When a vehicle's automated driving system, rather than a human, performs the driving task, Australian liability law built around the human 'driver' struggles to assign, prove and insure responsibility for accidents. Mackie (2018) argued that human drivers will often not be liable for highly and fully automated vehicle accidents, so victims must pursue manufacturers and other suppliers under negligence and the product-liability provisions of the Australian Consumer Law. That analysis, and the ongoing Australian automated-vehicle reform programme, leave five interconnected issues unresolved: the meaning of 'legal control' and the definition of a 'driver' across automation levels; the baseline technical and regulatory safety-assurance standards that should govern testing and validation of automated driving systems; how the law should respond to ongoing concerns about the explainability of machine-learning decision-making, which complicate the proof of causation and fault; the interpretation or reform of the consumer-expectations test for defectiveness and the development-risks defence; and the design of any specialised strict-liability or compensation scheme. As of August 2026, the five questions remain unresolved: no national automated-vehicle liability or safety statute is in force in Australia, the reform program remains in development, and later scholarship still treats responsibility attribution and legally sufficient explanations of automated-driving decisions as open problems.

Background

An automated driving system (ADS) is the machinery — sensors, software and actuators — that performs part or all of the dynamic driving task (steering, acceleration, braking and monitoring the environment) without a human needing to supervise it. Under the widely used SAE J3016 taxonomy, levels 0–2 denote driver-assistance features that still require a human driver; levels 4 and 5 (commonly called 'highly' and 'fully' automated vehicles, or driverless cars) allow the human occupant to be a mere passenger, at least within the system's designed operating conditions. Australia's road-transport laws, its state-based compulsory third-party (CTP) injury insurance schemes and the common law of negligence all presuppose a human 'driver' who is in control of the vehicle, so the arrival of vehicles without an effective human controller strains the concepts through which fault and responsibility are allocated.

Mackie (2018), in the Computer Law & Security Review, analysed how liability for such accidents would be resolved under then-current Australian law. He argued that whether the human remains liable depends on the degree of automation and the degree of control the human had over the events leading to the accident, so for most level 4–5 accidents the plaintiff (an injured road user) would instead have to establish liability against those who manufacture, maintain or contribute to the operation of the vehicle — in negligence, or in a claim for defective goods under Part 3-5 of the Australian Consumer Law (ACL), the national consumer-statute regime that imposes liability on manufacturers for goods with a safety defect. Part 3-5's notion of defectiveness turns on what safety consumers are generally entitled to expect (a consumer-expectations test), and it supplies a defence where the state of scientific and technical knowledge at the time of supply was not such as to allow the defect to be discovered (the development-risks defence). Mackie identified two compounded proof problems: first, causation — the technical complexity of an ADS and the opacity of machine-learning decision-making make it hard to determine why a crash happened; second, fault or defect — because completely testing for and eliminating all ADS errors is impracticable, a manufacturer will typically be held only to a duty of reasonable testing, and plaintiffs may struggle to prove the testing unreasonable. The National Transport Commission (NTC), the Australian government's transport-reform advisory body, had proposed clarifications of driver concepts for enforcement purposes, but their extension to civil liability was unsettled.

Since 2018 Australian governments have been building an end-to-end national regulatory framework for automated vehicles through five NTC-led workstreams — control (who is legally 'driving' when an ADS is engaged), safety at first supply, in-service safety, insurance, and data (Burke 2020). As of 2024, however, only levels 0–2 were permitted on public roads, and commentators emphasised that comprehensive legislative reform was still required before higher-level vehicles could be commercially deployed (Yates and Siskovic 2024). No Commonwealth Act or legislative instrument with 'automated' in its title is in force, so the liability framework Mackie analysed has not been replaced by statute. Meanwhile the surrounding problems have only grown more visible: empirical work shows that people attribute more responsibility to the occupants of driverless cars than to taxi passengers, contrary to normative analyses (Zhai, Wang and Liu 2024); mixed human–machine error cases remain contested (Zhai, Gao, Wang and Liu 2023); legal theorists argue existing attribution doctrines mishandle artificially intelligent systems (Soh 2023); and quantitative methods for apportioning responsibility among crash stakeholders are still being actively developed (Wang, Xue, Jiang and Jia 2026). The question of how a legal system should allocate and prove responsibility for automated-vehicle accidents is accordingly a live research front internationally (Roussou, Ziakopoulos, Kontaxi and Yannis 2026).

