Are Humans Required for Copyright Authorship? 

Mitchell Niles*

As the use of Artificial Intelligence (AI) has taken off in the past few years, people have voiced concerns regarding its encroachment on human creative agency.¹ Combined with the problems presented to our legal system,² this can create many issues for our creatives. Multiple lawsuits have been filed in recent years claiming copyright infringement against AI companies for training their Large Language Models (LLMs) on copyrighted materials.³ These companies, like Meta and Anthropic, have been able to avoid infringement through fair use.⁴ Using copyrighted works to train an LLM is “quintessentially transformative” to the original work.⁵ The same court has said that AI models are not replicating an author’s work using AI, but turning “a hard corner and creat[ing] something different,”⁶ and likening the training of LLMs on copyrighted works to an aspiring author who is training to write by reading the works of other authors.⁷ Such cases have allowed continued LLM training on inputs of copyrighted works, but have raised another question about their output: Can a non-human machine be considered an author of a creative work? 

Human Authorship of Creative Works 

The answer to this question comes from Thaler v. Perlmutter.⁸ Here, a computer scientist created a generative AI called the “Creativity Machine,” and produced an image using the model.⁹ The scientist applied for copyright registration of the image to the United States Copyright Office and listed the Creativity Machine as the author, with himself as the owner of the work.¹⁰ The Copyright Office denied the scientist’s application on the basis that the work lacked human authorship.¹¹ After the Copyright Review Board confirmed the denial, the scientist sought review from the courts.¹² At the district court level, the scientist gave multiple arguments in support of his position that the Creativity Machine should be a registerable author, including alternative arguments that he should be able to own the copyright under the work-made-for-hire doctrine of the Copyright Act of 1976, that allows someone who hires a creator to make the work to still become author of the work.¹³ These arguments failed and the district court confirmed the denial.¹⁴  

The United States Court of Appeals for the District of Columbia Circuit agreed with the district court, reasoning that multiple provisions of the Copyright Act of 1976 define authorship as distinctly human, such as describing potential authors’ “human lifespans, family members, domiciles, nationalities, mentes reae, or signatures.”¹⁵ The same human authorship is required under the work-made-for-hire doctrine.¹⁶ In the Copyright Act, Congress provided protections for non-human entities, like corporations, to own works made by creators, but these entities are considered the authors, not the actual authors that crafted the work.¹⁷ This is not to say that works generated using AI cannot be copyrighted, but they must retain human authorship.¹⁸ 

How Much Human Input is Required of an Author? 

Thaler is a clear-cut issue of authorship, but the case also brings up another lingering question of how much human input is required for a work to be considered authored by a human. At the passage of the Copyright Act, computers were considered dormant machines that required human input to create anything,¹⁹ but that may no longer be the case.²⁰ There is no bright-line rule for determining the level of human input required to make AI-generated works copyright eligible, but the Copyright Office has been trying to create such a distinction in the past few years.²¹  

With AI-generated materials, the Copyright Office will look to see whether such work is a “‘mechanical reproduction’ or instead of an author’s ‘own original mental conception, to which [the author] gave visible form.’”²² When an author gives just a prompt to an LLM to create a work and the creative parts of the work are determined and generated by the LLM, then it does not contain the traditional elements involved in human authorship.²³ There are other scenarios, where an author arranges AI-generated materials in a certain way that the work as a whole constitutes a work of original authorship.²⁴ In either case, copyright protections will only apply to those parts of the work that are human generated.²⁵ For example, in 2023 the Copyright Office allowed registration of a graphic novel that combined human-authored text with AI-generated images, but only the text could be copyrighted.²⁶ 

Based on these examples, it is not hard to see this becoming a gray area. Imagine you are authoring a book; if you used an LLM to help you generate ideas for that book, what will you be able to protect and what must you disclaim? How much of that book is your own mental conceptions given physical form and how much is a mechanical production? If you make an AI model generate a melody and lyrics for a song, what must you add and restructure for you to be the author of the song? In a world where we increasingly see AI-generated content in our commercials, on our phones, and in our artwork, one can imagine a future where the Copyright Office and the courts are parsing through AI prompts and running works through AI-detection software. 

1. See Ryan Watkins & Eran Barak-Medina, AI’s Influence on Human Creative Agency, 38 Creativity Rsch. J. 137, 137 (2026) (contending that when AI systems are successfully used in a profession, they will influence the creative agency of the humans using them).  ↩︎
2. A.B.A., Top Six AI Legal Issues and Concerns for Legal Practitioners (Nov. 14, 2025) https://www.americanbar.org/groups/law_practice/resources/law-technology-today/2025/ai-legal-issues-and-concerns-for-legal-practitioners/ %5Bhttps://perma.cc/226X-MJZH%5D. ↩︎
3. See Complaint at 3, Concord Music Group, Inc. v. Anthropic PBC, No. 3:26-cv-00880 (N.D. Cal. Jan. 28, 2026) (alleging Anthropic is training certain Claude AI models on copyright works of music publishers); see also Complaint at 2–4, UMG Recordings, Inc. v. Uncharted Labs, Inc., No. 1:24-cv-04777 (S.D.N.Y. June 24, 2024) (alleging that Udio created copies of copyrighted music to train its AI model without permission or compensation of the creators or publishers).  ↩︎
4. See Bartz v. PBC, 787 F. Supp. 3d 1007, 1034 (N.D. Cal. 2026); see also Kadrey v. Meta Platforms, Inc., 788 F. Supp. 3d 1026, 1060 (N.D. Cal. 2025).  ↩︎
5. Bartz, 787 F. Supp. 3d at 1022.  ↩︎
6. Id.  ↩︎
7. Id.  ↩︎
8. 130 F.4th 1039, 1041 (D.C. Cir. 2025).  ↩︎
9. Id.  ↩︎
10. Id.  ↩︎
11. Id.  ↩︎
12. Id. at 1044.  ↩︎
13. Id. at 1048; 17 U.S.C. § 201(b). ↩︎
14. Thaler, 130 F.4th at 1044, 1048.  ↩︎
15. Id. at 1045–47.  ↩︎
16. Id. at 1048.  ↩︎
17. Id.at 1048–49.  ↩︎
18. Id.  ↩︎
19. Id. at 1048.  ↩︎
20. See Gov’t Accountability Off., Science and Tech Spotlight: AI Agents (2025), https://www.gao.gov/assets/gao-25-108519.pdf [https://perma.cc/YNP7-ZLQU] (discussing the increase and potential of agentic AI models).   ↩︎
21. Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence, 88 Fed. Reg. 16190 (Mar. 16, 2023).  ↩︎
22. Id. (quoting Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53, 60 (1884)).  ↩︎
23. Id.  ↩︎
24. Id.  ↩︎
25. Id.  ↩︎
26. Id.  ↩︎

*Mitchell Niles, J.D. Candidate, University of St. Thomas School of Law, Class of 2027 (Associate Editor).