Who Owns the Machine’s Masterpiece? Rethinking Intellectual Property Rights in the Age of Generative Artificial Intelligence

1. Introduction

The rapid adoption of Generative Artificial Intelligence (AI) since late 2022 has significantly impacted creative industries by enabling the production of literature, music, and visual art at an unprecedented pace and scale. This shift challenges the anthropocentric foundations of traditional intellectual property frameworks, which have long conditioned protection on human authorship.[1] This article assesses whether AI-Generated Content constitutes legally protected intellectual property and explores the legal mechanisms for assigning ownership. Through a comparative jurisdictional analysis, this study identifies three models:

1. The United States of America (United States) maintains a rigid human-authorship requirement, given that courts deny protection to purely AI-Generated content, as seen in the case of Thaler v Perlmutter.[2]
2. The United Kingdom (UK), which uses the Copyright, Designs and Patents Act, 1998[3] to grant authorship of computer-generated works to the person who made necessary arrangements to create them
3. China, which uses a flexible approach, as the Beijing Internet Court protected an AI-generated image as art, vesting authorship in the user based on textual prompts and parameter setting.[4]

The article further examines the South African context. While Section 1(1)(h) of the Copyright Act[5] contains a computer-generated provision similar to the UK’s, it lacks modern judicial interpretation. The study concludes by suggesting that South Africa should provide legislative clarity through functional tests that examine the creative process or through the establishment of sui generis rights to promote innovation without devaluing human creativity.

Keywords: Generative AI; AI-Generated Content; Intellectual Property; human authorship; Li v Liu; Hegelian personality theory; Locke’s labour theory; prompt engineering

2. Why AI Challenges Traditional Copyright Theory

Generative AI breaks the link between a human creator and their work. Historically, it has been assumed that authorship is a unique human trait, but AI-Generated Content produced by artificial neural networks currently creates masterpieces at a speed and scale that defies traditional norms.[6]

The primary objective of copyright is to incentivise the production of original works for public benefit. The law grants short-term monopolies, thereby encouraging human authors to put in the effort of creation knowing they can garner rewards.[7] AI challenges this purpose because, unlike human authors, machines do not respond to economic incentives; computers need no rewards to generate content. As a result, traditional legal frameworks cannot easily bring themselves to accept the existence of valuable AI-Generated Content with a system designed to protect intellectual labour derived from the human mind.[8]

Numerous jurisdictions maintain that authorship and originality are linked to this human-centric requirement. Authorship is exclusively a human trait, as seen in the statutory language of the United States, which references human heirs such as widows and children, and by judicial precedents such as the Naruto v Slater[9] case, which established that non-human entities lack statutory standing.[10] Originality requires an author’s independent creation and a “modicum of creativity”.[11] Instead of making deliberate choices, AI systems create content through automated processes and a statistical recognition of patterns.[12] This leads to the discourse that their content lacks the necessary creative spark and human authorship, as it is an autonomous AI creation, which is required for protection.[13]

Philosophical justifications for intellectual property, specifically Locke’s labour theory and Hegel’s personality theory, further complicate the status of AI-Generated Content. Locke’s labour theory posits that property rights are earned when a person mixes their labour with common resources, implying that creators deserve rights in works produced through their own intellectual effort.[14] Hegel’s personality theory views property as an externalisation of the author’s will and personality, arguing that creative works are an extension of the self.[15] Because these theories are rooted in human agency and self-development through creation, they provide a strong basis for denying rights to a machine, which possesses neither a will to externalise nor the capacity for personal growth through effort.

