Introduction
Artificial Intelligence’s (“AI“) rapid development presents both innovation opportunities and legal challenges, especially in intellectual property (“IP“). As AI systems produce increasingly valuable outputs, questions arise about who, if anyone, can claim ownership, particularly from an originality and authorship perspective under the South African Copyright Act 98 of 1978 (“the Copyright Act“).
We explore how AI-generated works interact with copyright rules, especially in cases where AI tools are the primary creators and the requirements of authorship and originality become particularly contentious. For copyright protection to apply, the legal requirements of authorship and originality must be met, enabling a party to claim rights over the work. Without clear ownership or the ability to establish authorship, there is no legal basis to prevent third parties from using such works without authorization.
Authorship
The definition of “author” in the Copyright Act identifies the author of computer-generated works as the person who made the necessary arrangements for the creation of the work. This enables AI-generated content to potentially qualify for copyright protection, provided a human directed the process.
South African case law, including Payen Components SA Ltd v Bovic CC and Haupt v Brewers Marketing Intelligence (Pty) Ltd, distinguishes between computer-aided works (where the human retains authorship by using the computer as a tool) and computer-generated works (where minimal human input complicates authorship claims), with the legal focus of the latter shifting to whether the user providing the inputs or instructions qualifies as the author.
It is argued that AI developers could be considered authors based on their role in designing AI tools but this view is not yet settled in South African law. Current legal reasoning leans toward recognizing the human user as the author (providing instructions to the AI tool). However, granting authorship to AI developers could lead to impractical outcomes and undermine the core principles of copyright law and other intellectual property laws which seek to encourage creativity and innovation in exchange for a monopoly for a limited period of time.It is critical to note that authorship does not shield individuals who use generative AI tools from copyright infringement risks. Generative AI systems are trained on large datasets that often include copyrighted material, raising the possibility that AI-generated content may be “substantially similar” to protected works. In copyright law, this similarity can constitute infringement even without direct copying. Thus, users may be deemed authors of works that unintentionally infringe on existing copyrights, exposing them to legal liability despite their creative input.
Originality
Originality is a core requirement for copyright protection under the Copyright Act, which, while not requiring novelty, demands that a work result from a human author’s own skill, effort, and intellectual judgment. This presents challenges for AI-generated content where human involvement is minimal, such as simply inputting a prompt and making superficial edits, as such works are unlikely to meet the originality threshold.
Although works reflecting meaningful human input may still qualify, what constitutes “sufficient” involvement remains a fact-specific legal question. The issue is further complicated by the way generative AI systems are trained on vast datasets, often incorporating pre-existing copyrighted materials, which raises concerns about derivative works and the originality of outputs. Assigning copyright to AI system developers risks absurd outcomes, akin to granting software vendors rights over all user-generated content, while granting it to users based on minimal input could dilute the value of genuine creative effort. This tension highlights the difficulty of applying traditional copyright principles to AI-generated works and underscores the need for legal clarity. Ultimately, the current legal framework suggests that works lacking meaningful human input are unlikely to qualify for copyright protection under South African law.
Foreign Law
In the United States (“US“), copyright protection for AI-generated content is determined by the extent of human creative input. The US Copyright Office maintains that only works having human authorship are eligible for protection, a stance reinforced by case law such as the “monkey selfie” decision and a 2023 federal ruling rejecting copyright for an image created solely by AI.
While AI can be used as a tool, such as for editing or arranging content, mere reliance on detailed prompts is generally insufficient to establish authorship. However, when a human meaningfully selects, modifies, or integrates AI-generated material into a broader creative work, those contributions may qualify for protection. This approach has already led to the creation of hybrid works, such as comic books combining human-authored text with AI-generated images.
The European Union (“EU“) adopts a human-centric approach, requiring that a work reflect the author’s free creative choices and intellectual inventiveness. This presents challenges for AI-generated content, particularly where human input is minimal. Article 50 of the EU Artificial Intelligence Act introduces transparency obligations for generative AI providers but does not resolve the issue of copyright ownership. As a general rule, EU law excludes fully autonomous AI outputs from protection, while AI-assisted works may qualify if the user’s contribution is sufficiently creative. Where human involvement is limited to prompting with minimal editing, the resulting work is unlikely to meet the originality threshold or qualify for copyright.
