Authored By: Jocelyn Fortgens
University of Witwatersrand
Introduction
Modern patent law was built on the assumption that invention is a fundamentally human activity. It rewards individuals and organisations for their ingenuity by granting temporary monopolies in exchange for public disclosure. For most of its history, this model has functioned without serious challenge. However, recent developments in artificial intelligence (AI) have begun to unsettle that assumption.
In 2018, Dr Stephen Thaler filed a series of patent applications in multiple jurisdictions, not in his own name, but on behalf of an AI system known as the Device for the Autonomous Bootstrapping of Unified Sentience (DABUS). This decision by Dr Thaler was deliberate, as it forced patent offices and courts to confront a question that existing legislation had never clearly anticipated: can a machine be recognised as an inventor?
A clear judicial consensus has been established globally. Courts in the United States and the United Kingdom have held that inventorship is limited to natural persons, while Australia ultimately adopted the same position on appeal after a brief departure at first instance.[1],[2],[3] There is only one jurisdiction to have granted a DABUS patent, and it stands alone on this equally significant and controversial decision – South Africa.[4]
This article examines these developments through the lens of the DABUS litigation. It argues that, although the prevailing judicial approach is doctrinally coherent, it exposes a growing tension between established legal concepts and the realities of technological change. The article further considers whether incremental interpretation is sufficient, or whether more direct legislative intervention will be required as AI systems play an increasingly significant role in the inventive process.
Background: The DABUS Applications
DABUS, developed by Dr Thaler, is a connectionist AI system designed to reproduce aspects of human cognition, including the ability to generate novel ideas and recognise their practical utility. According to the World Intellectual Property Organisation (WIPO), the system functions by forming new associations and autonomously recognising the use of such concepts, without human intervention at the point of creation.[5] Two outputs generated by DABUS became the subject of global patent applications: first, a food container utilising fractal geometry, optimised for robotic handling and stacking; and second, a ‘neural flame’, a flickering light signal intended to draw human attention in emergency situations.[6]
The decision to name DABUS as the inventor, and not himself, was intentional by Dr Thaler, and was done to test the limits of existing patent law. He sought to determine whether current legal frameworks could provide for autonomous machine invention. The applications were submitted under the Patent Cooperation Treaty and examined on an individual basis by national offices. In each jurisdiction, the central question was one of statutory interpretation: whether the applicable legislation permits a non-human entity to be recognised as an inventor?[7],[8]
The cases were not purely academic exercises. In several jurisdictions, including the United Kingdom (UK), the inventions were recognised as both novel and inventive, with the UK Intellectual Property Office (UKIPO) raising no objections on their substantive patentability. The dispute focused solely on the inventor’s identity: a procedural and definitional matter with significant practical implications for the future of AI-driven innovation.[9]
How the Major Jurisdictions Ruled
The United States: Thaler v Vidal (Fed Cir, 2022)
The United States (US) was one of the first major jurisdictions to issue a definitive ruling on AI inventorship. The United States Court of Appeals for the Federal Circuit held that the Patent Act requirement that an inventor be an “individual” unambiguously refers to a natural person.[10] In reaching this conclusion, the court relied on the consistent statutory interpretation of 35 USC § 100(f)–(g), which explicitly defines inventors and joint inventors in terms of natural persons.[11] The US Supreme Court subsequently declined to hear a challenge to the ruling, leaving the Federal Circuit ruling as final authority under US law.[12]
The United Kingdom: Thaler v Comptroller-General [2023] UKSC 49
Proceedings in the UK commenced when the UKIPO rejected Dr Thaler’s patent applications in December 2019 on the basis that DABUS could not qualify as an inventor under the Patents Act 1977.[13] The Hearing Officer held that an “inventor”, as defined in section 7(3), refers to the “actual devisor” of the invention, a concept interpreted by the House of Lords in Yeda Research and Development v Rhone-Poulenc as the natural person responsible for conceiving the inventive concept.[14] Appeals to the High Court and subsequently to the Court of Appeal were unsuccessful, with the latter issuing a majority decision affirming this interpretation.[15]
The matter ultimately reached the Supreme Court of the United Kingdom, where Lord Kitchin, alongside Lord Hodge, Lord Hamblen, Lord Leggatt and Lord Richards, unanimously dismissed the appeal. Delivering the leading judgement, Lord Kitchin adopted a deliberately narrow textualist approach, emphasising that the Court’s role was not to engage with broader policy considerations surrounding AI-generated inventions and whether they ought to be patentable, but rather to interpret the statute as enacted. On this basis, the Court held that the structure and language of the 1977 Act admit of only one conclusion: an inventor must be a natural person possessing legal personality. As DABUS is not, and has never been, a legal person, it is incapable of holding or transferring rights, with the result that no entitlement could vest in Dr Thaler by virtue of ownership.[16]