The question of whether the human prompter satisfies these theories remains a point of intense legal debate. In the landmark Chinese case of Li v Liu, the court held that a human user could be an author because their tailored prompts and parameter settings reflected original textual input, personal judgment, and aesthetic choice,[16] effectively treating the AI as a tool similar to a camera.[17] This implies that the prompter’s intellectual labour in crafting prompts and the outputs could satisfy Locke’s theory of mixed labour. However, critics argue that prompts are often too abstract; they represent an idea instead of a concrete expression. This leans toward the criticism that the AI, rather than the human, performs the execution of the creative plan.[18] Under this view, if there is insufficient creative control over the final expressive elements, the work fails to embody the personality or labour required by traditional theories, potentially leaving AI-Generated Content in the public domain.[19]

3. Comparative Jurisdictional Analysis

3.1 The United States of America

The United States maintains a rigid, anthropocentric approach to human authorship, rooted in historical and philosophical justifications that link property rights to human intellectual labour and personal expression.[20] Historically, U.S. courts have defined an author as a natural person who creates original work through intellectual invention, a principle established in the landmark 1884 case Burrow-Giles Lithographic Co. v Sarony.[21] This approach is heavily influenced by utilitarianism and the labour theory, aiming to provide economic incentives for human creativity. Because AI systems do not respond to such rewards, they fall outside the primary purpose of the U.S. copyright framework.[22]

Under current legislation, the Copyright Act of 1976 requires all eligible work to be authored in the first instance by a human being.[23] The U.S. Copyright Office (USCO) reinforces this through its regulatory practices, as detailed in the Compendium of U.S. Copyright Office Practices, which explicitly states that the Office will refuse registration for works produced by nature, animals, or machines that operate without creative input from a human.[24] The statutory framework further implies humanity through provisions regarding the lives of authors and the rights of their “widows,” “children,” or heirs to inherit interests; attributes that non-human entities lack.[25]

This requirement has been upheld through major judicial precedents. In the “monkey selfie” case, Naruto v Slater, the court held that a crested macaque lacked statutory standing because the Copyright Act does not expressly authorise animals to file infringement suits.[26] This logic was extended to artificial intelligence in Thaler v Perlmutter, where the court affirmed the denial of copyright for an image created autonomously by an AI named the “Creativity Machine”.[27] The court noted that even the “work-made-for-hire” doctrine cannot be used to justify AI authorship, as that provision presupposes a protectable interest originally created by a human before ownership can legally transfer to an employer.[28]

Despite this strictness toward purely autonomous output, the U.S. approach does accommodate AI-assisted works where a human exercises significant creative control. In such instances, there is limited protection to the human contribution. For example, in the Zarya of the Dawn[29] decision, the USCO granted protection to the human-authored text and the original selection and arrangement of the images, but denied protection to the individual images themselves because they were generated by the AI tool Midjourney rather than by the human applicant.[30] Consequently, U.S. regulators apply a functional, case-by-case analysis to determine if the expressive elements of a work were actually conceived and executed by a human or a machine.[31]

3.2 The United Kingdom

The United Kingdom’s statutory approach is distinctive because its primary legislation, the Copyright, Designs and Patents Act, explicitly accommodates works created without a human author. Under Section 178 of the Act, a work is classified as “computer-generated” if it is produced in circumstances where there is “no human author”. For such works, Section 9(3) attributes authorship to the “person by whom the arrangements necessary for the creation of the work are undertaken”. This provision creates a “legislative fiction” that allows copyright to subsist in automated outputs by vesting rights in a human, rather than the machine itself.

This statutory framework creates a distinct class of protection that differs from traditional human authorship in several key ways. Consequently, computer-generated works are treated differently in terms of the scope and duration of protection. For instance, while traditional literary, dramatic, musical, or artistic works enjoy protection for the life of the author plus 70 years, the term for computer-generated works is limited to 50 years from the end of the calendar year in which the work was made.[32] Furthermore, these works are excluded from certain moral rights, such as the right to be identified as the author under Section 79(2)(c).

Despite statutory definitions, the rise of modern generative AI creates uncertainty in the application of the “arrangements necessary” test. It remains unclear whether simply providing a textual prompt to an AI system constitutes making the necessary arrangements for creation, or whether more substantial involvement, such as programming the AI or curating its training data, is required. Case law, such as Nova Productions Ltd v Mazooma Games Ltd, acknowledges the distinction between a computer-aided and a computer-generated work where the human contribution may be too trivial to satisfy traditional authorship principles.[33]

