (December 28, 2023), https://nyulawreview.org/forum/2023/12/who-is-the-bad-art-judge-aesthetic-nondiscrimination-in-andy-warhol-foundation-v-goldsmith/.
[3] Thaler v. Perlmutter, 130 F.4th 1039, 1041 (D.C. Cir. 2025), https://law.justia.com/cases/federal/appellate-courts/cadc/23-5233/23-5233-2025-03-18.html.
[4] Ibid.
[5] U.S. Copyright Office, Letter to Zarya of the Dawn (Registration # VAu001480196), February 21, 2023, https://copyright.gov/docs/zarya-of-the-dawn.pdf.
[6] William Carlos Williams, “The Red Wheelbarrow,” Poetry Foundation, https://www.poetryfoundation.org/poems/45502/the-red-wheelbarrow.
[7] Ibid.
[8] Robert Frost, “Stopping by Woods on a Snowy Evening,” Poetry Foundation, https://www.poetryfoundation.org/poems/42891/stopping-by-woods-on-a-snowy-evening.
[9] Bleistein v. Donaldson Lithographing Co., 188 U.S. 239 (1903), https://supreme.justia.com/cases/federal/us/188/239/.
[10] Feist Publications, Inc. v. Rural Tel. Serv. Co., 499 U.S. 340 (1991), https://supreme.justia.com/cases/federal/us/499/340/.
[11] “Amicus,” Wex, Legal Information Institute, Cornell Law School, last reviewed February 2025, https://www.law.cornell.edu/wex/amicus.