---

NNAI Livescu sponsors undergraduate research projects on various topics, overseen by faculty and staff in the initiative.

Please note that this is a working version and it will be updated after the research project is completed.

By Isabella Yuan, 4th-year UCLA Human Biology and Society Major and Philosophy Minor

AI, Copyright Law, and Work-Made-For-Hire

I. INTRODUCTION

On November 3, 2018, computer scientist and inventor Stephen Thaler filed an application with the United States Copyright Office (USCO) to register a copyright claim on a two-dimensional artwork titled “A Recent Entrance to Paradise” (“Recent Entrance”). In the application, Thaler claimed that a machine learning system named the “Creativity Machine” had autonomously generated the artwork. For this reason, Thaler listed the Creativity Machine as the sole author of the artwork.

Thaler’s application was ultimately rejected by the USCO, which cited its policy that, to be eligible for copyright registration, a work must be authored in the first instance by a human being.^{[i] [ii]} While Thaler’s attempt to register the Creativity Machine as an author in a copyright claim was ultimately unsuccessful, it has raised questions regarding the possibility of copyrighting works created exclusively with artificial intelligence (AI).

Driven by analysis of Thaler’s unsuccessful case in favor of extending copyright protection to autonomous AI-generated works, I raise the following three questions: (1) is AI-generated work copyrightable; (2) how might AI-generated work become eligible for copyright protection; and (3) should AI-generated work be copyrightable?

II. BACKGROUND

Copyright is a form of intellectual property that protects original works of authorship that are fixed in any tangible medium of expression.^[iii]The holder of a copyright has the exclusive right to reproduce, distribute, perform, display, or prepare derivative works based on the copyrighted work.^[iv] For most works created after 1978, copyright protection lasts for the life of the author plus 70 years.^[v] The USCO, a department within the Library of Congress, is in charge of administering U.S. copyright law. Its duties include registering copyright claims, maintaining public records of copyright claims, and developing regulations concerning copyright law.^[vi]

U.S. copyright law is grounded in the U.S. Constitution’s Intellectual Property Clause, which gives Congress the authority to enact laws that “promote the Progress of Science and useful Arts, by securing for limited Times to Authors and Inventors the exclusive Right to their respective Writings and Discoveries.”^[vii] The use of the term “author” in the Intellectual Property Clause has formed the basis of subsequent analysis and interpretations of ownership in copyright law. 

Currently, U.S. copyright law is primarily governed by the Copyright Act of 1976. According to the language of the Act, the author is generally the owner of the copyright except in specific cases, such as when the author assigns the copyright to another party, or when the author creates the work under a work-made-for-hire (WMFH) arrangement––when the work is created by an employee within the scope of their employment or when it is specially commissioned under a written agreement.^[viii] According to the WMFH doctrine, the employer or commissioning party is considered the legal author and copyright holder of the work.

III. IS AI-GENERATED WORK COPYRIGHTABLE?

After his application to register a copyright claim for “Recent Entrance” was denied, Stephen Thaler filed a lawsuit against the USCO and its director, Shira Perlmutter. In the lawsuit, Thaler contested the USCO’s requirement that the author of a work be human for the work to be eligible for copyright protection, arguing that the Copyright Act does not explicitly define an “author” to be human and that the human-authorship requirement is unconstitutional and unsupported by case law.^[ix]

Additionally, Thaler argued that he qualified as the legal author of “Recent Entrance” through the WMFH doctrine, claiming that the Creativity Machine functionally acted as his employee in generating the artwork. The WMFH doctrine allows non-human entities to be considered “authors”: corporations can be legal authors under the WMFH doctrine despite not being natural persons. For example, Microsoft holds copyrights to software developed by its employees during and as part of their employment. The corporation is thus considered the legal author, even though the actual creator may be an individual employee. Thaler essentially argued that, as the owner and developer of the Creativity Machine, the work was generated within the scope of a relation analogous to employment. He concluded that, pursuant to the WMFH doctrine, the resulting output should be attributed to him as the legal author.

The Court affirmed the USCO’s decision to deny Thaler’s application. In a Memorandum Opinion, Judge Howell addressed Thaler’s first argument––that the Copyright Act does not explicitly require that an “author” be human––by first conceding that the term “author” is undefined in the statute and that the definition of “author” has historically been adapted in response to technological advances.^[x] For example, Judge Howell cited Burrow-Giles Lithographic Co. v. Sarony (1884), in which the Supreme Court ruled that the term  “author” could extend to a photographer because the photograph was understood to represent the “original intellectual conceptions of the author.”^[xi]

However, despite acknowledging that the definition of “author” has historically evolved alongside technological advances, Judge Howell noted that the understanding that the legal author must be human is based on long-standing precedent.^[xii] She also emphasized that copyright law was defined to incentivize human intellectual labor, and that non-human entities like AI do not require or respond to such incentives.^[xiii] Based on this conclusion that human authorship is required for copyright protection under the Copyright Act, Judge Howell rejected Thaler’s second argument––that he qualified as legal author through the WMFH doctrine––by pointing out that because the work itself was not created by a human, it could not constitute a copyrightable work.^[xiv] As a result, Thaler still would not qualify as the legal author.

