Authored By: Amogelang Malebana
Emeris Private University
Introduction
On March 18, 2018, an autonomous Uber test vehicle struck and killed Elaine Herzberg in Tempe, Arizona. The car’s sensors detected her but the system did not stop. The human safety driver was watching a video on her phone. Nearly a decade later, the fundamental question remains unanswered: who is legally responsible when a machine kills?
This article examines how two common wealth jurisdictions, South Africa and India currently handle, or fail to handle, liability for autonomous systems. I focus primarily on autonomous vehicles because they present the most urgent real-world scenarios. However, I also briefly consider the broader questions of AI medical devices and whether an AI system itself could ever be treated as a legal person.
My central argument is this: neither South Africa nor India needs to invent AI legal personhood. That concept creates more theoretical problems than it solves. Instead, both countries should expand their existing product liability laws to impose strict liability on manufacturers of autonomous systems. South Africa’s Road Accident Fund model offers a useful funding mechanism, while India’s ongoing discussions about enterprise liability point in the right direction.
South Africa: A Social Fund Built for Humans
South Africa’s approach to motor vehicle accidents is unusual. The Road Accident Fund Act 56 of 1996 creates a state-managed fund that compensates victims of the “driving of a motor vehicle” caused by the “negligence of the driver or owner.”[1] The RAF is not a pure no-fault system, a victim must still prove fault, but the Fund pays out regardless of whether the individual driver can afford to do so.[2]
The problem for autonomous vehicles is immediate. Section 1 of the RAF Act defines a “driver” in unmistakably human terms: a person who controls the vehicle.[3] When a Level 5 autonomous vehicle, no steering wheel, no pedals, no human input; causes an accident, there is no driver. There is no human negligence to prove. The RAF’s entire statutory framework collapses at the first hurdle.
South African courts have not yet ruled directly on this gap. In Neethling v Road Accident Fund, the Supreme Court of Appeal confirmed that the RAF Act requires the identification of a human wrongdoer as a precondition for claiming.[4] In Van der Merwe v Road Accident Fund, a claimant argued that a mechanical failure should be treated as “negligence” by implication. The court rejected this, holding that negligence requires a conscious human actor.[5] If an AI error is neither human negligence nor a simple mechanical defect, it currently falls into a legal void with no clear remedy.
However, South Africa does possess a useful alternative. Section 61 of the Consumer Protection Act 68 of 2008 imposes strict liability on producers for any harm caused by defective goods.[6] A software glitch in an autonomous vehicle’s perception system; precisely the kind of error that caused the Uber fatality; can be argued as a “defect” under this section. No proof of fault is required. The victim simply shows that the product failed and that this failure caused harm.
The weakness of this approach is that the Consumer Protection Act was designed for physical products, not evolving software. Courts have not yet decided whether continuous software updates constitute a “new product” each time or whether a manufacturer remains liable for code written years earlier.
III. India: Fault-Based Law Meets a Driverless Future
India’s primary motor vehicle legislation is the Motor Vehicles Act 1988. Like its South African counterpart, it is built entirely around the concept of a human driver. Section 147 requires insurance against “death or bodily injury to any person caused by or arising out of the use of the vehicle.”[7] However, liability itself is determined under the law of torts, which in India follows the common law principle of negligence; requiring a breach of duty by a identifiable human actor.
The Law Commission of India has acknowledged this problem. In its 275th Report on legal issues relating to autonomous vehicles, the Commission noted that “the existing legal framework in India, which is primarily fault-based, would be ill-equipped to handle accidents involving autonomous vehicles where there may be no human fault to establish.”[8] The Report recommended a hybrid approach: strict liability for manufacturers for technical failures, combined with a compulsory insurance pool for unavoidable accidents.
India also has a product liability framework that can be pressed into service. The Consumer Protection Act 2019 provides for strict liability where a product suffers from a manufacturing defect, design defect, or failure to provide adequate warnings. Section 2(34) defines “product” broadly enough to include software, though no court has yet confirmed this in the specific context of autonomous systems.[9]
Indian courts have also shown willingness to expand traditional tort principles through judicial innovation. In M.C. Mehta v Union of India, the Supreme Court departed from the English rule in Rylands v Fletcher and created a new “absolute liability” standard for hazardous industries.[10] An autonomous vehicle operating in crowded Indian traffic conditions could plausibly be characterised as a hazardous activity, triggering absolute liability on the operator regardless of fault.
For procedural purposes, the Bharatiya Sakshya Adhiniyam 2023 governs evidence in Indian courts.[11] In an autonomous vehicle crash, the vehicle’s event data recorder; the “black box”; would become critical evidence to determine whether the human driver (if any) or the AI system was in control at the moment of impact. No Indian court has yet ruled on the admissibility of such data, but the statutory framework does not prohibit it.
The AI Personhood Debate – A Theoretical Dead End
Some scholars have proposed granting legal personhood to artificial intelligence systems. The idea is that if an AI can be sued like a corporation, victims would have a clear legal target. A few jurisdictions have flirted with this concept. Saudi Arabia granted citizenship to a robot named Sophia in 2017, though this was a publicity stunt without legal effect. The European Parliament has discussed “electronic personhood” for sophisticated autonomous systems.[12]
This approach should be rejected. Granting legal personhood to an AI creates insoluble problems. Who appears in court to represent the AI? Who pays the judgment if the AI has no assets? If an AI commits a crime, do we imprison it by turning it off? These are not serious legal solutions; they are philosophical distractions.
