When an autonomous vehicle injures someone, the victim typically cannot reach the vehicle's driving data or the manufacturer's development records, so the traditional rule that the claimant proves every element of the tort breaks down. The source paper (Chen, Cai and Wei, 2024) argues from German and EU law that China should shift the burden of proof for product defect and fault onto the manufacturer, and it leaves open how such a regime should be designed in detail. This problem asks how such a regime should be designed for China: how the burdens of proof for product defect, fault, and the causal link between defect and damage should be allocated among the parties; what disclosure and data-preservation duties must accompany the allocation so that the evidence victims need actually reaches the court; how to mitigate the adverse systemic consequences of burden shifting; and how the regime should respond to the takeover-time gap, to the obsolescence of law relative to evolving AI, and to multi-cause cases. The design question remains open: the EU's new Product Liability Directive introduces disclosure duties and presumptions applicable from December 2026, while the AI Liability Directive the paper relied on was withdrawn in February 2025, and China still has no national autonomous-driving liability statute.

Background

An autonomous (self-driving) vehicle is classified by the SAE J3016 industry standard into levels of driving automation. At Level 3 (conditional automation) the system drives itself within a defined operating domain, but a human fallback driver must be available to take over when the system asks; at Level 4 (high automation) the system is itself expected to manage the fallback within its domain. A takeover request is the system's prompt telling the Level 3 driver to resume manual control, and the takeover time is the interval within which the driver can reasonably be expected to do so safely. Human-factors experiments show that safe takeover depends on traffic complexity, driver state, and the warning given, with critical lead times of only a few seconds.

Tort (civil-damages) liability for a traffic accident in the civil-law tradition requires establishing damage, an act or product defect, causation, and, where the liability is fault-based, fault. The burden of proof has what German doctrine calls a dual meaning: it determines both which party must produce evidence for a contested fact and which party bears the loss if the fact remains unproven (the risk of non-persuasion). Under the classical normative theory of Leo Rosenberg, each party must prove the preconditions of the legal rule from which it claims a benefit, and the defendant must prove the preconditions of the rule it invokes as a defence.

German law, the comparator used by the source paper, distributes these burdens as follows. Section 823 of the German Civil Code (BGB) establishes fault-based liability. Section 18 of the Road Traffic Law (StVG) presumes the driver's fault in accidents, so the driver must exonerate themselves. Sections 1a and 1b StVG keep the person who activates a Level 3 or Level 4 function in the legal role of driver: they need not constantly watch the road, but must resume control promptly when the system requests it or when circumstances obviously require it - without any statutory definition of a numerical takeover time. Section 63a StVG requires an electronic driving recorder (a black box) that logs when the automated function was activated and when a takeover was requested or a technical failure occurred, so that these facts can be proven in court. The German Product Liability Act imposes strict liability for defective products but leaves the injured party to prove the defect and its causal link to the damage, and grants no general right to information. To close this gap, the European Commission proposed an Artificial Intelligence Liability Directive (AILD) with disclosure obligations and evidentiary presumptions, and a revision of the 1985 Product Liability Directive; these proposals were the state of the debate when the source paper was written.

In China, the Civil Code (articles 1202-1207) provides strict product liability for defective products, and road-traffic liability rules center on the driver and vehicle owner, but no national statute governs the evidentiary position in autonomous-driving accidents. Governance proceeds through national standards, admission pilot programs, and local regulation such as the Shenzhen regulation of 2022 and the Beijing Autonomous Vehicles Regulations of 2025, none of which spells out a burden-of-proof regime for litigation.

The unresolved difficulty is structural. Because driving algorithms are opaque and technical information sits with the manufacturer, victims face an information asymmetry that makes the defect and causation elements nearly unprovable. Simply shifting the burden of proof onto the manufacturer fixes this but creates countervailing risks: higher litigation and compliance costs, over-collection of data, chilling of innovation, and meritless claims. The source paper, 'Distribution of the Burden of Proof in Autonomous Driving Tort Cases: Implications of the German Legislation for China' (Chen, Cai and Wei 2024), concludes that China should shift the burden of proof for product defect and fault onto the manufacturer, leaving the victim only preliminary proof of a causal link between defect and damage, and it identifies the remaining design questions that such a regime must answer before it can be enacted.

Problem Statement

Design, for Chinese law, a detailed distribution of the burden of proof in autonomous-driving tort litigation. The objective is an evidentiary regime under which a court can decide an autonomous-driving accident case - who bears liability among the injured party, the driver/user, the vehicle holder, and the manufacturer, and who loses when a fact cannot be proven.

The source paper's core recommendation - shifting the burden of proof for product defect and fault onto the manufacturer, leaving the victim only preliminary proof of a causal link between defect and damage - leaves open how such a regime should be designed in detail. The following design questions arise from the paper and its context:

(i) The takeover-time gap: how to delineate a reasonable takeover time for Level 3 and Level 4 systems, grounded in actual circumstances and in the empirical evidence on takeover performance, so that courts can assess whether a driver met their supervisory and resumption duties - legislation requires prompt resumption of control but has not defined what constitutes a reasonable and timely takeover duration.