Problem Statement

How should Australian law and regulation be reshaped so that liability for accidents caused by highly and fully automated vehicles (SAE levels 4–5 — vehicles whose automated driving system performs the entire driving task without human supervision, at least within its designed operating conditions) can be assigned, proved and compensated? Five interconnected unresolved questions jointly define this problem: (1) how should 'legal control' and the statutory or common-law definition of a 'driver' be clarified — across the different automation levels — for civil liability and insurance entitlements, given that the National Transport Commission proposed clarifications for enforcement but it remains unsettled whether those positions should extend to civil liability; (2) what baseline technical and regulatory safety-assurance standards for ADS testing and validation should be mandated, and by what process, so that courts and regulators can consistently assess the reasonableness of a manufacturer's testing as the technology matures; (3) how should Australian law and evidence rules respond to the difficulty of establishing causation and fault in litigation, given that the technical complexity of ADSs and ongoing concerns about the explainability of machine-learning decision-making make it hard to determine why an accident happened; (4) how should the consumer-expectations test for defectiveness and the Australian Consumer Law's development-risks defence be interpreted or reformed to balance victim compensation, preservation of innovation incentives and practicable testing burdens for manufacturers of adaptive systems; and (5) if a specialised strict-liability or state-funded compensation scheme is adopted, how should it be designed — its scope, funding sources, interaction with existing insurance and tort rules, and thresholds for recovery — to balance compensatory and allocative efficiency against moral hazard and innovation costs? The five questions share one scientific objective: a coherent legal-regulatory framework under which responsibility for automated-vehicle accidents can be established and injured parties compensated. A resolution must address all five sub-questions; a defensible negative answer to any sub-question counts as addressing it.

References

  1. Mackie, Tom (2018). 'Proving liability for highly and fully automated vehicle accidents in Australia.' Computer Law & Security Review 34(6):1314–1332. DOI: 10.1016/j.clsr.2018.09.002. https://doi.org/10.1016/j.clsr.2018.09.002
  2. Burke, Marcus (2020). 'Developing End-to-End Regulation for Automated Vehicles in Australia.' In: Lecture Notes in Mobility, pp. 23–29. Springer. DOI: 10.1007/978-3-030-52840-9_3. https://doi.org/10.1007/978-3-030-52840-9_3
  3. Yates, Angelika; Siskovic, Samuel (2024). 'Steering the Transition to Automated Vehicles in Australia: Current Status and Regulatory Changes Ahead.' In: Data Science, Machine Intelligence, and Law, pp. 207–237. Springer. DOI: 10.1007/978-3-031-41992-8_10. https://doi.org/10.1007/978-3-031-41992-8_10
  4. Zhai, Siming; Wang, Lin; Liu, Peng (2024). 'Not in Control, but Liable? Attributing Human Responsibility for Fully Automated Vehicle Accidents.' Engineering 33:121–132. DOI: 10.1016/j.eng.2023.10.008. https://doi.org/10.1016/j.eng.2023.10.008
  5. Zhai, Siming; Gao, Shan; Wang, Lin; Liu, Peng (2023). 'When both human and machine drivers make mistakes: Whom to blame?' Transportation Research Part A: Policy and Practice 170:103637. DOI: 10.1016/j.tra.2023.103637. https://doi.org/10.1016/j.tra.2023.103637
  6. Soh, Jerrold (2023). 'Legal dispositionism and artificially-intelligent attributions.' Legal Studies 43(4):583–602. DOI: 10.1017/lst.2022.52. https://doi.org/10.1017/lst.2022.52
  7. Roussou, Julia; Ziakopoulos, Apostolos; Kontaxi, Armira; Yannis, George (2026). 'Literature review of legal frameworks and liability allocation in connected and automated vehicles.' Research in Transportation Business & Management 68:101804. DOI: 10.1016/j.rtbm.2026.101804. https://doi.org/10.1016/j.rtbm.2026.101804
  8. Wang, Qiuling; Xue, Ziwei; Jiang, Wanke; Jia, Wenjian (2026). 'Responsibility Attribution for Autonomous Vehicle Crashes Based on Causal Inference: Case Study in California, USA.' Transportation Research Record. DOI: 10.1177/03611981261438023. https://doi.org/10.1177/03611981261438023