3.3 China

China has adopted a flexible and pragmatic approach to AI-Generated Content, distinguishing its judicial outcomes from the more rigid standards seen in the United States. In the landmark 2023 case of Li v Liu, the Beijing Internet Court ruled that an image generated using the AI software Stable Diffusion constituted a work of fine art protected under Chinese copyright law.[34] The court attributed authorship to the human user, reasoning that their prompts and iterative parameter adjustments, such as choosing the model, setting the “Random Seed” index, and selecting the subject’s detailed features, reflected a personalised expression and “aesthetic choice”. This approach treats AI as an “ordinary tool” similar to a camera or a paintbrush, sidestepping the debate over machine creativity by focusing on the human user’s control and choice throughout the iterative refinement process.[35]

The Chinese legal framework is also actively evolving to address the copyright risks associated with the AI development process and training data. In 2023, China implemented the Interim Measures for the Management of Generative AI Services, which require providers to respect intellectual property rights and maintain transparency regarding their datasets.[36] These regulations, specifically Article 15, impose a “duty of care” on AI developers, compelling them to establish reporting mechanisms so that copyright holders can flag potential infringements.[37] The legal weight of these measures was demonstrated in the Shanghai Character License Administrative Co. (SCLA) v AI Company[38] case, where a court found an AI provider liable for generating images substantially similar to the copyrighted “Ultraman” characters, ruling that the company had violated the copyright holder’s exclusive right to prepare derivative works.[39]

However, the Chinese approach distinguishes between distinct levels of AI automation and human involvement. While Li v Liu vested rights in the user of a partially generative tool, earlier cases like Tencent v Yingxun[40] assigned copyright in an AI-generated news article to the developer, as the developer was found to have exercised significant control over the data input and templates.[41] This suggests that China’s framework is evolving into a differentiated model where ownership is assigned based on who, whether developer or user, exercised effective creative control over the specific output.[42] By framing AI as a tool rather than a creator, the Chinese judiciary maintains the anthropocentric foundations of copyright while pragmatically expanding the scope of protection to encompass masterpieces produced in the age of silicon.[43]

3.4 The Republic of South Africa

The Copyright Act 98 of 1978 serves as the primary legal framework for protecting various classes of work in South Africa, including literature, art, and music. Despite its longevity, there is a notable legislative silence regarding the specific legal status of autonomous generative AI, as the Act was not originally designed to contemplate machine-driven creativity. While the Act does provide a definition for the “author” of a computer-generated work, naming them as the “person by whom the arrangements necessary for the creation of the work were undertaken” under Section 1(1)(h), it lacks the clarity found in other jurisdictions, such as the United Kingdom, which defines computer-generated works as those having no human author. This silence creates significant regulatory gaps and increases the country’s vulnerability to challenges related to ownership, cybersecurity, and the protection of intellectual property in a rapidly digitalising economy.

This legislative ambiguity is compounded by a distinct lack of judicial guidance directly addressing AI authorship within a copyright context.[44] As it stands, no South African court has ruled on whether a purely autonomous AI output can be protected by copyright. The most relevant existing precedent is Haupt v Brewers Marketing Intelligence,[45] where the Supreme Court of Appeal held that databases generated by algorithms qualify as protectable literary works, provided they meet requirements for originality and expression.[46] Outside the courts, the discourse has been dominated by the Companies and Intellectual Property Commission (CIPC), which in 2021 became the first in the world to grant a patent to an AI system named DABUS as the inventor.[47] However, Oriakhogba, among many legal scholars, argues that this administrative grant may not withstand judicial scrutiny, as it arguably contradicts the human-centric foundation of the South African patent and copyright regimes.[48]

In the event of future litigation, South African courts would likely adopt a cautious, anthropocentric approach to AI-Generated Content. Courts are expected to interpret the existing “arrangements necessary” clause in the Act to mean that a natural or juristic person behind the AI must be identified as the author, effectively treating the technology as an assistive tool rather than a creator.[49] This would align South Africa with global judicial trends in the United States and China, which generally refuse copyright to purely autonomous AI outputs while protecting works that reflect a clear “human cause” or “intellectual labour”. Courts would likely hold that for a work to be original and copyrightable, it must be the product of a conscious agent’s creative choices. Thus, works generated without sufficient human oversight would likely be deemed unprotectable and enter the public domain. This interpretation would help prevent the proliferation of patent thickets and ensure that the law continues to focus on the social and economic well-being of human creators rather than machines.