Significantly, the Court left open an important possibility. Lord Kitchin noted that the outcome might have differed had Dr Thaler identified himself as the inventor and characterised DABUS merely as a highly advanced tool used in the inventive process. This distinction between AI as inventor and AI as instrument is central and is revisited later in this article.[17]
Australia: Commissioner of Patents v Thaler [2022] FCAFC 62
Australia offered a brief departure from the emerging judicial consensus. At first instance, the Federal Court of Australia held that, because the Patents Act 1990 does not define the term “inventor”, it need not be confined to a natural person as a matter of statutory interpretation. Justice Beach rejected the Deputy Commissioner’s reliance on dictionary definitions and concluded that the statutory framework was sufficiently flexible to accommodate an AI inventor.[18]
This position was short-lived. On appeal, the Full Court of the Federal Court of Australia overturned the decision, holding that the term “inventor”, when read in the context of the Australian patent law as a whole, has always been understood to require human agency.[19]
South Africa: The CIPC DABUS Grant (2021)
South Africa stands apart as the only jurisdiction to have granted a patent naming DABUS as inventor. The Companies and Intellectual Property Commission (CIPC) approved the application for the fractal food container in June 2021, making it the first patent office in the world to formally recognise an AI as an inventor.[20]
The grant was subsequently published in the South African Patent Journal in July 2021, attracting significant international attention. Two structural features of South African patent law largely account for this outcome. Firstly, the system does not involve substantive examination: patents are granted on the basis of formal compliance, without any assessment of novelty, inventiveness, or inventorship in the doctrinal sense. Secondly, the Patents Act 57 of 1978 does not define the term “inventor”.[21] This contrasts with jurisdictions such as the UK and US, where statutory language has been interpreted as implicitly requiring a natural person.
Academic and professional opinion in South Africa remains divided. Critics, including the IP firm Spoor & Fisher, maintain that the CIPC erred in granting the patent, arguing that the application failed to satisfy section 30(4) of the Act concerning proof of entitlement and should have been refused even within a non-examining system.[22] Others, however, maintain that the outcome is defensible under a purposive approach to interpretation. They point to South Africa’s broader intellectual property policy framework, particularly the Intellectual Property Policy Phase I and the Presidential Commission on the Fourth Industrial Revolution, which emphasises innovation and technological advancement as key drivers of economic development, and argue that an unduly restrictive reading of “inventor” would undermine these objectives.[23],[24],[25]
Critical Analysis: The Policy Tensions Exposed by DABUS
The “Human Inventor” Requirement
The requirement that an inventor be human has rarely been stated explicitly in patent legislation, largely because it was historically assumed to be self-evident. Patent law is grounded in a theory of natural rights, most closely associated with John Locke, according to which the act of creation is an expression of human intellect deserving of legal protection.[26] This foundation has not only justified the existence of the patent system but has also shaped its doctrinal structure – from the centrality of “mental conception” as the defining criterion of inventorship to rules governing assignment and succession, all of which presuppose a legal person capable of holding rights and bearing obligations.
AI systems such as DABUS, however, do not fit within this conceptual framework. DABUS does not “conceive” of inventions in the manner outlined by patent law; rather, it processes data, identifies patterns, and produces outputs without intention, interest, or legal personality. As Ryan Abbott has argued, the law’s insistence on human inventorship reflects a deeper assumption that creativity is inherently human, an assumption which is increasingly challenged by advances in AI.[27] The Supreme Court of the United Kingdom acknowledged this tension but declined to resolve it, emphasising that any reconsideration of the legal position is a matter for Parliament, rather than judicial, intervention.[28]
AI as a Tool
The “AI as sophisticated tool” distinction identified by Lord Kitchin offers a pragmatic solution in the short term. Where a human researcher employs an AI system as an aid in the inventive process, while retaining meaningful creative control, there is little difficulty in recognising that individual as the inventor. In this respect, the patent system can readily accommodate AI in the same way it has long accommodated laboratory instruments, computational tools, and collaborative research environments.
However, this distinction becomes increasingly fragile as AI capabilities evolve. Systems such as DABUS represent only an early stage of autonomous machine creativity. As more advanced systems emerge, capable of producing highly complex and novel inventions with minimal human input, the boundary between tool and inventor begins to blur. At what point does the use of AI cease to be assistance and instead become the deployment of an independently inventive system? If a researcher merely inputs a general problem and the AI generates a complete technical solution, it becomes difficult to justify, in doctrinal terms, attributing inventorship to the human. The existing legal framework offers no clear or satisfactory resolution to this dilemma.