Thaler then appealed this decision in the U.S. Court of Appeals for the District of Columbia, which upheld the USCO’s decision to deny registration of the copyright claim for “Recent Entrance.” In the Opinion, Judge Millett affirmed the view that human authorship is a prerequisite for copyright protection, adding that the Copyright Act makes sense only if the term “author” is understood to mean a human author.^[xv]  Judge Millett also elaborated that the human authorship requirement does not call for a universal denial of copyright protection for works involving AI, noting that, depending on the situation, the USCO had previously approved registrations for works where human authors made use of AI.^[xvi] For example, if a human has significantly modified AI-generated material in a sufficiently creative way, the work may be eligible for copyright protection. However, in such a case, only the modifications made by the human would be protected––not the AI-generated material itself.^[xvii]

According to Judge Millett, the main issue with Thaler’s application was that Thaler was claiming that the work had been created entirely by AI without any human input. Thaler’s WMFH argument would have worked only if a human creator was the actual author of the work and created it under a WMFH framework. Since that was not the case––the work was (according to Thaler) created entirely by AI––there was no authorship to attribute, and Thaler could not be considered the author under the statute.

The Thaler case reaffirmed the longstanding principle that purely AI-generated works are not eligible for copyright protection under U.S. copyright law. This position has remained unchanged since.           

IV. HOW MIGHT AI-GENERATED WORK BECOME ELIGIBLE FOR COPYRIGHT PROTECTION?

Though Thaler was ultimately unsuccessful in registering a copyright claim for an AI-generated work, his case raised interesting implications regarding legal authorship and legal personhood. While the U.S. District Court for the District of Columbia and the U.S. Court of Appeals for the District of Columbia both affirmed the USCO’s decision, Thaler’s point that the Constitution does not explicitly require that an “author” be human still stands. It is still plausible (though unlikely) that the USCO may revise its guidelines to allow for copyright registration of AI-generated works, or that a court may rule that human authorship is not necessary for copyright registration under the Copyright Act. As it is still possible for future legal developments to expand the definition of legal authorship to include AI-generated works, it is worth considering how current legal frameworks could accommodate such works if the human authorship requirement were lifted.

One path to eligibility could be under the WMFH doctrine, as argued by Thaler. By the WMFH doctrine, the employer is considered the legal author of any eligible works produced by an employee within the scope of their employment.^[xviii] To return to the Microsoft example, if a Microsoft employee were to develop an AI system as part of their employment, Microsoft would be the legal author and owner of the AI system. The question then is, if that AI were to autonomously build software, could Microsoft claim copyright ownership and legal authorship under the WMFH doctrine?

When Thaler made the WMFH argument in his case against the USCO, his argument failed not only because of the human-authorship requirement but also because he could not demonstrate that (i) the AI qualified as an employee within existing legal definitions, and (ii) a valid employment existed between the AI and himself as the purported employer.^[xix] These limitations stemmed from the fact that, under current law, AI systems are not recognized as legal persons capable of entering employment relationships.

Based on Thaler’s case, one way for AI-generated works to become eligible for copyright by the WMFH doctrine is to have AI recognized as a legal entity. This would involve granting the AI some form of legal personhood––a status granted by law to an entity that allows it to have rights and obligations like a human being.^[xx] The primary example of attributing personhood status to non-human entities is the concept of corporate personhood, through which corporations are able to sue, be sued, own property, enter into contracts, and much more despite lacking human consciousness, physical embodiment, or other markers of personhood. Considering that the legal system already recognizes non-human entities like corporations as legal persons, it is plausible that AI could one day be similarly recognized as a legal entity. Recognizing AI as a legal entity would thus align with existing precedent of assigning legal personhood to non-human entities.

One might argue that just because one non-human entity has legal personhood does not mean that AI should be entitled to legal personhood. Their situations are not entirely analogous. After all, corporations can be seen as collections of human stakeholders, and their legal personhood can be seen as deriving from the rights and responsibilities of the humans who comprise them.^[xxi] The same cannot be said of autonomous AI systems, which are theoretically non-sentient and would thus lack intent and accountability in a way that a corporation (or a corporation’s constituents) would not.

