Authored By: Ali Shahzad
University of Kent
Introduction
In March 2025, a United States federal court held that the Copyright Act requires all work to be authored in the first instance by a human being, refusing registration to an image produced entirely by an artificial intelligence system with no human contribution[1]. Months earlier, the UK Supreme Court had reached an analogous conclusion in patent law, holding that DABUS, another AI system created by the same litigant, could not be named as an inventor under the Patents Act 1977 because only a natural person qualifies[2]. Unlike US copyright law and UK patent law, however, UK copyright law already purports to answer the question these cases left open: section 9(3) of the Copyright, Designs and Patents Act 1988 (CDPA) grants authorship of a computer-generated work to the person by whom the arrangements necessary for its creation are undertaken[3]. Enacted decades before generative AI existed, section 9(3) was designed for deterministic software, not systems that produce unpredictable, unrehearsed output from open-ended prompts. This article argues that the arrangements test cannot coherently be applied to generative AI, and that the Government’s current preference for outright repeal, while doctrinally tidy, would leave a protection gap better filled by a narrowly tailored replacement right. The article first outlines the section 9(3) framework and its solitary case law, before examining why generative AI defeats the test, and finally assessing the repeal proposal against the alternative of reform.
The Legal Fiction: Section 9(3) and Its Solitary Case Law
Section 9(3) applies to literary, dramatic, musical or artistic works generated by computer in circumstances such that there is no human author of the work[4]. Where no human author exists, the CDPA nonetheless supplies one by legal fiction: the person who made the necessary arrangements. The only English decision squarely applying the provision is Nova Productions Ltd v Mazooma Games Ltd, where Kitchin J held that the programmer of a video game, rather than the player, had undertaken the arrangements necessary for images generated during gameplay, because he had devised the game’s logic and written the underlying code[5]. The Court of Appeal upheld this reasoning without material qualification[6].
The provision’s coherence was further strained by the separate development of the originality requirement. Following the Court of Justice’s ruling that a protected work must reflect the author’s own intellectual creation, expressing the author’s personal touch, the Court of Appeal in THJ Systems Ltd v Sheridan confirmed that this demanding, human-centred standard governs UK originality generally[7]. Because section 9(3) works are, by definition, authorless, it is unclear how they can simultaneously satisfy an originality test built around personal creative choice. Commentators have accordingly described section 9(3) as sound in seeking a human author-in-law behind every computer-generated work, but weak in explaining how that fictional author’s absent personality supplies the personal touch originality otherwise demands[8].
Generative AI and the Collapse of the “Arrangements” Test
Nova Productions concerned a program whose outputs, however numerous, were fully determined by code the programmer had written; the arrangements were traceable to a specific, identifiable act of authorship. Generative AI models break this chain. A user who types a prompt into a diffusion or language model does not write, configure or predict the output in the way the Nova programmer did; the model’s parameters are learned from training data the user never selected, and its output is probabilistic rather than deterministic. It is genuinely unclear whether prompting constitutes arrangements necessary for the creation of the work, whether that credit belongs instead to the developer who built and trained the model, or whether, for sufficiently generic prompts, no one has made the relevant arrangements at all[9]. The UK Intellectual Property Office’s own December 2024 consultation acknowledged this uncertainty, tentatively suggesting that the author will usually be the person who inputted the prompt, while conceding that the position is unsettled and typically left to platform terms of service[10]. Comparative practice underlines the difficulty: a Prague court refused copyright in an AI-generated image produced from a simple descriptive prompt, finding no human creative activity, whereas commentary on more elaborate, iterative prompting suggests the opposite conclusion might follow[11]. The provision that was meant to remove uncertainty about authorship has instead relocated that uncertainty to a new and harder question.
III. Why Repeal Alone Is Not the Answer
Faced with this incoherence, the Government’s preferred position, set out in its 2024 consultation and maintained in its subsequent report, is to remove section 9(3) altogether, citing an absence of evidence that computer-generated work protection has stimulated investment and noting that comparable protection is unavailable in AI-competitive jurisdictions such as the United States[12]. This article submits that repeal, without more, is the wrong response. It resolves the doctrinal contradiction between originality and authorless works by declining to protect purely machine-generated output at all, but this produces exactly the protection gap the DABUS reasoning warned against in the patent context: work with genuine commercial value would fall into the public domain the moment no identifiable human contribution can be shown, inviting exactly the kind of disguised authorship claims section 9(3) sought to prevent. A more defensible course is to retain a bespoke right, but recast it on the model already used elsewhere in the CDPA for entrepreneurial works, such as sound recordings, which attract protection without any originality requirement at all, in exchange for a shorter term and no moral rights[13]. Reframing computer-generated output as an investment-based, non-original right would remove the internal contradiction with the human-centred originality standard without abandoning protection altogether, and would sit more honestly with the reality that no human arranged the specific output in the way section 9(3) presently pretends.
