By: Ryan Zaitoun
As sophisticated artificial intelligence (“AI”) tools produce outputs that are increasingly mistaken for human creations, the efficacy of copyright enforcement and the value of the human creativity it protects have been called into question.[1] In January 2025, the Copyright Office published Part II of its report on AI-generated content and its copyrightability, determining that existing copyright law can address AI-generated works and allowing AI-assisted creations to be copyrightable but only with sufficient human creative control.[2] Further, the Copyright Office found no significant policy incentive to afford copyright to purely AI-generated content, as copyright’s purpose is to incentivize human innovation and artistic pursuit.[3]
Surprisingly, this distinction isn’t necessarily new. In 1965, International Business Machines Corporation employee, Michael Noll, successfully copyrighted what became the first instance of copyrightable computer-generated art in his “Gaussian Quadratic.”[4]After the initial denial of his copyright registration, the Copyright Office eventually accepted Noll’s copyright because, though the output of the image was seemingly random, the program used to create the image remained entirely mathematical, and Noll retained some creative control over aspects of the output.[5] Modern generative AI systems, however, operate differently. They use an iterative process to compare words in their prompts with trained sources to generate results that can differ infinitely, even with the same prompt.[6] This lack of human creative control over the output led the Copyright Office to revoke copyright for the images in a comic series called “Zarya of the Dawn” in 2023; however, it only did so after the author discussed using Midjourney AI to generate the images on social media.[7]
As AI tools improve, situations like with Zarya pose a serious risk to copyright law’s effectiveness if enforcement bodies cannot successfully distinguish entirely AI-generated works from AI-assisted or human-created content.[8] This is especially true given how rarely AI users report their own AI use and how AI detection systems often struggle.[9] Current Copyright Office practices require registrants to disclose the use of AI in submitted materials, and knowingly failing to do so constitutes a criminal offense that can trigger copyright revocation.[10] Generally, these negative consequences can be incentive-enough for risk-averse parties to disclose all known AI use; however, recent studies continue to show chronic under-reporting of AI use across professions.[11] For example, a 2025 study revealed that as much as sixteen point nine percent of academic submissions were flagged by AI detectors, yet only one point seven percent of submissions reported AI usage.[12] The study noted stigma against AI use in academia and concerns about authorial reputation as barriers to the reporting requirement, while similar findings have been noted by business journals in office contexts.[13] Additionally, public opinion against AI in art provides further commercial incentive for businesses to minimize their reported AI use.[14] This public stigma leaves businesses and authors with a difficult choice: disclose AI use, possibly leaving their work in the public domain and losing potential revenue due to negative public opinion, or remain silent about AI use, which may leave them criminally liable.[15]
To keep copyright meaningful in a future proliferated with human-passing AI-generated content, there should be stronger incentives for authors to publicly disclose AI use in their works. Previous scholarship on this topic discussed introducing certifications for human-created works, similar to labels used to identify “Organic” or “Fairtrade” goods, to give consumers informed choices and to incentivize authors to label how their work was made.[16] In addition to requiring AI-use disclosures in copyright registrations, the Copyright Office should seek to reward registrants for their transparency, as certification systems in the private sector already do.[17] One approach to implementing this system could be providing registrants with tiered copyright labels based on the extent of AI involvement in creating a work, similar to one suggested for academic publications.[18] This system may help positively reinforce voluntary reporting and responsible AI use by distinguishing these works from uncertified, entirely AI-generated content, and, like with other certification regimes, enable authors to receive greater revenue.[19] Overall, by recognizing copyright protections for some AI-assisted material in its 2025 report, the Copyright Office is well positioned, as both a regulatory body and a public voice of authority, to lead the normalization of responsible AI use and reporting.[20]
[1] See Judy Wang & Nicol Turner Lee, AI and the Visual Arts: The Case for Copyright Protection, Brookings (Apr. 18, 2025), https://www.brookings.edu/articles/ai-and-the-visual-arts-the-case-for-copyright-protection/ [https://perma.cc/RR6M-BB63] (noting that the Copyright Office’s inability to distinguish between human-authored and AI-authored work is worrisome and creates difficulties in enforcing its own guidance).
[2] See U.S. Copyright Off., Report on Copyright and Artificial Intelligence: Part 2: Copyrightability, 1, 41 (2025) [hereinafter The Copyright Office Report], https://www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-2-Copyrightability-Report.pdf [https://perma.cc/89EG-LD3J] (“[T]he Copyright Office concludes that existing legal doctrines are adequate and appropriate to resolve questions of copyrightability. Copyright law . . . can enable case-by-case determinations as to whether AI-generated outputs reflect sufficient human contribution to warrant copyright protection.”).
