INTRODUCTION
Autonomous Vehicles (Hereinafter “AVs”) have transformed in a new era of mobility by removing the need for human intervention, judgement and control, substituting them with artificial algorithms and complex control systems. However, while this signifies technological advancement, it also poses serious questions about traditional legal principles of liability, more specifically, on human based cause-effect relationship, and human intent actions required in the fault based system.1
The concerns surrounding the widespread use of AVs are neither empirical nor evidentiary, but are essentially conceptual. The fundamental aspect that is challenged by AVs, is the principle of causation which is at the essential of the law of torts. While traditionally, this has depended on human intent and actions, with the advent of AVs, where no singular cause is responsible for any given event due to the non-linear and complex interplay between machine learning, algorithms, real time data , so it is difficult to fix a direct link between the cause of action and human fault and causation.2
These disparities become more intensified when looked through the lens of Indian law. For instance, the Motor Vehicles Act (Hereinafter “MVA”), 1988, Consumer Protection Act (Hereinafter “CPA”), 2019 and Bharatiya Nyaya Sanhita, 2023 are all predicated on a human based approach, requiring the imputation of negligence on a human actor that can be said to have caused the said harm, with AVs making all such legislations futile as they rely on machine decisions, leading to fragmentation of responsibility among various interconnected systems.
FOUNDATIONS OF FAULT-BASED LIABILITY AND THE CENTRALITY OF CAUSATION
In modern tort law, fault-based liability, more specifically negligence, remains its dominant framework. Negligence basically has three components-duty of care, breach of duty and causation-the middle one-causation-plays the role of linking fault to liability. Without cause, no negligence. A claim in negligence fails not only on the ground that duty was owed and breached, but also that the harm was caused by the defendant.3
Causation in negligence has two aspects: factual and legal. Factual causation is normally assessed by the “but-for” test-would the harm have happened if it wasn’t for the defendant’s conduct.4 The importance of the test can be seen from Barnett v. Chelsea & Kensington Hospital Management Committee was held negligent but not liable on the ground that the injury would have been caused regardless of the defendant’s actions5 The factual causation is then narrowed down by the doctrine of proximate cause, which limits the liability only to such of the consequences that would reasonably have been foreseen, thus being Human Centric.
These exceptions still rely on the concept that the damage is attributed to certain identifiable human actions. This principle fails when applied to AVs since in the conventional driving scene decisions are concentrated in one actor-the human driver, thereby rendering it easy to trace the causes. An autonomous vehicle on the other hand is composed of components like a computer, a machine learning algorithm, and sensors etc; these elements together interact to generate an outcome, which means the cause is derived from the interaction of the components, not any specific component, hence a distributed causality.6 Thus, the question isn’t only that causation would be more difficult to ascertain, but the very structure of the cause does not fit the concept of autonomous systems, unlike the traditional law is human centric. This either results in an inability to prove causes and liability thus becoming indeterminate, or that the assignment of liability becomes arbitrary without relation to the moral concept of cause in the realm of tort law.
ALGORITHMIC DECISION-MAKING AND THE DISRUPTION OF LEGAL RESPONSIBILITY
AVs change the way in which driving decisions are made. In the way driving was done before, all decisions during the driving task were made by the human being controlling the vehicle. AVs use sensors, software and machine learning to make decisions on behalf of the human beings and driving is no longer done by one single person, but by a technical system.7
One prominent characteristic of autonomous driving decisions is their distributed nature. AVs involve a multitude of players and actions and there are various actors that participate in the decision-making process: manufacturers, software engineers, data providers etc. Autonomous driving decisions are made at several points in time of the system life cycle. As there are several actors making decisions at different times, it becomes quite difficult to establish a single person, whose decision may be considered responsible for a consequence derived from a certain decision.8
Another characteristic of the problem is the nature of machine learning. With traditional computer programs, decisions are made according to pre-determined rules. Machine learning systems decide on patterns learned from given data. As machine learning makes decisions by learning from data and experiences, the system might sometimes lead to unexpected or incomprehensible outcomes; people who built the system may not fully be able to explain why such a decision is being made by the system.9
