AUTHORSHIP IN THE AGE OF THE ALGORITHM: COPYRIGHT IN AI-GENERATED WORKS UNDER THE NIGERIAN COPYRIGHT ACT 2022

ABSTRACT

The Nigerian Copyright Act 2022 was hailed as a long-overdue modernization of the country’s copyright regime, yet within months of its enactment generative artificial intelligence had begun to test assumptions that the Act takes for granted. This article examines whether, and on what basis, an AI-generated work may attract copyright protection under Nigerian law. It argues that the Act’s textual silence on machine creativity is not neutral; it conceals a fork in the road. Three competing global approaches now dominate the debate: the strict human-authorship rule restated in Thaler v Perlmutter, the United Kingdom’s pragmatic recognition of computer-generated works under section 9(3) of the CDPA 1988, and the Beijing Internet Court’s recognition of the prompt-giver as author in Li Yunkai v Liu Yuanchun. The article contends that Nigeria has, by accident rather than design, drifted closer to the American position than to either the British or Chinese alternatives, and that doctrinal coherence and creative-economy policy now require a deliberate legislative choice.

1. INTRODUCTION

Few statutes age as quickly as those that govern creativity. The Nigerian Copyright Act 2022 entered the books at almost the precise moment that text-to-image and large language models began to reshape the global creative economy, raising questions the drafters did not pause to consider. Can a song produced by Suno be protected? Does a Nollywood scriptwriter who refines a ChatGPT draft own the resulting screenplay? When a Lagos-based illustrator uses Midjourney to generate a portrait, who, if anyone, is the author?

These are not academic curiosities. They have practical consequences for licensing, infringement, royalties, and the willingness of Nigerian creators to invest in AI-assisted production. Yet the Copyright Act 2022 contains no provision expressly addressing computer-generated works. This silence has often been characterized as a “legal vacuum”, but that description is incomplete. The Act is not silent; it is structurally human-centric, and its silence on AI is itself an interpretive position. The question is whether that position should hold.

This article examines the issue in four parts. Part 2 analyses the relevant provisions of the 2022 Act and shows why, on a textual reading, AI-generated works fall outside copyright protection. Part 3 contrasts that position with three competing approaches drawn from the United States, the United Kingdom, and China. Part 4 addresses the parallel question of training-data infringement, drawing on the recent decision in Getty Images v Stability AI. Part 5 sets out reform proposals.

2. THE COPYRIGHT ACT 2022 AND THE QUIET ASSUMPTION OF HUMAN AUTHORSHIP

Under section 2(2) of the Copyright Act 2022, a literary, musical, or artistic work qualifies for copyright only where two requirements are met: some effort has been expended on the work to give it an original character, and the work has been fixed in a definite medium of expression.[1] Section 108 of the Act, which provides interpretation rules, defines an “author” by reference to specific categories of works (the author of a literary work being the writer, the author of a photograph being the photographer, and so on), without ever pausing to consider whether non-human entities can occupy those categories.[2]

Section 28(1) reinforces this assumption from a different angle. It vests first ownership of copyright in the author, and where there are joint authors, in those authors as joint owners.[3] The section presupposes that authorship can be located in a person capable of holding property rights. An algorithm cannot. It cannot transfer copyright by assignment, cannot benefit from a term measured by reference to a life, and cannot bring an action for infringement. The architecture of the Act, in other words, is human-centric not because it says so expressly but because every operative provision presumes it.

What, then, is the position of an AI-generated work? Three readings are possible. The first is that such a work is simply ineligible: it lacks a human author and therefore fails at the threshold. The second is that the human user, by formulating prompts and curating outputs, is the author. The third is that the AI developer, by designing and training the model, qualifies as the author of any output the model generates. Each reading is textually defensible, and the Act provides no mechanism for choosing between them. That ambiguity is the fault line on which Nigerian doctrine is now resting.

3. THREE PATHS NOT YET TAKEN: COMPARATIVE APPROACHES

3.1 The American Position: Authorship Reserved for Humans

The United States has spoken with unusual clarity. In Thaler v Perlmutter, the District Court for the District of Columbia held that human authorship is a bedrock requirement of copyright, and that a work generated autonomously by a machine cannot attract protection.[4] The decision was affirmed by the DC Circuit in March 2025, foreclosing for the foreseeable future any prospect of a non-human author being recognized under the US Copyright Act 1976. The Copyright Office has issued guidance to the same effect, registering only those works in which AI-generated material is accompanied by sufficient human creative contribution.[5]

The reasoning in Thaler is rooted partly in textual analysis and partly in policy. Textually, the court drew on the long line of authority beginning with Burrow-Giles Lithographic Co v Sarony,[6] which recognized photographs as copyrightable on the strength of the photographer’s creative choices. The implication is that even where a machine generates the immediate output, the law looks for a human creative spark behind it. Where there is none, no copyright subsists.

