Authored By: Ntombifuthi Precious Tlapu
University of South Africa
Abstract
Artificial intelligence (AI) is changing how creative works are made and exposing the weaknesses of intellectual property (IP) laws that were written for humans. This article looks at the main problems raised by AI, especially in copyright law, which still depends on human authorship. Using examples from South Africa, the United States, European Union, China, Indonesia and the Czech Republic, it argues that current laws do not fully deal with AI-generated works. Giving AI legal rights is not practical, but refusing protection may discourage investment and creativity. The article concludes that clearer rules, limited rights for AI works, and openness about training data are needed to protect human creativity and guide future legal reform.
- Introduction
The rapid diffusion of digital technologies and machine learning tools has highlighted tensions between innovation and the legal protection of creative expression. AI systems including language models, image generators, and virtual assistants now regularly create text, code, images, and audio by learning from large collections of data that often include copyrighted material. This raises difficult questions about authorship, ownership, and liability when output resembles or reproduces antecendent works. The South African context currently lacks AI-specific IP legislation, but comparative examples show multiple approaches that can inform reform. This article therefore explores the problem space, asks why IP laws appear difficult to adapt to AI, and outlines potential pathways forward.
The main argument of this paper is that current intellectual property laws, especially copyright rules based on human authorship, are not suitable for works created by AI. A balanced mix of legal changes including clearer laws, narrow sui generis protection for1 certain AI-generated content, rules to improve data transparency, and better ownership guidance when humans play a meaningful role in directing AI would help protect creators and support ongoing cultural production.
- Research Methodology
This article follows a doctrinal method, focusing on the study of legal rules, statutes, and case law. The research draws on South Africa’s Copyright Act 98 of 1978, leading cases such as Haupt t/a Softcopy v Brewers Marketing Intelligence (Pty) Ltd and MoneyWeb (Pty) Ltd v Media 24 Ltd, as well as court decisions and laws from other jurisdictions, including the United States, European Union, China, and Indonesia. Academic works, policy papers, and commentaries are also used to support the analysis. By relying on authoritative sources, the methodology examines how intellectual property law currently applies to AI-generated works and highlights areas where South African law may require reform.
- Legal Framework
Intellectual property law covers a number of categories of intangible creations copyright, patents, trademarks, and trade secrets2and each category raises different challenges when AI is involved. Copyright traditionally requires an original expression fixed in a tangible form and has been interpreted to presuppose human authorship in many jurisdictions.3 South African copyright law under section 2(1) of the Copyright Act of 1978 protects literary works and computer programs where originality is present and eligibility of other works are not considred relevant to the matter at hand.4 At the level of principles, the key questions are who, if anyone, is the author of an AI-generated work, whether training on copyrighted datasets constitutes infringing use, and how exceptions such as ‘fair dealing’5 should apply to machine training and output.
- Judicial Interpretation
Judicial responses have been varied but consistent in one key respect which is a focus on human creativity. In South Africa, the Supreme Court of Appeal has affirmed that original literary works and computer programs are eligible for copyright protection when originality exists, the factual inquiry looks to human creative input. Likewise, the Johannesburg courts have applied ‘fair dealing’ standards in a way that requires attribution where an exception is claimed. 6Internationally, a Prague Municipal Court decision rejected copyright protection for an AI-generated image in which there was insufficient human creative involvement, aligning with United States precedent emphasising human authorship.7 China presents an interesting counterpoint. In Fellin v Baidu, the Beijing Internet Court examined an AI-assisted report and while finding the contested product not protectable as a traditional copyright work, the court recognised the investments of the user and developer and protected contractual and licensing arrangements.8 Instances such as Zarya of the Dawn and other disputes have reinforced the point that automated output without demonstrable human creative input which falls outside of copyright protection.9In Indonesia, AI is not recognised as a creator or copyright holder under Law No. 8 of 2014 on Copyright.10 However, I think when a human provides clear direction and makes creative choices that shape the final output, courts have generally been more open to recognising copyright in such cases.