A decision made in the event of a certain accident might have been determined by the system design, a learning process or the implementation of a particular version, long before the event takes place. This might create an insurmountable difficulty in tracing the liability of the specific act that causes the harm at the given time.The above challenges are all so different from the presumption on which the current tort law is based that it appears difficult to attribute fault liability in a conventional sense.10
THE COLLAPSE OF CAUSATION AND THE EVIDENTIARY CRISIS IN AUTONOMOUS VEHICLE LIABILITY
Causation has been its foundational element since the concept of fault based liability arose, it is where the defendant’s actions were the cause of the plaintiff’s harm. In light of autonomous vehicle technologies, not only is the proof of causation becoming problematic, the very idea of it is under attack. The concept of causation issues can be clearly distinguished by way of the SAE levels of automation: from 0-2 when a human driver is operating the vehicle standard negligence can be said to apply and the action or inaction can be traced to a human driver. But on who will the fault be placed when there are numerous, intertwined and interdependent elements of an autonomous vehicle, it seems as if product liability wouldn’t be the appropriate route as it doesn’t encompass the fact that all elements have a role to play, or shared control from level 3 onwards and complete autonomy at levels 4 and 5 will mandate a new conception of liability11. While there exists a human being and a direct chain of events, and as the theories behind causation stand the argument can always be made that the act or omission by the defendant is what caused the harm to the plaintiff, that ceases to be an applicable avenue once complex, adaptive, automated processes begin to function rather than human judgment.12
Doctrinally causation issues arise due to the deficiencies of the ‘but for’ test. This states that had it not been for the defendant’s negligent act the harm would not have occurred , this is identified in the case law Barnett v. Chelsea & Kensington Hospital Management Committee.13 In a system as complex as AVs, all elements of the system including the sensor input, programming of software, training data as well as the environment come together to make one functioning entity and if it is impossible to pinpoint one specific cause it is equally so for determining the impact each part has in the causal chain. The argument, however, is still valid in a broader context in Indian law where although “causal nexus” is recognized between negligence and harm it is predicated on the idea that the link between the two can be clearly identified in case Jacob Mathew v. State of Punjab14 and Kusum Sharma v. Batra Hospital15. But with respect to automated systems where the flaw lies this is even less credible due to the opaque and multifaceted workings of these systems. Proximate cause is also inadequate because it only covers foreseeable damages, but autonomous vehicle harm may arise from the system interacting with the environment, which leads to an issue of whose foresight will determine liability – programmer, manufacturer, user, or the system.16
The issues are with evidential challenges, with the plaintiff required to prove causation on the balance of probabilities. While Sec 101-102 of The BSA, 2023 establishes this burden, Sec 106 allows for an exception if the information required for the plaintiff to establish their case is only in the possession of the defendant. Since most of the vital information that is necessary to ascertain the system’s failure will only be in the possession of the manufacturer/ designer/ programmer, the plaintiff will not be able to succeed on this premise. This also increases issues with relation to multiple parties involved as the court may assign the responsibility to a party of its choice, but this wouldn’t lead to establishment of a definite cause for the harm. The issue isn’t only of having too many parties, but their respective contributions being indeterminable.17
Indian laws are in no way sufficient to address the matter, the MVA, 1988 deals with only drivers and insurance, Sec 166-168 and Sec 140 refer to drivers.18 The CPA, 2019 focuses on compensation from defects or services. But the defendant has to still establish the causal link in product liability sections 82-87 as well as Sec 2(10) and 2(11). Moreover the complexity of a system, as interconnected and as vast as an autonomous vehicle makes it impossible to identify a defect. Product liability itself may not be applicable as its focus is on the defendant.19
It can be clearly said that because of doctrinal, evidential and legal shortcomings causation’s viability in law is severely compromised, where due to inability to prove a causal link victims are left uncompensated and offenders are left without liability, which defeats the purpose of liability,and it isn’t feasible for all responsible parties to be held liable under a comprehensive tort liability scheme; what needs to be ascertained is not multiple tortfeasors but their distinct causal contribution.20 Causation needs to be redefined from being a mandatory element for AVs to being replaced by a concept of enterprise liability.