The Patent Gap
A central practical concern is what Ryan Abbott has described as the “innovation gap”: where AI-generated inventions fall outside the scope of patent protection, the incentive structure underpinning the patent system is weakened.[29] Companies investing heavily in autonomous AI research and development may find that their most valuable outputs are ineligible for patent protection. This is not due to any deficiency in novelty or inventiveness, but because no qualifying human inventor can be identified. This gives rise to two problematic outcomes: firms may choose to withhold AI-generated inventions from disclosure, thereby undermining the public disclosure function of the patent system, or they may misattribute inventorship by naming a human contributor who did not truly devise the invention.
Neither consequence advances the public interest. The former restricts the dissemination of technological progress, while the latter compromises the integrity of the patent register. As AI becomes increasingly embedded in fields such as pharmaceuticals, materials science, and engineering, these tensions are likely to intensify.
The Ownership Question
Even if the inventorship issue was resolved, a further question arises: who should own the rights to an AI-generated invention? Three principal models can be identified. The first assigns ownership to the developer of the AI system, drawing an analogy to employer-ownership rules in the context of employee inventions.[30] The second attributes ownership to the user – the individual or entity that deploys the system to generate the invention.[31] The third contemplates a sui generis approach, under which AI-generated inventions would automatically enter the public domain, potentially subject to a shorter period of protection, on the basis that the traditional patent bargain, reward in exchange for human creativity, does not apply where no human inventor exists.[32]
Each model entails significant trade-offs. Allocating rights to developers risks concentrating substantial patent power in the hands of a limited number of AI firms. A user-based approach may prove difficult to administer in practice and could incentivise strategic reliance on AI systems to circumvent inventorship requirements. By contrast, a public domain model, while conceptually coherent, may dampen investment in AI research and development by removing the commercial incentive to disclose innovations.[33]
- The Path Forward
WIPO has engaged in consultations on AI and intellectual property since 2019. In its revised Issues Paper of 2020, WIPO identified AI inventorship as one of the most pressing challenges confronting the global IP system, inviting member states to consider whether existing frameworks require amendment.[34] To date, no binding international consensus has been reached.
The current ‘patent gap’ necessitates a choice between three distinct reform paths. One approach would involve a targeted amendment to national patent legislation, introducing a category of “AI-assisted invention” that mandates disclosure of the AI system used while allowing the human operator or developer to be named as the nominal inventor. This preserves the human-inventor principle while acknowledging AI’s substantive role. A second option is a sui generis right, analogous to the European database right, granting short-term protection to the owner of an autonomous AI system without engaging the traditional inventorship doctrine. A third and more ambitious approach would involve an international treaty under WIPO guidance to harmonise the treatment of AI-generated inventions, reducing the potential for regulatory arbitrage highlighted by the DABUS litigation.[35]
South Africa is well-positioned to contribute meaningfully to this debate, particularly within the African context. Its Intellectual Property Phase I committed to reviewing the patent system to ensure alignment with development objectives. The planned introduction of substantive search and examination (SSE) of patents, which is still awaiting legislative finalisation, will compel South Africa to confront the inventorship question more directly than its current formalities-based system allows.[36] This presents an opportunity to adopt a progressive and clearly reasoned approach to AI inventorship, one that could shape the broader African Union IP agenda.[37]
Conclusion
The DABUS litigation has performed a vital function: it has compelled patent systems worldwide to confront a question for which they were never designed. Courts in major jurisdictions have responded in a cautious and predictable way, confirming that existing statutory frameworks do not accommodate non-human inventors. From a strictly legal perspective, these decisions are difficult to fault, yet they do little to resolve the underlying issue.
As AI systems become more capable, the gap between legal doctrine and technological practice is likely to widen. Maintaining a rigid human-inventor requirement may prove increasingly difficult to justify, particularly in cases where human involvement is minimal or difficult to characterise in traditional terms. At the same time, any move away from that requirement raises complex questions about ownership, accountability, and the purpose of the patent system itself.
South Africa’s decision to grant a DABUS patent, notwithstanding its procedural limitations, at least acknowledged that the current legal framework may be ill-equipped to address emerging technological realities. As the Fourth Industrial Revolution accelerates, patent regimes that remain rigidly tied to a human-inventor requirement risk marginalising themselves from the most innovative sectors of the global economy. What is required is not mere reinterpretation, but legislative innovation: a willingness to rethink and redesign the foundational architecture of patent law.
What emerges from the DABUS cases is not a settled answer, but a clear need for further development. Whether through legislative reform, judicial adaptation, or international coordination, patent law will need to engage more directly with the realities of AI-driven innovation. The challenge is not simply to accommodate new technology, but to do so in a way that preserves the underlying objectives of the patent system while remaining workable in practice.
Bibliography
Cases
Commissioner of Patents v Thaler [2022] FCAFC 62 (Full Federal Court of Australia).
Thaler v Comptroller-General of Patents, Designs and Trade Marks [2023] UKSC 49.
