Self-Driving Cars (SDCs) pose unique issues under South African law, specifically relating to liability for accidents caused by SDCs. Liability under the common law for accidents caused by SDCs was discussed in a previous blog, specifically from a product liability perspective.
In this blog, the focus will be on a liability claim under the Consumer Protection Act 68 of 2008 (CPA) and whether there is an avenue of recourse under the CPA.
Product liability
Internationally, the move has been towards strict liability for defective products under the umbrella term of "product liability". The South African legislature has attempted to codify strict liability for product defects to a certain extent in the Consumer Protection Act. It is important to note that section 2(10) of the CPA clearly states that the statute does not abolish the common law claim for damages arising from defective products. A claim may still be brought in terms of the common law, but it could be preferable to institute a claim in terms of the CPA.
The following section will focus on whether a claim can be brought in terms of the CPA for motor vehicle accident damages caused by a malfunctioning or defective artificial intelligence (AI).
Strict liability under the CPA
Strict liability is a type of liability that excludes fault. This means that negligence is not required for liability to be attributed to the producer or manufacturer.
Section 61 of the CPA provides for strict liability for goods. This strict liability is attributed to any importer, producer, distributor, or retailer. Products that are unsafe, defective, or hazardous are included in this scope. Section 61(1)(c) extends the scope even further by stating that inadequate instructions or warnings also have the effect of attributing strict liability to the manufacturer. It is interesting to note that the definition of "goods" specifically includes software and code, while section 61(5)(c) of the CPA includes immovables in its scope. It can thus be readily assumed that an AI is covered in the concept of a "good", which renders the strict liability imposed in section 61 of the CPA applicable to an AI — in this scenario applied to an AI driving an SDC.
Defences against strict liability
The logical assumption that follows from here is that a malfunctioning or defective AI will be subject to strict liability imposed in the CPA. However, section 64(4) of the CPA provides defences against a claim utilising the strict liability of section 61(1), or instances where strict liability will not be applicable. This includes if:
- the unsafe product characteristic, failure, defect, or hazard is attributable to the compliance with a public regulation;
- if the party is only involved on a marketing basis and cannot reasonably be expected to be aware of the defect;
- if the alleged unsafe product characteristic, failure, defect, or hazard did not exist at the time that it was supplied, then there arises no liability.
This raises questions about the liability of an AI that continuously evolves and artificially gathers intelligence.
A coder codes a baseline code in which the AI exists and provides ways in which it gathers intelligence. Suppose that the AI works on the same principles as Tesla and that millions of pieces of intelligence are uploaded to a central data holding facility. This data is then processed by the AI, whereafter an update is generated and sent to the cars. This update can then be installed manually or automatically by the user. There is thus little to no human interaction with the AI after the baseline code has been written, except for reviews of the baseline coding to facilitate better development of the AI.
The only way in which a defect can be flagged for possible amendment is when a real-world incident occurs that highlights the issue. This is due to the countless variances of real-world scenarios and the dependence on the AI and all the available data to make the right decision. In such a case, a defect did not exist up until the moment the car was involved in a motor vehicle accident.
Does the defect then exist at the point that the car is handed over to the seller?
If the AI gathered the data that led to the accident after the car was sold, the manufacturer or producer may have a complete defence against liability based on section 61(4)(b)(i). Not only does the wording of section 61(4)(b)(i) offer a potential complete defence, but Kriek, on p. 392 of her doctoral dissertation, suggests that a defendant (manufacturer) can prove a prima facie case by merely adducing sufficient evidence about their quality control measures.
Onus
The onus would then shift to the plaintiff to rebut this. Manufacturers are inevitably better disposed to present expert evidence than the people who suffered harm from the "goods". All these factors together may lead to many litigants opting not to follow litigation to recover damages. The word "supply" is, when relating to goods, defined in section 1 of the CPA as "selling, renting, exchanging or hiring in the ordinary course of business." As the defence in section 61(4)(b)(i) requires for a "good" to be "supplied", an automotive manufacturer or supplier can ensure the use of this defence by merely including the AI as being rented to the buyer on a subscription basis — an approach already adopted by Tesla.
Conclusion
The CPA does not provide a comprehensive framework, and there may be a need to adapt before full SDCs are launched in South Africa. Accidents will inevitably happen, and the legislation concerning liability will not be comprehensive enough to deal with these issues. The CPA was enacted to protect vulnerable consumers. However, consumers will be left vulnerable if the legislation does not evolve to protect them.
About the author
Hein Steenberg completed his BCom Law degree, followed by an LLB and LLM in mercantile law, with a specific focus on space law at the North-West University. Hein is currently pursuing his LLD in space law.
Last updated on 15 May 2026.