COMPARATIVE JUDICIAL APPROACHES AND LESSONS FOR INDIA
It is seen from the approaches taken in different countries that the concept of liability for AVs is also transforming. While in the existing legal regime, the human drivers are assigned liability, due to increasing levels of automation, the courts and legislation in US, UK and Germany are increasingly adapting to system-oriented liability regimes, in which the manufacturer, operator, or insurers bear liability instead of the individual user.21
In the UK, the Automated Vehicles Act, 2024 itself provides the concept of ‘user in charge’ (UiC) for the AVs. According to this act, when the vehicle is in authorised self-driving mode and being operated in it, human user in control of the car is exempt from the liability of the driving operation and the liability is attributed to the insurer that Compensates the victim and in turn seeks indemnification from Authorised Self-Driving Entity (ASDE), usually the vehicle manufacturer or designer. The UK approach, thereby, is moving away from assignment of liability from drivers, moving toward the concept of assigning liability on the basis of control.22
Germany’s approach appears to adopt a two- tiered approach wherein strict liability is imposed on the keeper of the vehicle according to the Road Traffic Act (Straenverkehrsgesetz) which is further supplemented by manufacturer liability for system failures. The introduction of a “Technical Supervisor” under the more advanced level of automation, also highlights the adaptation of the legal system for the new technological environment. One specific aspect of the German approach is the combination of legal rules and insurance liability with evidence gathering technology like event-data recorders that addresses the issues of compensation and the establishment of negligence.23
In the US, the evolving notions of liability for AVs are primarily driven by court decisions and state tort laws. Whereas the product liability principles and rules apply for the higher level of automation with blame assigned to the systems design rather than human negligence, the lower level of automation sees the traditional principles of driver liability being applied.24
In all these countries we can witness a systematic shift from the fault-based liability to system-oriented liability that connects responsibility with the control of the system, risk creation and access to the information in concern. These advancements have some implications for Indian law as in India the liability of AVs is still rooted in human drivers as provided under the MVA, 1988. So the lessons that could be derived from the global experiences for Indian courts are: Firstly, need to establish an insurance based compensated scheme similar to UK; Secondly, adoption of rules of product liability and enterprise liability like US and Germany, and Thirdly, making event data recorders a part of technology, so that evidence is obtained to establish the control of the vehicle. India needs to rapidly recognize the transition from partial automation to full automation (Level 3 to Level 4) and establish system-oriented liability rather than fault-based liability on the drivers, else the present MVA framework shall appear outdated.
RETHINKING LIABILITY: BEYOND FAULT-BASED MODELS IN INDIAN LAW
The problem is not merely that there is insufficient evidence to establish that cause of action can be proven in case of AVs but that the law based on fault principle is inherently incompatible with such a technology. Once the act of a human driver is replaced by an operational logic of a system, and a decision is an intricate integrated decision, identifying the fault and cause of action in the human or an event cannot continue functioning. 25 Furthermore, broadening multiple causes doctrines is unlikely to be an appropriate solution. Traditionally multiple cause doctrines were developed and implemented to allocate fault and compensate victims for losses caused by a variety of factors. However, even in situations where there are multiple causes, traditional law assumes that these causes are atomized and allocated among the parties involved. With autonomous systems causality is not multiple but entirely unknown.
This can be an enormous difficulty to the Indian legal system where law of causation has been treated as a prerequisite for fault liability. There is a risk of not getting compensation by an injured person simply because it may not be possible to identify the harm caused by a complex opaque autonomous system. The Indian system has already complicated things with Sections 101 and 102 of the BSA, 2023 which places the burden of proof on the claimant and proving information from the manufacturers or designers would be a distant dream for a person injured by a self-driving vehicle.
Experiments conducted in other jurisdictions can be used as guidance. The Law Commission in England proposes the idea of ‘Authorised Self-Driving Entity’ (ASDE) which will replace the current driver-based fault liability framework. In America also, the movement is towards product liability as well as strict liability. Germany has amended the Road Traffic Act (Straenverkehrsgesetz) to keep up with high levels of automation which place the obligation on manufacturers and also include mandatory insurance provision. All such suggestions indicate the trend towards moving away from fault liability.26
One of the likely paths in developing the law of causation would be to introduce a presumption of liability. Instead of asking a person to prove how the harm has been caused, we could have a presumption of liability for a person or entities such as manufacturers or operators, and it will be up to them to show otherwise. Section 106 of the BSA also provides for cases of knowledge of a fact especially known to a person. Such principles can be extended in cases of AVs to compensate for the information deficit.
Another solution applicable to Indian circumstances would be an enterprise liability approach. Enterprise liability, in contrast to individual fault based liability regime, does not focus on personal culpability but the entity that creates and derives benefit from the risk. The entity creating an autonomous system is a complicated nexus of manufacturers, developers, service providers, and data disseminators and therefore, it would be appropriate to hold all of them responsible for harm. This could evolve in the Indian framework by extending the present framework of MVA, 1988. Although Sections 166 and 168 are fault based claims, Section 140 is a limited no fault provision and similar frameworks could be used to deal with the harm generated by AVs.27
A similar challenge is also observed in the context of the CPA, 2019. Though there is provision for bringing a claim for defects in the definitions in Section 2(10) and 2(11) and Sections 82-87 for product liability.28 However, in cases of an autonomous vehicle, it is hard to pin down a specific defect because harm may result from a combination of system operations and therefore, traditional product liability will not be sufficient to address such a complex claim.