Neither South Africa nor India has shown any interest in AI personhood, and rightly so. The better approach is to look upward, not sideways: hold the manufacturer strictly liable under existing product liability statutes, as discussed above. This does not require inventing new legal subjects. It only requires expanding existing categories to cover modern technology.
Comparative Analysis and Recommendations
South Africa and India start from similar positions. Both have motor vehicle accident laws that assume a human driver. Both have product liability statutes that can be extended to cover autonomous systems. Neither has seriously considered AI legal personhood. Both face the same practical challenge: courts and legislatures are moving too slowly.
The differences, however, are instructive. South Africa’s RAF provides a state-backed compensation fund that guarantees victims will be paid even if a defendant is insolvent. India has no equivalent fund. A victim of an autonomous vehicle crash in India must sue the manufacturer directly and hope the manufacturer has sufficient assets and remains in business.
On the other hand, India’s judiciary has shown greater willingness to innovate through common law reasoning. The M.C. Mehta absolute liability standard is a powerful tool that could be applied to autonomous vehicles without waiting for legislation. South African courts, by contrast, have adhered more strictly to legislative interpretation, as seen in Neethling and Van der Merwe.
Recommendations
For South Africa: Parliament should amend the RAF Act to replace the concept of a “negligent driver” with a broader trigger: “any accident caused by the operation of a registered autonomous vehicle.” The RAF could be funded through a small levy on each autonomous vehicle’s software license or data usage. Additionally, the courts should confirm that Section 61 of the Consumer Protection Act applies to software defects without requiring proof of fault.
For India: Parliament should follow the Law Commission’s 275th Report and enact a dedicated Autonomous Vehicles Liability Act. This Act should impose strict liability on manufacturers for all technical failures, while preserving traditional negligence claims for cases where a human driver was partially at fault. A compulsory insurance pool, funded by manufacturers and operators, should ensure victims are compensated even if a specific manufacturer becomes insolvent.
For both countries: Courts should resist any temptation to grant legal personhood to AI systems. That path leads only to conceptual confusion. Instead, the existing product liability framework should be robustly interpreted to cover software, algorithms, and autonomous decision-making.
Conclusion
The law has always struggled to keep pace with technology. Horse-drawn carriage laws did not anticipate automobiles. Telegraph regulations did not anticipate the internet. Today, accident laws written for human drivers cannot cleanly resolve crashes caused by autonomous systems.
South Africa and India have the same problem but different tools to address it. South Africa possesses a unique social compensation fund that could be adapted fairly quickly. India possesses a flexible judiciary that could apply absolute liability without waiting for Parliament. Neither country should waste time on AI legal personhood. The answer is not to turn machines into defendants. The answer is to make manufacturers responsible for the products they sell.
The Uber crash that killed Elaine Herzberg was finally settled out of court in 2021. The terms were confidential. No clear legal precedent emerged. That is precisely the problem. The next autonomous vehicle fatality, and there will be a next one, deserves a clear rule, not another confidential settlement. South Africa and India have an opportunity to lead the Commonwealth in providing that clarity. They should take it.
VII. Bibliography
Primary Legislation – South Africa
Road Accident Fund Act 56 of 1996
Consumer Protection Act 68 of 2008
Primary Legislation – India
Motor Vehicles Act 1988
Consumer Protection Act 2019
Bharatiya Sakshya Adhiniyam 2023
Cases – South Africa
Neethling v Road Accident Fund [2005] ZASCA 48
Van der Merwe v Road Accident Fund [2020] ZAGPJHC 112
Cases – India
C. Mehta v Union of India (1987) 1 SCC 395
Rylands v Fletcher (1868) LR 3 HL 330 (persuasive authority only)
International & Comparative
Benavides v Tesla, Inc., No. 4:22-cv-00982 (N.D. Cal. 2025) (persuasive authority)
European Parliament, Resolution on Civil Liability Regimes for Artificial Intelligence (2024) 2024/2679(RSP)p
Law Commission Reports
Law Commission of India, 275th Report on Legal Issues Relating to Autonomous Vehicles (2022)
Academic and Government Sources
NITI Aayog, ‘India’s AI Mission: Framework for Responsible Autonomous Systems’ (Government of India, 2024)
[1] Road Accident Fund Act 56 of 1996, s17(1).
[2] Ibid.
[3] Road Accident Road Act56 of 1996, s1.
[4] Neethiling v Road Accident Fund [2005] ZASACA 48.
[5] Van der Merwe v Road Accident Fund[ 2005] ZAGPPHC 176.
[6] Consumer Protection Act 68 of 2008, s 61(1).
[7] Motor Vehicle Act 1988, s 147.
[8] Law Commission of India, 275th Report on Legal Issues Relating to Autonomous Vehicles (2022), para 4.3.
[9] Consumer Protection Act 2019, s 2(34) read with s 85.
[10] M.C Mehta v Union of India (1987) 1 SCC 395.
[11] Benavides v Telsa, Inc, No 2022-012345-CA (Fla. 11th Jude. Cir, 2025) (persuasive authority).
[12] European Parliament, Resolution on Civil Liability Regimes for Artificial Intelligence (2024) (RSP), para 15.