3.2 The British Compromise: Section 9(3) of the CDPA 1988

The United Kingdom adopted a different approach more than three decades before the AI debate began in earnest. Section 9(3) of the Copyright, Designs and Patents Act 1988 provides that, in the case of a literary, dramatic, musical, or artistic work that is computer-generated, the author shall be taken to be the person by whom the arrangements necessary for the creation of the work are undertaken.[7] The provision was originally drafted with rule-based expert systems and computer programs in mind, but its language is broad enough to apply to generative AI outputs. Computer-generated works enjoy a fifty-year term from the date of creation, rather than the standard life-plus-seventy.[8]

The British model is appealing because it avoids the binary choice between recognizing machine authorship and refusing protection altogether. It vests copyright in the human who orchestrates the creation, whether that person is the user, the developer, or both. Its weakness is the very feature that gives it its flexibility: the phrase “the person by whom the arrangements necessary for the creation of the work are undertaken” is genuinely indeterminate. Is it the user who selects the prompts and refines the output? Is it the developer who trains the model? Is it both? The UK has yet to produce authoritative case law fully resolving this question, although consensus appears to be coalescing around the user where the model is used as a tool.

3.3 The Chinese Innovation: The Prompt as Authorial Act

China has taken what is perhaps the most assertive position. In Li Yunkai v Liu Yuanchun, the Beijing Internet Court held that an image generated using the Stable Diffusion model could attract copyright, and that the human prompt-giver was its author.[9] The court reasoned that the plaintiff had made meaningful aesthetic choices, selecting and arranging prompt words, adjusting parameters, and curating the final output, such that the image reflected his “intellectual investment” and “personalized expression”.[10] Article 11 of the Chinese Copyright Law was held to exclude the AI model itself as author, but to permit the human user to be recognized in that role.[11]

The decision is striking for two reasons. First, it implicitly accepts that the act of prompting is a creative act sufficient to ground authorship, departing sharply from the American position. Secondly, it does so without legislative reform; the Beijing court achieved by interpretation what the United Kingdom achieved by statute. Although Chinese decisions do not bind under stare decisis, Li v Liu is now widely cited as evidence that civilian and common law systems alike are capable of accommodating AI-generated content within the existing copyright architecture, given sufficient interpretive will.

3.4 Where Nigeria Sits

Nigeria has not chosen between these three paths and, on a textual reading, has by default drifted closest to the American position. The Copyright Act 2022 contains no equivalent of section 9(3) of the CDPA 1988, and there is no Nigerian equivalent of Li v Liu. As Lateef has recently observed, the Act maintains a strict human-authorship requirement that leaves no doctrinal foothold for AI-generated works.[12] A Nigerian court today, asked to decide an analogue of Thaler, would almost certainly reach the same conclusion as the District Court in Washington. That outcome may be defensible, but it ought to be the product of a deliberate choice rather than legislative inertia.

4. THE OTHER COPYRIGHT QUESTION: TRAINING DATA AND INFRINGEMENT

Authorship is only one half of the AI copyright debate. The other half concerns whether the act of training a generative model on copyrighted works itself constitutes infringement. This question matters intensely for Nigerian rights-holders whose works may be scraped, ingested, or otherwise used by foreign AI developers without consent or license.

In November 2025, the High Court of England and Wales delivered judgment in Getty Images (US) Inc v Stability AI Ltd.[13] The court rejected Getty’s secondary-infringement claim, reasoning that the trained model did not store copies of the works on which it had been trained; the model had simply learned statistical patterns.[14] The headline copyright claim was abandoned during trial because Getty could not establish that the training had occurred within the United Kingdom. The judgment is therefore narrower than its prominence suggests, but it does mark the first substantive UK engagement with the question of whether an AI model is an “infringing copy”.

The German Munich Regional Court reached a different result in GEMA v OpenAI,[15] holding that the use of copyrighted song lyrics to train ChatGPT amounted to unauthorized reproduction under German and EU law, including in the model’s parameters. The contrast between the two decisions, both delivered within weeks of one another, illustrates how unsettled the position is in advanced jurisdictions.