Insurance mechanisms would be a vital alternative to victims who may otherwise not get compensation at all. The insurance, instead of a consequence of liability, could form the bedrock of a legal system. Such a model can be compared to the Section 140 of MVA, 1988 although extended more broadly for autonomous systems. The Motor Vehicle Act itself embodies an element of no fault liability in Section 140 and has been used successfully for compensations when fault could not be proved, as in the case of other technology based risks such as autonomous systems internationally.29
This does not mean that we should discard the concept of fault liability completely. We can still have a hybrid system which would apply traditional fault liability where a machine has been used and controlled by a human or misused, while the systems based liability would be applicable for the cases where the vehicle has been controlled by the autonomous system.
LEVEL 5 AUTOMATION AND THE TRANSFORMATION OF LEGAL RESPONSIBILITY
In the higher levels of automation, specifically level 5 (total automation), the whole notion of legal liability itself has a tectonic shift in paradigms. While in the earlier levels the human driver has some degree of supervision or control over the system, in case of level 5, there is no driver at all: the machine stops being an instrument at the hands of a human, and becomes a decision making agent on its own; it cannot then carry the human occupant in the vehicle and is no longer conceived as a sort of “driving agent” for whom a human driver has a duty to monitor.30
There would be profound effects in terms of apportionment of liability. Standard conceptions of negligence as an inchoate human behavior would not apply to the automated system, because of absence of a human decision maker. Liability will begin to be levied on the ‘Automated Driving System Entity’ that is the manufacturer, then technician, and may be both alone or together the user of the automated system, and this liability will be on the basis of provided system parameters, its design and the risks it incurs.31
A practical outcome of this advancement is that we can no longer refer to “human error” as a defence which will enable drivers to defend their actions for example the failure to maintain adequate control of the vehicle and prevent an accident arising there from. In ordinary circumstances, the manufacturer will be able to point to the driver as having caused the accident that is for example if he was too slow in reaction or inattentive to road signs). At level 5, this defense cannot be made use of, since a human driver cannot even interact with the automated system and the human occupant has no input in driving decisions.
The advances in automation have therefore created a basis for implementation of both enterprise-based liability and strict liability regimes, especially in the U.S., wherein manufacturer and user of automation may be held liable irrespective of fault. The rationale is that since they profit from designing, selling, or utilizing automation technologies, they should bear the burden of risk, and internalize the cost by being held liable. Following similar logic, regulatory actions with regard to road traffic offenses would, more realistically, be viewed as regulatory infringements rather than criminal offenses committed by the passenger, and liability would fall on the system contractor or fleet operator.This raises the complexity level to theoretical arguments, with the idea of granting “electronic personhood” to autonomous systems, though this idea is still subject to debate and exists only in theory.32
However, the Indian legal system is also on the verge of witnessing an age where human beings are no longer behind the wheel of automobiles, and cannot be blamed when accidents happen. Existing Indian laws, such as the MVA of 1988 and its associated laws, are built upon the premise of the human driver operating the vehicle. As the world shifts towards fully autonomous driving systems, the Indian legal system will need to create new rules and precedents, and depart from existing liability models that center around the human driver and focus more on compensation, moving towards strict liability modes for system and enterprise execution functions, and insurance mechanisms instead of fault-based doctrines that have little theoretical relevance today.
CONCLUSION
This paper has argued that the rise of AVs fundamentally challenges the continued reliance on fault-based liability within Indian law. The traditional principles of negligence are unsuited to complex, technologically mediated, and inscrutable systems.
The crux of the problem lies in the issue of causation. A machine’s causation is a socio-technical process, unlike that of an individual human actor where causality could be broken down and viewed separately. Even the doctrines of concurrent causation are not capable of handling these issues as they presuppose the separation and independent appraisal of various causal agents.
The limitations are compounded by evidentiary problems, under the BSA, 2023, whereby the burden is on the party without indispensable technical information. Also the problem lies in laws such as the MVA, 1988 and CPA, 2019; both rely heavily on human fault and defect, thereby not having a viable tool for addressing harms caused by machine logic. A growing disconnect between law and technology, thus compromises fundamental principles of accountability and access to justice.
Therefore, it is submitted that there exists a pressing need for a system of liability, based on the inherent systemic character of automated technologies. Liability presumptions, enterprise liability and insurance based compensation are some readily available alternatives that could move away from tracing causality and effect in detailed steps. This evolution of legal thought, away from an individually-focused regime and towards systemic liability, is visible in the legal frameworks of the UK, USA and Germany.
This does not imply a need to entirely do away with fault based liability. A hybrid fault based system, which could potentially hold the human agent responsible in cases of human error and use a system oriented approach in cases of algorithmic error, could prove effective.