(ii) Calibrated disclosure obligations and data-preservation rules for autonomous-driving data (what must be logged, for how long, and on whose demand), so that the data needed to prove defect, fault, and causation actually reaches the court.

(iii) The allocation itself: which party bears the burden of proof (in both its senses - production and risk of non-persuasion) for product defect, fault, and the causal link between defect and damage, including any presumptions or preliminary-proof thresholds that trigger a shift, together with safeguards against the adverse systemic consequences of broad burden shifting (higher litigation costs for manufacturers, defensive over-collection of data, chilling of innovation, and meritless claims) while preserving victims' access to compensation.

(iv) An updating mechanism that keeps the regime aligned with rapidly evolving AI algorithms and deployment practices, so that the liability rules do not become obsolete or unpredictable as the technology changes.

(v) Rules allocating evidentiary burdens in multi-cause cases, where multiple components, suppliers, and external events (for example cyberattacks or unforeseeable environmental events) may each have contributed to the harm.

These design questions are interdependent: the allocation depends on the evidence that disclosure and preservation duties make available to the court, and the extent of burden shifting a regime can adopt is conditioned by the safeguards against its adverse systemic consequences. The design is therefore a single integrated regime that addresses all five questions (i) to (v) together.

References

  1. Chen, Zhihua; Cai, Qianyi; Wei, Hanbing (2024). Distribution of the Burden of Proof in Autonomous Driving Tort Cases: Implications of the German Legislation for China. World Electric Vehicle Journal 15(7):305. DOI 10.3390/wevj15070305. https://doi.org/10.3390/wevj15070305
  2. 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. DOI 10.1016/j.rtbm.2026.101804. https://doi.org/10.1016/j.rtbm.2026.101804
  3. Li, Zhendong; Shi, Lefeng; He, Weijun (2025). A adaptability mechanisms study of traffic liability-division in autonomous vehicle era. Transport Policy. DOI 10.1016/j.tranpol.2025.06.023. https://doi.org/10.1016/j.tranpol.2025.06.023
  4. Lujan Tutusaus, Carlos; Hidalgo, Justin; Flix, Oriol (2026). Regulatory Divergence in Autonomous Vehicle Deployment: A Comparative Study of the US, EU, and China. SAE International Journal of Advances and Current Practices in Mobility. DOI 10.4271/2026-26-0243. https://doi.org/10.4271/2026-26-0243
  5. Ba, Zhuo; Zhao, Ziyu; Zhang, Bokang (2025). Liability for Autonomous Vehicle Torts: Who Should Be Held Responsible? World Electric Vehicle Journal 16(12):665. DOI 10.3390/wevj16120665. https://doi.org/10.3390/wevj16120665
  6. Zhang, Ruixuan (2025). Autonomous Vehicles: Legal Governance of Civil Liability Risks. Lecture Notes in Education Psychology and Public Media 96:108-114. DOI 10.54254/2753-7048/2025.BO24130. https://doi.org/10.54254/2753-7048/2025.bo24130
  7. Jacquemin, Zoe (2024). Product Liability Directive: Disclosure of Evidence, the Burden of Proof and Presumptions. Journal of European Tort Law. DOI 10.1515/jetl-2024-0009. https://doi.org/10.1515/jetl-2024-0009
  8. Nannini, Luca (2025). When Less Regulation Means More Complexity: The EU AI Liability Directive Withdrawal and Its Impact on European Technological Competitiveness. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society. DOI 10.1609/aies.v8i2.36679. https://doi.org/10.1609/aies.v8i2.36679
  9. European Union (2024). Directive (EU) 2024/2853 of the European Parliament and of the Council of 23 October 2024 on liability for defective products and repealing Council Directive 85/374/EEC. Official Journal of the European Union, CELEX 32024L2853. https://eur-lex.europa.eu/eli/dir/2024/2853/oj
  10. Xu, Yao; Wei, Jixin; Mi, Ting; Chen, Zhihua (2024). Data Security in Autonomous Driving: Multifaceted Challenges of Technology, Law, and Social Ethics. World Electric Vehicle Journal 16(1):6. DOI 10.3390/wevj16010006. https://doi.org/10.3390/wevj16010006
  11. Ebers, Martin (2025). Civil Liability for Autonomous Vehicles in Germany. In: Autonomous Vehicles and the Law. Brill Nijhoff. DOI 10.1163/9789004711891_007. https://doi.org/10.1163/9789004711891_007
  12. Ding, Chunyan (2025). Self-Driving Cars Facing the Law: the Case of China. In: Autonomous Vehicles and the Law. Brill Nijhoff, pp. 432-459. DOI 10.1163/9789004711891_013. https://doi.org/10.1163/9789004711891_013