4. Proposed Solution for South Africa

A proposed solution for South Africa is for courts to adopt a substantial human contribution test when interpreting the existing “arrangements necessary” clause in the Copyright Act. This test would move beyond mere administrative arrangements and assess whether the technology was used as a creative tool rather than an autonomous creator, ensuring that protection is granted only when there is a clear human cause behind the final expression.[50] By requiring evidence of material human intervention, courts can maintain the human-centric focus of South African intellectual property law while accommodating modern, AI-assisted creative workflows.

Under this proposed test, the quality and specificity of prompts would serve as a primary indicator of whether the user defined the work’s expressive elements or merely provided an abstract idea. As seen in Li v Liu, tailored and specialised prompts can reflect a user’s original textual input, aesthetic choices, and personal judgment, effectively transforming the AI into a tool similar to a camera.[51] Additionally, the number of iterations and the overall iterative process would be critical. A user who continually modifies parameters and prompts to move away from “random tensors” and toward a specific vision demonstrates the level of creative control necessary for authorship.

The selection of outputs, along with subsequent editing and refinement, would further distinguish protectable works from purely automated ones. Furthermore, redaction or the use of external software to refine the image shows that the human has taken direct responsibility for the work’s expressive features. Ultimately, courts should weigh the degree of creative judgment exercised throughout the process, granting copyright only when the human made meaningful creative choices that are “objectively revealed” in the final work. This balanced approach prevents the public domain from being shrunk by autonomous machine outputs while continuing to reward genuine human ingenuity.

5. Conclusion

The age of generative artificial intelligence has brought us to a legal and philosophical crossroads, forcing a global re-evaluation of what it means to be an “author” in the 21st century. As this article has demonstrated, the global response to AI-Generated Content is deeply fragmented, ranging from the United States’ rigid human-authorship requirement to China’s flexible, user-centric acceptance of original prompts as creative labour. While the United States remains anchored in the belief that only the “creative powers of the human mind” deserve protection, China and the UK have recognised that the path forward requires a more pragmatic “tool narrative” or legislative fiction to accommodate human-AI collaboration.

For South Africa, the path toward resolving the ownership of the “machine’s masterpiece” is already partially paved. Section 1(1)(h) of the Copyright Act provides a unique statutory bridge, the “arrangements” clause, which allows computer-generated works to have an author without requiring that the author be the machine itself. However, to transform this into a robust framework for the digital era, South African courts and policymakers must adopt a substantial human contribution test. By assessing the quality of prompts, the number of iterations, and the degree of creative judgment exercised by the user, the law can ensure that protection is only granted when a human has truly “mixed their labour” or “externalised their personality” into the final output.

Adopting such a test would allow South Africa to bypass the anthropocentric gridlock seen in other jurisdictions while avoiding the danger of patent thickets or the flooding of the public domain with low-quality, autonomous machine outputs. This balanced approach honours the Hegelian justification for property; protecting works as an extension of the creator’s will, while providing the legal certainty necessary to incentivise investment in South Africa’s burgeoning AI sector.

Ultimately, the question of “Who owns the machine’s masterpiece?” should not be answered by the machine’s capabilities, but by the human agency that directs them. By providing clear legislative and judicial guidance on what constitutes “necessary arrangements,” South Africa can protect its creative industries, promote technological innovation, and ensure that in the age of silicon, human creativity remains the heart of the law.

BIBLIOGRAPHY

Books

Caroline B. Ncube et al., Artificial Intelligence and the Law in Africa. 2023.

John Locke, Second Treatise of Government. 2010

Panashe C. Zowa, Intellectual Property Rights and Sustainable Development Goals in Africa, 1^st ed. 2025.

United States Copyright Office, Compendium of U.S. Copyright Office Practices, 3^rd ed. Chapter 300: Copyrightable Authorship: What Can Be Registered. (updated 28 January 2021).

Case Law

Burrow-Giles Lithographic Co. v Sarony 111 U.S. 53 (1884).