Ultimately, the transition from human negligence to machine logic necessitates a corresponding transformation in legal thought. AVs not only challenge the traditional legal system, but also pose a threat to its entire framework. To remain blind to this reality is to compromise the ability of law to dispense justice.
*****
Footnotes
1. Bo Long, Ziyu Zhao & Qianyi Cai, Comparing Tort Liability Frameworks in Autonomous Vehicle Accident Governance, 17 World Elec. Vehicle J. 32 (2026).
2. Zhuo Ba, Ziyu Zhao & Bokang Zhang, Liability for Autonomous Vehicle Torts: Who Should Be Held Responsible?, 16 World Elec. Vehicle J. 665 (2025).
3. Mariolaine Monot-Fouletier, Liability for Autonomous Vehicle Accidents, in The Cambridge Handbook of Artificial Intelligence: Global Perspectives on Law and Ethics 163 (2022).
4. Amit Chougule, Vinay Chamola, Aishwarya Sam, Fei Richard Yu & Biplab Sikdar, A Comprehensive Review on Limitations of Autonomous Driving and Its Impact on Accidents and Collisions, 5 IEEE Open J. Vehicular Tech. 142 (2023).
5. Barnett v. Chelsea & Kensington Hospital Management Committee, [1969] 1 Q.B. 428 (Q.B.).
6. Vahid Yazdanpanah et al., Reasoning About Responsibility in Autonomous Systems: Challenges and Opportunities, 38 AI & Soc’y 1453 (2023).
7. Maarten Herbosch, Liability for AI Agents, 26 N.C.J.L. & Tech. 391 (2024).
8. Darsh Parekh et al., A Review on Autonomous Vehicles: Progress, Methods and Challenges, 11 Electronics 2162 (2022).
9. Gourav Bathla et al., Autonomous Vehicles and Intelligent Automation: Applications, Challenges, and Opportunities, 2022 Mobile Info. Sys. 7632892 (2022).
10. Kareem Othman, Exploring the Implications of Autonomous Vehicles: A Comprehensive Review, 7 Innovative Infrastructure Sols. 165 (2022).
11. Prasenjit Ghorai et al., A Causation Analysis of Autonomous Vehicle Crashes, 16 IEEE Intell. Transp. Sys. Mag. 33 (2024).
12. Ibid
13. Barnett v. Chelsea & Kensington Hosp. Mgmt. Comm., [1969] 1 Q.B. 428.
14. Jacob Mathew v. State of Punjab, (2005) 6 S.C.C. 1 (India).
15. Kusum Sharma v. Batra Hospital & Med. Research Centre, (2010) 3 S.C.C. 480 (India)
16. Qidi Zhang, Tingru Zhang & Liang Ma, Human Acceptance of Autonomous Vehicles: Research Status and Prospects, 95 Int’l J. Indus. Ergonomics 103458 (2023).
17. Bharatiya Sakshya Adhiniyam, 2023.(India)
18. Motor Vehicles Act, 1988,(India)
19. Consumer Protection Act, 2019(India)
20. Cassandra Burke Robertson, Litigating Partial Autonomy, 109 Iowa L. Rev. 1655 (2023).
21. Antonio Artruñedo, Marcos Moreno-Gonzalez & Jorge Villagra, Lateral Control for Autonomous Vehicles: A Comparative Evaluation, 57 Ann. Rev. Control 100910 (2024).
22. Law Comm’n, Automated Vehicles (Law Com No. 404, 2022), https://www.lawcom.gov.uk.
23. Straßenverkehrsgesetz [StVG] [Road Traffic Act], as amended (Ger.).
24. Md Masud Rana & Kamal Hossain, Connected and Autonomous Vehicles and Infrastructures: A Literature Review, 16 Int’l J. Pavement Res. & Tech. 264 (2023).
25. Conor Rowan & Alireza Doostan, On the Definition and Importance of Interpretability in Scientific Machine Learning, arXiv (2025),
26. Brittany Eastman et al., A Comparative Look at Various Countries’ Legal Regimes Governing Automated Vehicles, J.L. & Mobility 1 (2023).
27. Motor Vehicle Act,1988(India)
28. Consumer Protection Act,2019(India)
29. Motor Vehicle,1988(India)
30. Siming Zhai, Lin Wang & Peng Liu, Not in Control, but Liable? Attributing Human Responsibility for Fully Automated Vehicle Accidents, 33 Engineering 121 (2024).
31. Manzoor Ahmed Khan et al., Level-5 Autonomous Driving—Are We There Yet? A Review of Research Literature, 55 ACM Computing Surv. 1 (2022).
32. Jingyuan Zhao et al., Autonomous Driving System: A Comprehensive Survey, 242 Expert Sys. with Applications 122836 (2024).