- Critical Analysis
There are three main practical and legal challenges. Firstly, copyright law generally requires a human author. This creates a problem for AI-generated works, as they may not be protected under current laws. If these works aren’t protected, people might not see the value in investing in new creative projects. However, going too far in the other direction by giving legal rights to AI could be risky, since machines can’t take moral or financial responsibility.
Secondly, it’s often unclear what data is used to train AI systems. If the training data includes copyrighted material without permission, the AI might copy protected parts. This raises concerns about both copyright and privacy, especially because it’s difficult for users or others to tell whether an AI’s output is based on a specific existing work.
Thirdly, different laws in different countries make it hard to enforce rules consistently. AI models are often trained on data taken from many places around the world, which means they could be subject to several legal systems at once. As a result, platforms operating globally may struggle to keep up with all the legal requirements, and their responses to takedown requests might be inconsistent.
When looking at how different legal systems deal with these issues, four main priorities become clear such as protecting creators’ rights, encouraging innovation, ensuring someone is held responsible when things go wrong, and providing clear and manageable rules for developers and online platforms.
- Recent Developments
Policymakers and courts across the globe are actively debating targeted responses. The EU has signalled regulatory approaches that include transparency obligations for AI systems and potential tailored protections. 11 The US continues to rely on existing copyright doctrines, particularly fair use 12 while some legislators and commentators argue for statutory updates. China’s courts have, in some cases, recognised rights for AI-assisted works where human creative contribution is identifiable and have emphasised licensing frameworks to protect investments.13 Indonesia’s position remains conservative, treating a ‘creator’ as a human natural person or juridical entity, not AI.14 South Africa currently lacks AI-specific IP regulation, though data protection (POPIA) governs certain automated processing activities such as machine learning. 15 These developments suggest a likely move toward a mixed models such as clearer human authorship thresholds, limited sui generis protections in narrow circumstances, and transparency or licensing requirements for training data.
- Way Forward
Update the Law: Copyright rules should be changed to clearly explain who owns what when AI tools are involved in creating content.
Special Rights for AI Output: There could be limited, short-term legal protection for certain AI-generated content or datasets that have real commercial value.
Transparency in Training Data: Companies using AI models for high-risk commercial purposes should be required to clearly explain where their training data comes from.
Clear Guidance on Fair Use: Authorities should provide solid guidance on how existing copyright exceptions apply when AI systems are trained on protected works.
Cross-Disciplinary Oversight: Committees made up of legal experts, tech professionals, and members of the public should be set up to keep track of how AI is affecting intellectual property.
- Conclusion
Artificial intelligence is changing the way intellectual property law works. It creates works without human authors and often depends on large amounts of data that may include copyrighted material. This shows the limits of laws that were written for another time. Looking at other countries, we see that there is no single answer, but ideas such as clearer rules on authorship, short-term rights for some AI works, openness about training data, and stronger licensing systems are useful steps. For South Africa, change is especially important because without it both creators and developers face uncertainty.
This issue matters because intellectual property law is not only about money. It also protects human creativity, supports culture, and encourages innovation in a digital world. If AI continues to grow without proper legal reform, there is a risk that creators will not be treated fairly and people will lose trust in new technologies.
We cannot stop AI, but we can choose how it is used. The real question is whether our legal systems will rise to the challenge and guide AI in a way that protects creativity and fairness for the future.
