Today, well-known generative AI systems such as ChatGPT and DALL·E can produce text, images, and code. These systems, often referred to as large language models (LLMs), operate by processing user inputs (or prompts) and generating outputs based on vast datasets and learned patterns. While impressive in capability, they challenge traditional notions of authorship.
In South Africa, copyright protection arises automatically once certain requirements are met. The work must be original and must be reduced to a material form, such as writing, recording, or digital storage. Unlike some forms of intellectual property, most works eligible for copyright protection do not require registration.
The owner of copyright enjoys broad rights, including the ability to reproduce, distribute, perform, or license the work, typically for financial gain. While not legally required, creators often indicate their ownership by including a notice such as “©” or “copyright reserved” on the work.
Copyright protection extends to a wide range of works, including literary works, artistic creations, musical compositions, films, sound recordings, broadcasts, and computer programs.
The position becomes more complex when considering AI-generated content. South African copyright law recognises that works generated by computers may be eligible for protection. However, the key issue is determining who should be regarded as the author, and therefore the owner, of such works.
Several possibilities arise: ownership could vest in the developer who created the AI system, the company that owns or operates it, or the user who provides the prompt that generates the output.
Under current interpretations of South African law, the user is generally regarded as the most likely copyright owner, on the basis that they direct the creation process through their input. However, this approach is not without controversy. Where the user’s contribution is minimal, for example, where a simple or generic prompt produces a highly complex output, it becomes difficult to justify full ownership based on human creativity alone.
To address this uncertainty, many AI service providers include provisions in their terms and conditions stating that users retain ownership of the content generated through their prompts. While helpful in practice, these contractual terms do not fully resolve the underlying legal ambiguity.
As AI systems become more sophisticated, they are increasingly capable of generating content with little or no meaningful human input. This creates a growing tension within copyright law, which has traditionally been built on the premise of human authorship and originality.
The central challenge, therefore, is whether it remains appropriate to attribute ownership to a human user in circumstances where the creative contribution is minimal or indirect. If AI can independently produce complex works, the law must grapple with whether such outputs should be protected at all, and, if so, in whose hands those rights should rest.
AI is pushing the boundaries of traditional copyright principles, particularly regarding authorship and ownership. While South African law provides a framework for protecting computer-generated works, it leaves important questions unanswered in an era of increasingly autonomous systems.
As technology continues to evolve, legal reform or clearer judicial guidance may be required to ensure that copyright law remains both fair and relevant. Until then, the ownership of AI-generated content will remain a developing and, at times, uncertain area of law.
Disclaimer: This article is the personal opinion/view of the author(s) and does not necessarily present the views of the firm. The content is provided for information only and should not be seen as an exact or complete exposition of the law. Accordingly, no reliance should be placed on the content for any reason whatsoever, and no action should be taken on the basis thereof unless its application and accuracy have been confirmed by a legal advisor. The firm and author(s) cannot be held liable for any prejudice or damage resulting from action taken based on this content without further written confirmation by the author(s).
