- within International Law, Law Department Performance and Consumer Protection topic(s)
The convergence of Artificial Intelligence (AI), intellectual property (IP), and class action law is creating a new and complex risk landscape for businesses worldwide. Two recent developments epitomise this shift. In South Africa, the introduction of Rule 11A into the High Court Rules which came into effect on 19 September 2025, has formalised the pathway for class action certification, creating a predictable framework for collective redress. Simultaneously, landmark litigation in the United States, such as Bartz v Anthropic, is testing the legality of training Large Language Models ("LLMs") on copyrighted material, exposing developers to significant infringement claims. For businesses operating in or exposed to the South African market, the interplay between a restrictive copyright regime and a structured class action framework signals a heightened risk of high-stakes litigation. This article analyses these parallel developments and offers key strategic lessons for navigating this evolving legal frontier.
The Bartz v Anthropic case in the U.S. District Court for the Northern District of California provides a crucial, albeit jurisdiction-specific, lens into this conflict. The case involved a putative class action brought by authors alleging that Anthropic committed mass copyright infringement by training its AI model, Claude, on vast datasets that included their copyrighted books. The plaintiffs argued that this unauthorised copying for training purposes constituted infringement, regardless of whether the final output of the AI reproduced their work verbatim. This case directly confronts the central tension between an LLM's operational need for massive amounts of high-quality training data-with books being a prime source-and the fundamental rights of copyright holders.
The court's ruling was nuanced. While it found that certain activities, like training on lawfully acquired works, could fall under the U.S. doctrine of "fair use," it decisively rejected this defence for training on data sourced from pirate websites. The court held that building a permanent internal library from pirated books was not fair use, certifying this part of the claim for class action proceedings. Before the case could proceed to trial, the matter was settled and has received preliminary approval in the amount of USD1.5 billion, culminating in a landmark agreement that highlights the immense financial liability at stake. At USD1.5 billion, allocated across approximately 482,460 to 500,000 works, this averages about USD3,000 per book before deductions for fees and costs. The certification for class action status was pivotal, as it enabled the aggregation of thousands of individual claims into a single, high-value action, demonstrating the power of collective redress in holding AI developers accountable for their data sourcing practices.
The South African Class Action Framework: Rule 11A
In South Africa, the procedural gateway for such collective actions is Rule 11A of the High Court Rules. This rule mandates that no class action can proceed without court certification, which is granted only if it is in the "interests of justice." To secure certification, applicants must demonstrate several key factors: the existence of an identifiable class, a triable cause of action and common issues of law or fact that are central to the claims of all class members. Where damages are sought, a viable method for calculating and distributing them must also be presented.
Rule 11A also scrutinises the suitability of the class representative, who must be able to fairly represent the class, manage conflicts and competently conduct the litigation. The court determines the appropriate form of notice to potential members and whether the class will be "opt-in" or "opt-out." For respondents, this structured certification process offers clear grounds to challenge a proposed class action, particularly on the basis of weak commonality or an unsuitable representative. For applicants, it establishes a clear but rigorous roadmap for bringing a collective action.
Fair Use vs. Fair Dealing: A critical distinction for AI training
The legal outcome in Bartz hinged on the U.S. doctrine of "fair use," a flexible, four-factor test that allows for unlicensed use of copyrighted works for purposes like criticism, research and transformative uses. South Africa's Copyright Act contains no such general defence. Instead, it provides for "fair dealing," a significantly narrower concept that permits limited use only for specific, enumerated purposes such as private study, research, criticism or reporting current events. Crucially, there is no express exception for text and data mining or the wholesale ingestion of literary works for the purpose of training an LLM.
This distinction means the Bartz ruling would likely have been very different under South African law. Largescale ingestion of unlicensed books from any source - pirated or not - for commercial LLM training would almost certainly constitute infringement, as it does not fit neatly into any of the prescribed fair dealing exceptions.
Key lessons and strategic guidance
For AI developers, there is a clear and distinct advantage to conducting training activities in jurisdictions where a valid legal defence for such activities exists.
For copyright owners, while some AI training may be permissible, it is confined to specific guardrails. This presents an opportunity not only to police the infringement of your IP but also to potentially monetise it by licensing the use of your content to AI platforms for training purposes.
It is important to remember that piracy is still piracy, regardless of the intended purpose, and this holds true in any jurisdiction.
You should document your data sources by maintaining clear records of licensing and acquisition to defend against potential infringement claims.
To future-proof your strategy, you should expect a global convergence toward stricter copyright compliance and transparency in AI training.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