9.Bibliography
Journal Articles
Alesia Zhuk, ‘Navigating the legal landscape of AI copyright: a comparative analysis of EU, US and Chinese approaches’ (2023) AI and Ethics
<https://link.springer.com/article/10.1007/s43681-023-00299-0> accessed 9 September 2025
Johnson Ong and Khai Yi Lo, ‘AI-Generated Work and Copyright Law: A Comparative Analysis of US and EU Perspective’ (2024) HHQ <https://hhq.com.my/posts/ai generated-work-and-copyright-law-a-comparative-analysis-of-us-and-eu-perspectives/> accessed 9 September 2025
Tasya Safiranita Ramli and others, ‘Artificial intelligence as object of intellectual property in Indonesian law’ (2023) 26(2) The Journal of World Intellectual Property <https://onlinelibrary.wiley.com/doi/full/10.1111/jwip.12264> accessed 10 September 2025
Case Laws
Haupt t/a Softcopy v Brewers Marketing Intelligence (Pty) Ltd 2006 (4) SA 458 (SCA) MoneyWeb (Pty) Ltd v Media 24 Ltd [2016] 3 All SA 193 (GJ); 2016 (4) SA 591 (GJ) Websites
Celeste Snyders, ‘Unpacking the legal side of Artificial Intelligence’ (GoLegal, 27 June 2022) <www.golegal.co.za/legal-artificial-intelligence> accessed 10 September 2025
European Parliament, ‘EU AI Act: first regulation on artificial intelligence’ (2023) <www.europarl.europa.eu/topics/en/article/20230601STO93804/eu-ai-act-first regulation-on-artificial-intelligence> accessed 10 September 2025
Georgetown Law, ‘Intellectual Property Law’ <www.law.georgetown.edu/your-life career/career-exploration-professional-development/for-jd-students/explore-legal careers/practice-areas/intellectual-property-law> accessed 8 September 2025
1 Alesia Zhuk, ‘Navigating the legal landscape of AI copyright: a comparative analysis of EU, US and Chinese approaches’ [2023] Springer link <https://link.springer.com/article/10.1007/s43681-023-00299-0> accessed 09 September 2025.
2 Georgetown Law, ‘Intellectual Property Law’ <www.law.georgetown.edu/your-life- career/career exploration-professional development/for-jd-students/explore-legal-careers/practice-areas/intellectual property-law> accessed 08 September 2025.
3Johnson Ong and Khai Yi Lo, ‘AI-Generated Work and Copyright Law: A Comparative Analysis of US and EU Perspective’ (2024) HHQ <https://hhq.com.my/posts/ai- generated-work-and-copyright-law-a comparative-analysis-of-us-and-eu- perspectives/ HHQ> accessed 09 September 2025. 4 Haupt t/a Softcopy v Brewers Marketing Intelligence (Pty) Ltd 2006 (4) SA 458 (SCA) [26].
5 MoneyWeb (Pty) Ltd v Media 24 Ltd [2016] 3 All SA 193 (GJ); 2016 (4) SA 591 (GJ) 194, 194. 6 The MoneyWeb Case.
7 Ong and Lo, ‘AI-Generated Work and Copyright Law: A Comparative Analysis of US and EU Perspective’ 8 Zhuk, ‘Navigating the legal landscape of AI copyright: a comparative analysis of EU, US and Chinese approaches’
9 Ong and Lo, ‘AI-Generated Work and Copyright Law: A Comparative Analysis of US and EU Perspective’ 10 Tasya Safiranita Ramli and others, ‘Artificial intelligence as object of intellectual property in Indonesian law’ (2023) 26(2) The J of World Intellectual Property <https://onlinelibrary.wiley.com/doi/full/10.1111/jwip.12264> accessed
10 September 2025.
11 European Parliament, ‘EU AI Act: first regulation on artificial intelligence’ <www.europarl.europa.eu/topics/en/article/20230601STO93804/eu-ai-act-first-regulation-on-artificial intelligence> accessed 10 September 2025.
12 Alesia Zhuk, ‘Navigating the legal landscape of AI copyright: a comparative analysis of EU, US and Chinese approaches’ [2023] Springer link <https://link.springer.com/article/10.1007/s43681-023-00299-0> accessed 09 September 2025.
13 Zhuk, ‘Navigating the legal landscape of AI copyright: a comparative analysis of EU, US and Chinese approaches’
14 Ramli and others, ‘Artificial intelligence as object of intellectual property in Indonesian law’
15 Celeste Snyders, ‘Unpacking the legal side of Artificial Intelligence’ (Go legal, 27 June 2022) <www.golegal.co.za/legal-artificial-intelligence> accessed 10 September 2025.