Haupt v Brewers Marketing Intelligence 2006 (4) SA 458 (SCA).

Li v Liu (2023) Jing 0491 Min Chu No. 11279.

Lin Chen v Hangzhou Gauss Inflatable Tech Co., Ltd (2024) Su 0581 Min Chu 6697.

Naruto v Slater No. 16-15469 (9^th Cir. 2018).

Nova Productions Ltd v Mazooma Games Ltd [2006] EWHC 24 (Ch).

Shanghai Character License Administrative Co. (SCLA) v AI Company (Alias) [(2024) 0192 113].

Shenzhen Tencent v Shanghai Yingxun (2019) Yue 0305 Min Chu 14010.

Thaler v Perlmutter 130 F.4th 1039 (D.C. Cir. 2025).

U.S. Copyright Office, Letter re: Zarya of the Dawn (Registration # VAu001480196).

Legislation

Copyright Act of 1976.

Copyright Act 98 of 1978.

Copyright, Designs and Patents Act, 1998.

Articles

DO Oriakhogba, Centering Intellectual Property in Artificial Intelligence Strategies in Africa Through a Techno-legal Analysis (2025).

Kanu Priya, Intellectual Property and Hegelian Justification, 1 NUJS L. Rev. (2008).

Mikhalien Du Bois et al., South African Intellectual Property Law Journal, 9 IPLJ. (2021).

Xi Lin, Protecting AI-Generated Images as Works of Fine Art in China: Magnifying the Legacy of Art to Copyright, 7(1). (2025).

Yiran Li, The Human Authorship Requirement in AI-Generated Works: A Comparative Analysis of Copyright Protection Frameworks, Journal of Law and Governance. (2025).

[1] Yiran Li, The Human Authorship Requirement in AI-Generated Works: A Comparative Analysis of Copyright Protection Frameworks, Journal of Law and Governance. (2025).

[2] Thaler v Perlmutter 130 F.4^th 1039 (D.C. Cir. 2025).

[3] Copyright, Designs and Patents Act, 1998.

[4] Li v Liu (2023) Jing 0491 Min Chu No. 11279.

[5] Section 1(1)(h) of the Copyright Act 98 of 1978 provides that, “unless the context otherwise indicates, “author”, in relation to a literary, dramatic, musical or artistic work or computer program which is computer generated, means the person by whom the arrangements necessary for the creation of the work were undertaken”.

[6] Thaler v Perlmutter.

[7] Thaler v Perlmutter, Xi Lin, Protecting AI-Generated Images as Works of Fine Art in China: Magnifying the Legacy of Art to Copyright. 7(1). (2025).  110.

[8] United States Copyright Office, Compendium of U.S. Copyright Office Practices (3rd edn, 2021) ch 300, para 306.

[9] Naruto v Slater No. 16-15469 (9^th Cir. 2018).

[10] Thaler v Perlmutter.

[11] Yiran Li, The Human Authorship Requirement in AI-Generated Works: A Comparative Analysis of Copyright Protection Frameworks, Journal of Law and Governance. 22.

[12] Yiran Li, The Human Authorship Requirement in AI-Generated Works: A Comparative Analysis of Copyright Protection Frameworks, Journal of Law and Governance. 22.

[13] Yiran Li, The Human Authorship Requirement in AI-Generated Works: A Comparative Analysis of Copyright Protection Frameworks, Journal of Law and Governance. 22, United States Copyright Office, Compendium (n X) para.308.

[14] John Locke, Second Treatise of Government, sec. 27.

[15] Kanu Priya, Intellectual Property and Hegelian Justification. 360.

[16] Li v Liu, Xi Lin, Protecting AI-Generated Images as Works of Fine Art in China: Magnifying the Legacy of Art to Copyright. 107.

[17] Li v Liu, Xi Lin, Protecting AI-Generated Images as Works of Fine Art in China: Magnifying the Legacy of Art to Copyright. 107.

[18] Xi Lin, Protecting AI-Generated Images as Works of Fine Art in China: Magnifying the Legacy of Art to Copyright. 118, Thaler v Perlmutter.

