From Human Fault to Algorithmic Accountability: Tort Law in the AI Era

Artificial Intelligence (AI) is transforming the way decisions are made across critical sectors worldwide. This shift from automation to autonomy presents significant challenges for traditional tort law, which has historically been built around concepts such as human fault, foreseeability and direct causation.

Introduction

Unlike traditional software systems that operate according to predefined instructions, modern AI systems increasingly rely on machine learning and adaptive decision-making processes. These systems allow them to function with varying degrees of autonomy across sectors such as:

• Healthcare
• Transportation
• Finance
• Public administration

Traditional tort principles evolved in an era where harmful conduct could generally be traced to an identifiable individual or entity. Today, however, many AI systems operate through complex algorithms whose decision-making processes may be difficult to understand even for their developers.

The “black box” nature of artificial intelligence complicates efforts to determine fault, establish causation and assign legal responsibility. At the same time, the fragmented AI ecosystem — comprising software developers, data providers, hardware manufacturers and system deployers — further complicates liability assessments.

As AI systems continue to exercise greater decision-making authority, legal systems worldwide are increasingly examining whether traditional tort law remains adequate to address AI-related harm.¹

The Nature of AI and the Shift from Automation to Autonomy

Understanding AI liability requires distinguishing between automated and autonomous systems.

Automated vs. Autonomous Systems

• Automated systems operate according to predefined rules established by human programmers. When harm occurs, liability can generally be traced to programming errors, design defects or human oversight failures.
• Autonomous systems rely on machine learning and environmental inputs to generate decisions and adapt their behaviour over time. Their outputs may not always be directly predictable because they are based on probabilistic models rather than fixed instructions.

This distinction has significant implications for legal liability because traditional tort law assumes a degree of predictability and human control that may not exist in autonomous systems.

The Autonomous Vehicle Example

The development of autonomous vehicles illustrates this challenge. The Society of Automotive Engineers (SAE) classifies vehicle automation on a scale from Level 0 to Level 5. While Levels 0 to 2 require meaningful human supervision, Levels 3 to 5 increasingly transfer decision-making authority to the vehicle itself.

Particularly difficult legal questions arise at the transition between Levels 2 and 3, where human operators remain legally responsible for intervention despite being largely disengaged from active vehicle control. This phenomenon has been described as the “human-machine interaction paradox,” where liability continues to rest on human actors even though critical operational decisions are made by autonomous systems.

Tort Law and the Crisis of Traditional Liability Principles

Traditional tort law allocates liability through doctrines such as:

• Negligence
• Strict liability
• Product liability

These doctrines depend heavily upon concepts of human conduct, reasonable care and foreseeability. Artificial intelligence challenges these assumptions because harmful outcomes may result from algorithmic processes rather than direct human instructions.

The Negligence Framework Under Strain

The negligence framework is particularly strained because the traditional “reasonable person” standard was developed to evaluate human behaviour. AI systems, however, process information differently and often outperform humans in specific tasks.

Some scholars have therefore proposed a “reasonable computer” standard, under which AI behaviour would be assessed against industry standards, accepted technological practices and comparable algorithmic systems rather than human conduct.

Foreseeability in an AI Context

Foreseeability also becomes more complex in the context of AI. Autonomous systems may achieve their intended objectives while simultaneously producing harmful unintended consequences.

For example, a healthcare AI designed to optimise patient triage may incorrectly classify patients, resulting in delayed treatment or medical harm. While the precise outcome may not have been foreseeable, the broader risks associated with deploying autonomous systems may nevertheless be foreseeable to developers and operators.

Causation and the But-For Test

Causation presents an equally significant challenge. Tort law traditionally relies upon the “but-for” test, requiring claimants to demonstrate that the harm would not have occurred but for the defendant’s conduct.

However, AI systems often operate through vast datasets, self-generated correlations and adaptive learning processes that obscure the causal relationship between human design decisions and resulting harm. As a result, victims may be able to demonstrate injury without being able to identify precisely which actor within the AI ecosystem caused the harm.

The Black Box Problem and Evidentiary Challenges

One of the most significant obstacles in AI liability litigation is the opacity of algorithmic decision-making. Many advanced AI systems, particularly those based on deep learning, do not generate outcomes through transparent rule-based processes.

Instead, decisions emerge through multiple layers of internal computational processes that may be difficult or impossible for users, regulators or courts to fully interpret. This creates a substantial information asymmetry between AI developers and those affected by AI-generated decisions.