In the UK case of Nova Productions Ltd v Mazooma Games Ltd, the court had to consider the authorship of computer-generated works. The definition of “author” of computer generated works in the UK Copyright, Designs and Patents Act mirrors the definition in the Copyright Act (being the person who makes the arrangements necessary for the creation of the work). The court considered the question of who the author of a computer-generated artistic work was which consisted of bitmap files which were graphics (created by one of the directors) and found that the frames were computer-generated artistic works and held that the director had undertaken the arrangements necessary for the creation of the work as he designed the appearance of the elements of the game, the rule and logic of the computer program and he wrote the program that generated the artistic work. China has taken a progressive stance, recognising copyright in AI-assisted works where the human user demonstrates sufficient intellectual input. In the case of Li v Liu, decided in 2023, the Beijing Internet Court found that an image created using Stable Diffusion was copyrightable because the user’s prompt engineering and aesthetic decisions reflected original intellectual achievement. Similarly, in Tencent v Shanghai Yingxun, the court upheld copyright in an AI-generated article, attributing authorship to Tencent as it had creative control over the process. While Chinese courts have affirmed protection for AI-assisted works, they have not yet ruled on the copyrightability of fully autonomous AI outputs. Nonetheless, the prevailing position in the country is that copyright vests in the natural or juristic person who exercises creative direction over the AI system.
Evaluation and the Way Forward
A consistent theme across the US, EU, and China is the emphasis on meaningful human involvement as a prerequisite for copyright protection in AI-generated content.
Each jurisdiction recognises that when a human exercises sufficient control, such as selecting inputs, refining outputs, or integrating AI-generated material into a broader creative work, authorship and ownership may be validly attributed. Conversely, works created autonomously by AI, without human creative input, generally fall outside the scope of copyright protection. While the specific thresholds for originality may differ, the core requirement of substantive human authorship remains universally upheld.
This international consensus offers valuable guidance for South Africa, particularly as the Copyright Act’s requirements of authorship and originality. Though our courts have yet to directly confront the issue, existing legal principles suggest that AI-assisted works with significant human input could qualify for protection as a computer-generated work. In contrast, entirely AI-generated works are unlikely to meet the standards of authorship and originality, both of which demand human skill, labour, and intellectual contribution.
Granting copyright to AI developers would also be problematic, as it would extend rights to parties who did not create the specific work or exercise creative control, an outcome that undermines the fundamental purposes of copyright law.
These positions remain speculative in the absence of statutory or judicial clarity. As AI technology continues to evolve rapidly, South Africa must adopt a proactive legal stance. This should provide certainty, align with the foundational principles of the Copyright Act, and reflect a balanced understanding of modern technological realities.
Drawing on international approaches, such as the US’s insistence on human authorship, the EU’s emphasis on intellectual effort, and China’s recognition of human-guided AI outputs, can help shape a workable and principled framework. Notably, China’s case-by-case method, which considers prompt selection, parameter refinement, and aesthetic judgment, offers a practical standard for assessing authorship. By grounding copyright protection in demonstrable human creativity and control, South Africa can maintain a balanced and credible system that rewards genuine intellectual effort while avoiding the pitfalls of extending rights to purely algorithmic outputs.
Notwithstanding the above, successfully claiming authorship does not resolve the underlying infringement risk. Because AI models train on existing, often copyrighted, data, their outputs may unintentionally replicate protected elements to a degree that constitutes infringement. Creators must therefore remain vigilant, as they could be held liable if their AI-assisted work is deemed substantially similar to an existing copyrighted piece, irrespective of their own creative contributions.
Therefore, when using AI systems or tools to create works, it is advisable to maintain detailed records of the human input involved in the creative process. While such documentation may help support claims of authorship and ownership, it is important to note that courts assess these matters on a fact-specific, case-by-case basis, and there is no guarantee that such records will be accepted as proof of copyright entitlement.
Authors: Janine Hollesen (Director), Preeta Bhagattjee (Director) & Malique Ukena (Candidate Attorney) at LEX Africa member Werksmans Attorneys, South Africa. www.werksmans.com