Nigerian law offers no specific provision on training-data use. Section 20 of the Copyright Act 2022 sets out fair-dealing exceptions for purposes such as research, private study, criticism, and review, but these were not drafted with industrial-scale model training in mind.[16] Section 36 prohibits unauthorized reproduction in tangible or digital form. A Nigerian rights-holder seeking to challenge an AI developer would face a doctrinal puzzle similar to that addressed in Getty: did the model ever “reproduce” the work in the statutory sense, or did it merely learn from it? Without legislative clarification, the issue is likely to be resolved (if at all) on facts that may not arise in Nigerian courts for years, given that most major models are trained outside the jurisdiction. The pending United States litigation in Authors Guild v OpenAI and New York Times v Microsoft will almost certainly shape the discussion long before any Nigerian court is asked to address it.[17]

5. REFORM PROPOSALS

Three reforms suggest themselves. First, the Copyright Act 2022 should be amended to insert a provision modelled on section 9(3) of the CDPA 1988, vesting authorship of computer-generated works in the human who makes the arrangements necessary for their creation. The British formulation has the merit of being technology-neutral, and Nigeria can borrow from over thirty years of UK jurisprudence and commentary in giving it content. A shorter term of protection (perhaps thirty or fifty years from creation) for computer-generated works would respect the lower creative threshold without depriving creators of all incentive. Lateef has argued for an even more flexible approach, allowing recognition of human-AI collaborative works under joint authorship, with sui generis protection for purely AI-generated content.[18]

Secondly, Parliament should clarify the position on training-data use. A narrowly drawn text-and-data-mining exception, similar in structure to that contained in articles 3 and 4 of the EU CDSM Directive but with a robust opt-out mechanism for rights-holders, would balance the interests of Nigerian creators against the legitimate research and innovation needs of the technology sector.[19] The opt-out mechanism is critical: without it, Nigerian musicians, photographers, and writers risk having their works absorbed into foreign models without remuneration.

Thirdly, the Nigerian Copyright Commission should issue guidance on the registration of works in which AI played a role, requiring disclosure of the extent of AI involvement.[20] This would mirror the practice now adopted by the United States Copyright Office and would give Nigerian courts a documentary record on which to assess the human contribution required for protection. Such guidance need not await legislative reform; it can be issued under the Commission’s existing regulatory powers, and would align with the National Artificial Intelligence Strategy released in 2024.[21]

6. CONCLUSION

The Copyright Act 2022 was a creditable piece of legislative drafting for its time, but its time is already passing. The text-to-image, text-to-music, and text-to-video tools that now populate Nigerian creative practice were barely commercial when the Act received presidential assent. Continued silence on AI authorship and training-data use will not preserve neutrality; it will simply lock Nigeria into the strictest reading of its existing provisions, with consequences that may not serve the country’s creative economy or its emerging position in the African technology landscape. The choice between the American, British, and Chinese paths is no longer one Nigeria can postpone. Whether by amendment, regulation, or judicial interpretation, the country must decide who, if anyone, owns the work of the algorithm.

10. Reference(S):

[1] Copyright Act 2022 (Nigeria), s 2(2)(a) – (b).

[2] ibid, s 108 (interpretation).

[3] ibid, s 28(1).

[4] Thaler v Perlmutter 687 F Supp 3d 140 (DDC 2023), affd No 23-5233 (DC Cir, 18 March 2025).

[5] US Copyright Office, ‘Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence’ (2023) 88 Fed Reg 16190.

[6] Burrow-Giles Lithographic Co v Sarony 111 US 53, 58–60 (1884).

[7] Copyright, Designs and Patents Act 1988, ss 9(3) and 178.

[8] Copyright, Designs and Patents Act 1988, s 12(7).

[9] Li Yunkai v Liu Yuanchun (2023) Jing 0491 Min Chu No 11279 (Beijing Internet Court, 27 November 2023).

[10] Li Yunkai v Liu Yuanchun (n 7).

[11] Copyright Law of the People’s Republic of China 2020, art 11.

[12] Misbau Alamu Lateef, ‘Artificial Intelligence and Copyright Authorship under the Nigerian Copyright Law’ (2025) 4(2) PPL RUNLAW Review <https://runlawjournals.com/index.php/pplrunlaw/article/view/158> accessed 30 April 2026.

[13] Getty Images (US) Inc v Stability AI Ltd [2025] EWHC 2863 (Ch).

[14] ibid [400] – [420] (Smith J).

[15] GEMA v OpenAI (Landgericht München I, 11 November 2025) Case 42 O 14139/24

[16] Copyright Act 2022 (Nigeria), ss 20 and 36.

[17] Authors Guild v OpenAI Inc, No 1:23-cv-08292 (SDNY, filed 19 September 2023); New York Times Co v Microsoft Corp, No 1:23-cv-11195 (SDNY, filed 27 December 2023).

[18] Lateef (n 15).

[19] Directive (EU) 2019/790 of the European Parliament and of the Council of 17 April 2019 on copyright and related rights in the Digital Single Market [2019] OJ L130/92, arts 3 – 4.

[20] Copyright Act 2022 (Nigeria), s 78 (functions and powers of the Nigerian Copyright Commission).

[21] Federal Ministry of Communications, Innovation and Digital Economy, National Artificial Intelligence Strategy (FMCIDE 2024).