Conclusion
Section 9(3) was drafted for a world in which computers executed instructions their programmers had fully specified. Generative AI has severed the link between a programmer’s arrangements and a model’s output, exposing both the provision’s internal tension with the human-centred originality standard confirmed in THJ Systems and its practical unworkability when applied to prompted, probabilistic systems. The Government’s preferred solution, outright repeal, resolves the doctrinal contradiction but risks leaving commercially significant AI output entirely unprotected, repeating in copyright the uncertainty DABUS exposed in patent law. This article has argued that the better reform is not repeal but recharacterisation: treating purely computer-generated output as a narrow, non-original, investment-based right akin to the CDPA’s protection for sound recordings, rather than forcing it into an authorial framework it was never equipped to satisfy. Parliament should legislate for this bespoke right when it finally revisits section 9(3), rather than leave the question, once again, to be settled by accident in litigation.
Reference(S):
Table of Cases
Infopaq International A/S v Danske Dagblades Forening (C-5/08) EU:C:2009:465
Nova Productions Ltd v Mazooma Games Ltd [2006] EWHC 24 (Ch)
Nova Productions Ltd v Mazooma Games Ltd [2007] EWCA Civ 219
Š v Taubel, Ref No 10 C 13/2023-16 (Municipal Court, Prague)
Thaler v Comptroller-General of Patents, Designs and Trade Marks [2023] UKSC 49
Thaler v Perlmutter 130 F4th 1039 (DC Cir 2025)
THJ Systems Ltd v Sheridan [2023] EWCA Civ 1354
Legislation
Copyright, Designs and Patents Act 1988
Secondary Sources
Guadamuz A, ‘Do Androids Dream of Electric Copyright? Comparative Analysis of Originality in Artificial Intelligence Generated Works’ (2017) 2 Intellectual Property Quarterly 169
Intellectual Property Office, Department for Science, Innovation and Technology and Department for Culture, Media and Sport, Copyright and Artificial Intelligence: Consultation (17 December 2024)
Department for Science, Innovation and Technology, Intellectual Property Office and Department for Culture, Media and Sport, Report on Copyright and Artificial Intelligence (2026)
Söğüt Atilla, ‘Dealing with AI-Generated Works: Lessons from the CDPA Section 9(3)’ (2024) 19 Journal of Intellectual Property Law & Practice 43
[1]Thaler v Perlmutter 130 F4th 1039, 1041 (DC Cir 2025).
[2]Thaler v Comptroller-General of Patents, Designs and Trade Marks [2023] UKSC 49.
[3]Copyright, Designs and Patents Act 1988, s 9(3).
[4]ibid s 178 (definition of ‘computer-generated’).
[5]Nova Productions Ltd v Mazooma Games Ltd [2006] EWHC 24 (Ch) [105]–[106] (Kitchin J).
[6]Nova Productions Ltd v Mazooma Games Ltd [2007] EWCA Civ 219.
[7]Infopaq International A/S v Danske Dagblades Forening (C-5/08) EU:C:2009:465, [37]; THJ Systems Ltd v Sheridan [2023] EWCA Civ 1354.
[8]Söğüt Atilla, ‘Dealing with AI-Generated Works: Lessons from the CDPA Section 9(3)’ (2024) 19 Journal of Intellectual Property Law & Practice 43, 47–48.
[9]Andres Guadamuz, ‘Do Androids Dream of Electric Copyright? Comparative Analysis of Originality in Artificial Intelligence Generated Works’ (2017) 2 Intellectual Property Quarterly 169, 184–86.
[10]Intellectual Property Office, Department for Science, Innovation and Technology and Department for Culture, Media and Sport, Copyright and Artificial Intelligence: Consultation (17 December 2024) paras 128–130.
[11]Š v Taubel, Ref No 10 C 13/2023-16 (Municipal Court, Prague).
[12]Intellectual Property Office (n 10) paras 132, 142; Department for Science, Innovation and Technology, Intellectual Property Office and Department for Culture, Media and Sport, Report on Copyright and Artificial Intelligence (2026).
[13]Copyright, Designs and Patents Act 1988, ss 5A, 9(2)(aa), 13A.