Unlike traditional defective products, which can often be physically inspected and tested, AI-related claims may require access to:

• Proprietary source code
• Training datasets
• System logs
• Technical documentation

Such information is frequently protected as confidential business information or trade secrets. Even where access is available, the non-deterministic nature of many AI systems may make it difficult to reproduce a particular outcome.

Consequently, plaintiffs may struggle to satisfy evidentiary burdens relating to negligence, causation and defectiveness. These challenges have prompted increasing calls for greater transparency, explainability and documentation requirements for high-risk AI systems.

The Fragmented AI Supply Chain

Liability becomes further complicated by the fragmented nature of AI development and deployment. Responsibility may be distributed among multiple actors, including:

• Data suppliers
• Software developers
• Model trainers
• Hardware manufacturers
• Deploying entities such as hospitals, banks or transportation companies

Traditional tort law generally seeks to identify a proximate cause and a responsible defendant. AI systems, however, function through interconnected technological contributions that blur traditional distinctions between creators, operators and users.

This has led scholars and policymakers to explore alternative approaches such as joint liability, enterprise liability and risk-based allocation frameworks. Under such models, liability may be imposed on the entity best positioned to prevent harm, manage risks or compensate victims.

The 2018 Uber Autonomous Vehicle Accident

The 2018 Uber autonomous vehicle accident illustrates these difficulties. Although the vehicle’s systems detected the pedestrian before impact, technical design choices and disabled safety features contributed to the collision.

Yet much of the legal scrutiny focused on the human safety driver rather than the broader technological and organisational factors that contributed to the incident. The case highlighted the continuing challenges associated with assigning responsibility for AI-enabled harm.

Comparative Approaches to AI Liability

Different jurisdictions have adopted varying approaches to regulating AI-related risks and liability.

European Union

The European Union has emerged as a global leader in AI regulation through measures such as the EU AI Act and reforms to product liability legislation. These frameworks adopt a risk-based approach and recognise that software and AI-enabled products may generate liability even where traditional product concepts are difficult to apply.

The reforms seek to ensure that victims are not deprived of remedies simply because harm was caused by autonomous or self-learning systems.

United Kingdom

The United Kingdom has adopted a more flexible, principles-based approach. Rather than creating a single AI regulator, the UK relies on existing regulators to apply overarching principles within their respective sectors, including:

• Safety
• Transparency
• Accountability
• Fairness

United States

In the United States, AI regulation continues to develop primarily through litigation and sector-specific regulation. Courts have increasingly examined the extent to which manufacturers may be liable when users place excessive reliance on semi-autonomous systems.

India

India currently lacks a dedicated legal framework governing AI liability. Existing laws address certain aspects of technology regulation but do not comprehensively address liability arising from autonomous decision-making systems. These include:

• Consumer Protection Act, 2019
• Information Technology Act, 2000
• Digital Personal Data Protection Act, 2023

However, India’s jurisprudence on strict and absolute liability may offer useful insights. The doctrine of absolute liability established in M.C. Mehta v. Union of India demonstrates the willingness of Indian courts to impose liability on enterprises engaged in inherently hazardous activities.

While this doctrine was developed in an environmental context and does not presently apply to AI systems, some scholars have suggested that risk-based liability frameworks may provide a useful model for regulating high-risk AI applications.

Conclusion

The rapid advancement of artificial intelligence has exposed the limitations of traditional tort law principles developed for a world in which human actors exercised direct control over decision-making. Concepts such as negligence, foreseeability and causation remain foundational to civil liability, yet their application becomes increasingly complex when autonomous systems operate through adaptive and often opaque algorithms.

The “black box” nature of AI, the fragmented technological supply chain and the growing autonomy of machine-learning systems have created significant accountability challenges. Victims of AI-related harm may struggle to identify the responsible party, establish causation or obtain access to the technical information necessary to prove their claims.

Comparative approaches adopted by jurisdictions such as the European Union, the United Kingdom and the United States demonstrate a growing recognition that existing legal frameworks must evolve to address the risks posed by artificial intelligence. While India does not yet have a dedicated AI liability regime, existing principles of tort law, consumer protection and regulatory oversight may provide a foundation for future reforms.

As artificial intelligence becomes increasingly integrated into critical sectors such as healthcare, transportation, finance and public administration, legal systems will need to develop frameworks that balance innovation with accountability. The future of tort liability for artificial intelligence will likely depend on creating mechanisms that ensure effective compensation for victims while promoting responsible development and deployment of AI technologies.

Last Updated on 11 June, 2026