This does not present a valid concern. Corporations are not necessarily collections of human stakeholders. For example, decentralized autonomous organizations (DAOs) are blockchain-based entities that embed AI and automation into their operations and corporate structure extensively, such that they can operate with minimal human intervention.^[xxii] In 2021, Wyoming became the first state to recognize DAOs as a form of Limited Liability Company (LLC). Tennessee and Vermont have since recognized DAOs as legal business entities, granting them legal status as well. Therefore, corporations do not necessarily need to involve natural persons to qualify for legal personhood––legal personhood is not contingent on human involvement. As a result, it is possible that if AI is granted legal personhood, it could (i) qualify as an employee and (ii) enter into a valid employment relationship, fulfilling the prerequisites of Thaler’s WMFH argument.

Even if (i) and (ii) were fulfilled, however, it is realistically unlikely that the WMFH doctrine in particular would be applied to extend copyright protection to AI-generated works because AI does not fit neatly into existing understanding of an employee in labor law (hence the need to grant it some form of legal personhood in order for it to fulfill the WMFH doctrine.) It is more likely that the courts or the legislature would simply grant copyright ownership directly to the entity that deploys the AI, rather than going through the extra step of vesting the AI with legal personhood to satisfy the requirements of the WMFH doctrine. However, even if AI is not directly applied, the WMFH doctrine can still be relevant for the purpose of considering whether AI-generated works can be copyrighted: it demonstrates that an entity can be considered an author without contribution expression, as long as that entity is providing some form of direction, initiative, or economic investment in the creation process.^[xxiii] The WMFH doctrine thus establishes that copyright can be grounded in economic relationships.

Regardless, even if the rationale behind the WMFH doctrine can be extended to AI, it does not mean that AI is actually capable of creating “original” works in the sense required by copyright law in the first place. On the one hand, weaker AI such as large language models (LLMs) or image generators are trained on large datasets to recognize patterns and generate outputs based on statistical correlations rather than genuine understanding. Consequently, these types of AI thus lack intentionality; they function more as tools than independent agents.

Regardless, even if the rationale behind the WMFH doctrine can be extended to AI, it does not mean that AI is actually capable of creating “original” works in the sense required by copyright law in the first place. On the one hand, weaker AI such as large language models (LLMs) or image generators are trained on large datasets to recognize patterns and generate outputs based on statistical correlations rather than genuine understanding. Consequently, these types of AI thus lack intentionality; they function more as tools than independent agents.

On the other hand, stronger AI systems––particularly the hypothetical artificial general intelligence (AGI), which would possess human-like cognitive capabilities––could potentially generate entirely original works in the sense required by copyright law. If stronger AI systems were to demonstrate intentionality in their outputs, they could challenge current legal standards of authorship and originality. However, simulating human cognition does not guarantee that stronger AI systems would exhibit genuine intentionality. Thus, even if granting AI legal personhood allows it to meet the prerequisites of the WMFH doctrine, further questions regarding the nature of creativity and authorship must be addressed before fully AI-generated works can be eligible for copyright protection.

V. SHOULD AI-GENERATED WORK BE COPYRIGHTABLE?

Even if AI could qualify to create copyrightable works, that does not mean that AI-generated works should be granted copyright protection. The normative question remains: what is copyright meant to protect, whom is it designed to serve, and what are the consequences of extending protection to AI-generated works?

V.1  Incentive Theory

Incentive theory is a central justification for copyright protection. According to incentive theory, copyright protection incentivizes creators to create by providing them exclusive rights to their intellectual property. Compared to monetary compensation, copyright protection provides further incentive for creators to innovate in the context of the inherent underproduction problem associated with non-rivalrous and non-excludable goods: essentially, when people can easily access a good without paying, creators have less financial incentive to innovate and create those goods.^[xxiv]

Copyright protection for AI-generated work cuts both ways when considering incentive theory. On one hand, by granting copyright protection to AI-generated works, developers and corporations may be incentivized to invest more time and money into developing AI systems. However, on the other hand, granting copyright protection to AI-generated works might disincentivize human creatives as their labor may be devalued in a market saturated with AI-generated content.

V.2  Market Monopolization

Allowing AI-generated work to be copyrighted may lead to corporations using AI to mass-produce creative works and secure copyrights for those works. This could lead to a select few dominant tech companies monopolizing creative markets. This would undermine the original purpose of copyright: to encourage individual authorship and innovation.