[3] Id. at 36; see also Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence, 88 Fed. Reg. 16190, 16191 (Mar. 16, 2023) (“Copyright can protect only material that is the product of human creativity.”).
[4] See Grant Taylor, Routing Mondrian: The A. Michael Noll Experiment, 8 Media-N: J. New Media Caucus (2012), https://median.newmediacaucus.org/routing-mondrian-the-a-michael-noll-experiment/ [https://perma.cc/Z3AM-3F98].
[5] See id.; Gaussian-Quadratic, ZKM Ctr. Art & Media Karlsruhe, https://zkm.de/en/artworks/gaussian-quadratic [https://perma.cc/QQ87-HWMX] (last visited Oct. 9, 2026) (describing how Noll iteratively tweaked the program until it created a picture he desired).
[6] Letter from U.S. Copyright Office, to Van Lindberg, Re: Zarya of the Dawn (Registration # VAu001480196), 5–8 (Feb. 21, 2023), https://cdn.arstechnica.net/wp-content/uploads/2023/02/AI-COPYRIGHT-decision.pdf [https://perma.cc/3TTG-9NXW]; The Copyright Office Report, supra note 2, at 7.
[7] Letter from U.S. Copyright Office, to Van Lindberg, Re: Zarya of the Dawn (Registration # VAu001480196), 2–3, 9 (Feb. 21, 2023), https://cdn.arstechnica.net/wp-content/uploads/2023/02/AI-COPYRIGHT-decision.pdf [https://perma.cc/3TTG-9NXW].
[8] See Jacob Noti-Victor, Regulating Hidden AI Authorship, 111 Va. L. Rev. 139, 205 (2025).
[9] See e.g., Ahmad S. BaHammam, The Transparency Paradox: Why Researchers Avoid Disclosing AI Assistance in Scientific Writing, 17 Nat. & Sci. Sleep 2569, 2569 (2025); Shilin Yan et al., A Sanity Check for AI-Generated Image Detection, Int’l Conf. on Learning Representations 1, 9 (2025) (finding the accuracy of various AI image detectors to be between around 55% to 65%).
[10] See Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence, 88 Fed. Reg. 16190, 16192 (Mar. 16, 2023); 17 U.S.C. § 506(e) (“Any person who knowingly makes a false representation of a material fact in the application for copyright registration . . . shall be fined not more than $2,500.”).
[11] See BaHammam, supra note 9, at 2569; Jordan Claes, Canadian Workers Experiencing AI Guilt on the Job, Emp. Hero (Jul. 6, 2026), https://employmenthero.com/en-ca/news/canadian-workers-ai-guilt-workplace-paradox/ [https://perma.cc/F8QA-CVME].
[12] See BaHammam, supra note 9, at 2569.
[13] See id. at 2570; Eric Anicich & Jeslyn Brouwers, Why Employees Aren’t Transparent About their AI Usage, Harv. Bus. Rev (June 10, 2026), https://hbr.org/2026/06/why-employees-arent-transparent-about-their-ai-usage (on file with the American University Business Law Review); see also Claes, supra note 11 (finding that nearly 34% of individuals surveyed admitted hiding AI use from their employers).
[14] See Noti-Victor, supra note 8, at 154.
[15] Id. at 139; see 17 U.S.C. § 506(e).
[16] See Noti-Victor, supra note 8, at 192-193.
[17] See, e.g., Lucy Buchholz, Fairtrade: US Consumers Prioritise Supply Chain Transparency, Sustainability Mag. (Sep. 8, 2023), https://sustainabilitymag.com/articles/fairtrade-us-consumers-prioritise-supply-chain-transparency [https://perma.cc/JR95-6VS3].
[18] BaHammam, supra note 9, at 2572 (describing a four-tiered system, the first tier not requiring disclosure if AI use was for minor proof reading and grammar; the second tier only requiring simple acknowledgement of AI use if it was used for editing and stylistic input; the third tier requiring a detailed disclosure if AI was used for drafting language and brainstorming; and the fourth tier requiring a comprehensive disclosure of AI use if there was extensive use of AI throughout the work’s creation).
[19] See id.; Noti-Victor, supra note 8, at 194 (“Studies have shown that consumers are more willing to pay more for goods and services that carry desirable certifications . . . .”).
[20] The Copyright Office Report, supra note 2, at 41.