Thaler v Vidal, 43 F.4th 1207 (Fed. Cir. 2022), cert. denied, 143 S. Ct. 1783 (2023).
Yeda Research and Development Co Ltd v Rhone-Poulenc Rorer International Holdings Inc [2007] UKHL 43.
Legislation
Patents Act 1990 (Cth) (Australia).
Patents Act 57 of 1978 (South Africa).
Patents Act 1977 (United Kingdom).
35 USC § 100 (United States).
Secondary sources
Abbott R, The Reasonable Robot: Artificial Intelligence and the Law (Cambridge University Press 2020).
Intellectual Property Policy of the Republic of South Africa: Phase I (Department of Trade and Industry, 2018).
Locke J, Two Treatises of Government (first published 1689, Cambridge University Press 1988).
Presidential Commission on the Fourth Industrial Revolution, Report (South Africa, 2020).
South African Companies and Intellectual Property Commission, Patent No. 2021/03242, published in the South African Patent Journal (28 July 2021).
Spoor & Fisher, ‘DABUS, the Rise of the Inventive Machines’ (2021) <https://spoor.com/dabus-the-rise-of-the-inventive-machines/> accessed 28 March 2026.
Thadlar T and Naidoo D, ‘Artificial Intelligence as Inventor: A South African Perspective’ (2021) 34 South African Intellectual Property Law Journal 1.
WIPO, ‘Revised Issues Paper on Intellectual Property Policy and Artificial Intelligence’ (WIPO/IP/AI/2/GE/20/1 REV, May 2020).
[1] Thaler v Vidal, 43 F.4th 1207 (Fed. Cir. 2022), cert. denied, 143 S. Ct. 1783 (2023).
[2] Thaler v Comptroller-General of Patents, Designs and Trade Marks [2023] UKSC 49.
[3] Commissioner of Patents v Thaler [2022] FCAFC 62 (Full Federal Court of Australia).
[4] South African Companies and Intellectual Property Commission, Patent No. 2021/03242, published in the South African Patent Journal, 28 July 2021.
[5] WIPO, ‘Revised Issues Paper on Intellectual Property Policy and Artificial Intelligence’ (WIPO/IP/AI/2/GE/20/1 REV, May 2020) 4–5.
[6] Thaler v Comptroller-General of Patents, Designs and Trade Marks (n 2).
[7] UK Patents Act 1977, ss 7 and 13.
[8] South African Patents Act 57 of 1978, s 30.
[9] Thaler v Comptroller-General of Patents, Designs and Trade Marks (n 2).
[10] Thaler v Vidal (n 1).
[11] 35 USC § 100(f)–(g) (United States).
[12] Thaler v Vidal (n 1).
[13] Thaler v Comptroller-General of Patents, Designs and Trade Marks (n 2).
[14] Yeda Research and Development Co Ltd v Rhone-Poulenc Rorer International Holdings Inc [2007] UKHL 43.
[15] Thaler v Comptroller-General of Patents, Designs and Trade Marks (n 2).
[16] Thaler v Comptroller-General of Patents, Designs and Trade Marks (n 2).
[17] Thaler v Comptroller-General of Patents, Designs and Trade Marks (n 2).
[18] Thaler v Comptroller-General of Patents, Designs and Trade Marks (n 2).
[19] Thaler v Comptroller-General of Patents, Designs and Trade Marks (n 2).
[20] South African Companies and Intellectual Property Commission (n 4).
[21] South African Patents Act 57 of 1978 (n 8).
[22] Spoor & Fisher, ‘DABUS, the Rise of the Inventive Machines’ (2021) <https://spoor.com/dabus-the-rise-of-the-inventive-machines/> accessed 28 March 2026.
[23] Tshiamo Thadlar and Desigan Naidoo, ‘Artificial Intelligence as Inventor: A South African Perspective’ (2021) 34 South African Intellectual Property Law Journal 1.
[24] Intellectual Property Policy of the Republic of South Africa: Phase I (Department of Trade and Industry, 2018).
[25] Presidential Commission on the Fourth Industrial Revolution, Report (South Africa, 2020).
[26] John Locke, Two Treatises of Government (first published 1689, Cambridge University Press 1988) bk II, ch 5.
[27] Ryan Abbott, The Reasonable Robot: Artificial Intelligence and the Law (Cambridge University Press 2020) 56.
[28] Thaler v Comptroller-General of Patents, Designs and Trade Marks (n 2).
[29] Ryan Abbott (n 27) 61.
[30] Patents Act 57 of 1978, s 59.
[31] Ryan Abbott (n 27).
[32] WIPO (n 5).
[33] Ryan Abbott (n 27).
[34] WIPO (n 5).
[35] WIPO (n 5).
[36] Intellectual Property Policy of the Republic of South Africa: Phase I (n 24).
[37] Presidential Commission on the Fourth Industrial Revolution (n 25).