V.3  Limited Liability

Limited liability is a legal principle that protects shareholders or members from being held personally liable for debts incurred by a business entity.^[xxv] Limited liability stems from corporate personhood: when corporations are treated as separate legal entities from their owners, the owners are shielded from direct responsibility for corporate actions.^[xxvi] Similarly, extending copyright to AI-generated works by granting AI legal personhood could shield developers from full legal responsibility for the outcomes of their creations as adverse outputs could be attributed to the AI rather than the developer. Therefore, granting copyright to AI-generated works might lead to a form of limited liability in which human actors can benefit from the output of AI without bearing the full legal consequences of its contents. After all, the AI has no use for royalties or reputation, but it would still produce new creative works. This may lead to a system that deprioritizes accountability and incentivizes developers and companies to pursue riskier projects.

At the end of the day, the language of the Constitution is ambiguous enough to accommodate new interpretations. In the face of this possibility, it is important to consider whether extending copyright protection to AI-generated works aligns with the original intent of copyright: to “promote the Progress of Science and useful Arts.”^[xxvii] Ultimately, given that extending protection would devalue and most likely disincentivize human creatives, from a consequentialist standpoint, purely AI-generated works should not be eligible for copyright protection.

VI. CONCLUSION

As AI technology continues to advance, the legal and ethical questions surrounding copyright ownership and authorship will only become more pressing. The Thaler case has made it clear that under current U.S. copyright law, purely AI-generated works are not eligible for copyright protection. However, as the boundaries of legal status and personhood continue to shift, it is important to keep an eye on how copyright law might reinterpret AI authorship accordingly. Ultimately, decisions about the copyrightability of AI-generated works raise foundational questions about the purpose of copyright law and how it balances between promoting innovation and protecting creators. Given the risk of devaluing human creativity and enabling corporations to monopolize creative output, AI-generated works should not be copyrightable.

References

1. – Thaler v. Perlmutter, No. 23-5233 (D.C. Cir. 2025)
2. The Library of Congress. United States Copyright Office. Compendium of U.S. Copyright Office Practices § 101. (3 ed, 2021). https://www.copyright.gov/comp3/docs/compendium.pdf
3. 17 U.S.C. § 102(a)
4. 17 U.S.C. § 106
5. 17 U.S.C. § 302
6. “Overview.” copyright.gov. Accessed May 12, 2025. https://www.copyright.gov/about/
7. U.S. Const. art. I, § 8, cl. 8. 
8. 17 U.S.C. § 101
9. United States Copyright Office. Re: Second Request for Reconsideration for Refusal to Register A Recent Entrance to Paradise. Perlmutter, Shia, Wilson, Suzanne, and Isbell, Kimberly. Correspondence ID 1-3ZPC6C3; SR # 1-7100387071. Washington, D.C.: Copyright Office, 2022. Letter, https://www.copyright.gov/rulings-filings/review-board/docs/a-recent-entrance-to-paradise.pdf.
10. Thaler v. Perlmutter, No. 22-1564 (BAH) (D.D.C. 2023)
11. Burrow-Giles Lithographic Company v. Sarony, 111 U.S. 53 (1884)
12. Thaler v. Perlmutter, No. 22-1564 (BAH) (D.D.C. 2023)
13. Ibid.
14. Ibid.
15. Thaler v. Perlmutter, No. 23-5233 (D.C. Cir. 2025)
16. Ibid.
17. Library of Congress, “Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence,”  Federal Register 88, no. 51 (March 16, 2023): 16190-16194. https://www.govinfo.gov/app/details/FR-2023-03-16/2023-05321.
18. 17 U.S.C. § 101
19. Thaler v. Perlmutter, No. 23-5233 (D.C. Cir. 2025)
20. “legal person,” Legal Information Institute, Cornell University, June 2023. https://www.law.cornell.edu/wex/legal_person.
21. Reyes, Carla, “Autonomous Corporate Personhood,” Washington Law Review 96, no. 4 (2021).
22. Halsey, Linda, “Article: Why Idaho Should Revisit Its Prohibition of Personhood for AI,” Idaho Law Review 60 (2021).
23. 17 U.S.C. § 101
24. Golger, Brian, “Comment: Copyright in the Artificially Intelligent Author: A Constitutional Approach Using Philip Bobbitt’s Modalities of Interpretation,” University of Pennsylvania Journal of Constitutional Law 22 (2020).
25. “Limited liability,” Legal Information Institute, Cornell University, June 2023.  https://www.law.cornell.edu/wex/limited_liability
26. David Millon, “The Ambiguous Significance of Corporate Personhood,” 2 Stan. Agora: Online J. Legal Persp. 39 (2001).^[1] U.S. Const. art. I, § 8, cl. 8. 
27. U.S. Const. art. I, § 8, cl. 8. 

---