[19] Xi Lin, Protecting AI-Generated Images as Works of Fine Art in China: Magnifying the Legacy of Art to Copyright. 120, 121.

[20] Thaler v Perlmutter, Yiran Li, The Human Authorship Requirement in AI-Generated Works: A Comparative Analysis of Copyright Protection Frameworks, Journal of Law and Governance. 22.

[21] Burrow-Giles Lithographic Co. v Sarony 111 U.S. 53 (1884).

[22] United States Copyright Office, Compendium (n X) para. 304-306.

[23] Thaler v Perlmutter.

[24] United States Copyright Office, Compendium (n X) para. 306.

[25] Naruto v Slater, Thaler v Perlmutter.

[26] Naruto v Slater.

[27] Thaler v Perlmutter.

[28] Thaler v Perlmutter.

[29] U.S. Copyright Office, Letter re: Zarya of the Dawn (Registration # VAu001480196)

[30] Yiran Li, The Human Authorship Requirement in AI-Generated Works: A Comparative Analysis of Copyright Protection Frameworks, Journal of Law and Governance.

[31] United States Copyright Office, Compendium (n X) para. 313.2.

[32] Xi Lin, Protecting AI-Generated Images as Works of Fine Art in China: Magnifying the Legacy of Art to Copyright. 113.

[33] Xi Lin, Protecting AI-Generated Images as Works of Fine Art in China: Magnifying the Legacy of Art to Copyright. 111.

[34] Xi Lin, Protecting AI-Generated Images as Works of Fine Art in China: Magnifying the Legacy of Art to Copyright.  107, Li v Liu.

[35] Xi Lin, Protecting AI-Generated Images as Works of Fine Art in China: Magnifying the Legacy of Art to Copyright. 107, 112, 113, 121 Lin Chen v Hangzhou Gauss Inflatable Tech Co., Ltd., (2024) Su 0581 Minchu 6697 (Changshu People’s Court., Oct. 18, 2024).

[36] Zowa, Intellectual Property Rights and Sustainable Development Goals in Africa. 319.

[37] Zowa, Intellectual Property Rights and Sustainable Development Goals in Africa. 319.

[38] Shanghai Character License Administrative Co. (SCLA) v AI Company [(2024) 0192 113].

[39] Zowa, Intellectual Property Rights and Sustainable Development Goals in Africa. 318.

[40] Shenzhen Tencent v Shanghai Yingxun (2019) Yue 0305 Min Chu 14010 (Changshu People’s Court., Dec. 24, 2019).

[41] Yiran Li, The Human Authorship Requirement in AI-Generated Works: A Comparative Analysis of Copyright Protection Frameworks, Journal of Law and Governance. 23.

[42] Xi Lin, Protecting AI-Generated Images as Works of Fine Art in China: Magnifying the Legacy of Art to Copyright. 21, 22.

[43] Xi Lin, Protecting AI-Generated Images as Works of Fine Art in China: Magnifying the Legacy of Art to Copyright. 107.

[44] Oriakhogba, Centering Intellectual Property in Artificial Intelligence Strategies in Africa Through a Techno-legal Analysis (2025). 17.

[45] Haupt v Brewers Marketing Intelligence 2006 (4) SA 458 (SCA).

[46] Haupt v Brewers Marketing Intelligence, Ncube et al., Artificial Intelligence and the Law in Africa. 2023. 276

[47] Ncube et al., Artificial Intelligence and the Law in Africa. 276, South African Intellectual Property Law Journal. 87.

[48] Ncube et al., Artificial Intelligence and the Law in Africa. 279

[49] Oriakhogba, Centering Intellectual Property in Artificial Intelligence Strategies in Africa Through a Techno-legal Analysis.19, 24.

[50] Xi Lin, Protecting AI-Generated Images as Works of Fine Art in China: Magnifying the Legacy of Art to Copyright. 112.

[51] Xi Lin, Protecting AI-Generated Images as Works of Fine Art in China: Magnifying the Legacy of Art to Copyright. 107, Panashe C. Zowa, Intellectual Property Rights and Sustainable Development Goals in Africa. 321.