1. Introduction

Artificial Intelligence (AI) is increasingly used in autonomous vehicles, medical diagnosis, financial services, industrial robotics, predictive policing, consumer platforms, and automated decision-making. Although these technologies can improve efficiency and accuracy, they can also cause physical injury, financial loss, privacy violations, discrimination, property damage, and other legally recognizable harms.

AI-related tort liability concerns the circumstances in which a person or organization may be held civilly liable for harm caused by an AI system. It examines the duties owed by developers, manufacturers, deployers, operators, professionals, and other parties, together with the legal requirements for establishing breach, causation, damage, and available remedies.

The central difficulty is that an AI system may operate through complex algorithms, changing data, and interactions among multiple parties. It may be unclear whether harm resulted from defective software, inadequate training data, unsafe design, negligent deployment, insufficient human supervision, or an unforeseeable external event.

Traditional tort law remains relevant, but its application varies across jurisdictions. Civil-law countries often organize liability through codified provisions concerning fault, causation, defective products, and responsibility for things or activities. Common-law jurisdictions rely more heavily on judicial precedent, negligence, product liability, and statutory causes of action.

A comparative approach therefore asks three fundamental questions:

Who owed the injured person a legally recognized duty?

Can the AI-related conduct or defect be shown to have caused legally recoverable harm?

Which liability model provides a fair and effective remedy without imposing unjustified responsibility?

An important qualification is that relatively few reported judgments directly establish liability for harm caused by modern AI systems. The case laws discussed below include direct automated-system disputes and established precedents that provide relevant principles by analogy.

2. Meaning and scope of AI-related tort liability

Tort liability generally arises when a person's legally wrongful act or omission causes harm for which civil law provides a remedy.

AI-related tort disputes may arise in several contexts.

A. Autonomous vehicles

An autonomous vehicle may cause a collision because of a perception error, unsafe navigation decision, defective sensor, inadequate training, or failure to respond appropriately to road conditions.

Potentially responsible parties include the manufacturer, software developer, vehicle operator, maintenance provider, or another road user. Responsibility depends on the facts and governing law.

B. Medical AI

AI systems may assist clinicians in diagnosing diseases, interpreting medical images, recommending treatments, or identifying high-risk patients.

A harmful outcome may result from inaccurate predictions, unsuitable training data, inadequate validation, defective software, or unreasonable reliance on an AI recommendation.

The legal question is not simply whether the AI made an error. It is whether a healthcare professional, manufacturer, hospital, or other party breached an applicable duty and thereby caused compensable harm.

C. AI-based financial decisions

Automated lending, insurance underwriting, fraud detection, and investment systems may cause financial loss through inaccurate classifications, defective models, discriminatory processing, or inadequate safeguards.

Claims may arise under negligence, contract, consumer-protection legislation, privacy law, or other applicable rules. Not every incorrect automated decision constitutes a tort.

D. Privacy and data-related harm

Facial recognition, biometric identification, automated profiling, and large-scale data analysis may cause unlawful interference with privacy or other protected interests.

Liability depends on whether the conduct violates an applicable legal duty or statutory provision and whether the claimant has access to the relevant remedy.

E. Industrial robotics and workplace automation

AI-enabled machinery may injure workers or damage property because of defective design, inadequate maintenance, insufficient safety procedures, or failures of supervision.

These cases may involve product liability, employer responsibility, occupational-safety legislation, and ordinary negligence principles.

F. Generative AI and automated content

AI-generated material may produce reputational injury, misinformation-related losses, unlawful disclosure, or intellectual-property infringement.

Whether a tort claim succeeds depends on the applicable legal rules concerning defamation, privacy, negligence, economic loss, or other protected interests. The mere production of inaccurate content does not automatically establish tort liability.

3. Principal models of AI-related tort liability

Model 1: Fault-based liability

Under a fault-based model, the claimant generally must establish that the defendant failed to exercise the care required by law and that this failure caused legally recoverable harm.

For example, a manufacturer may negligently release an AI-enabled device without appropriate safety testing, or a hospital may deploy a diagnostic model without adequate validation.

This model encourages responsible conduct but may create evidentiary difficulties when the relevant system is technically complex.

Model 2: Strict product liability

Strict product liability may impose responsibility for a defective product without requiring the claimant to prove that the manufacturer was negligent. The claimant must still establish the legally required elements, which commonly include a relevant defect, causation, and damage.

An AI-enabled product might be defective because of unsafe design, manufacturing defects, inadequate warnings, or software-related failures, depending on the applicable law.

Strict liability does not automatically apply to every AI service or every incorrect algorithmic output.

Model 3: Vicarious liability

Vicarious liability may make an employer or other legally responsible principal liable for torts committed by an employee or agent in circumstances covered by the relevant legal doctrine.

Where AI is used within a business, the organization may also have direct liability for its own negligent selection, deployment, supervision, or monitoring of the system.

The fact that an AI system itself is not a legal person does not prevent liability from attaching to the human or corporate actors responsible under existing law.

Model 4: Liability for dangerous things or activities

Some civil-law systems provide special rules concerning harm caused by things under a person's control, dangerous activities, or defective products.

These rules may be relevant when an AI-enabled machine, autonomous vehicle, or industrial robot causes injury. Their application depends on statutory wording, judicial interpretation, and the relationship between the defendant and the source of harm.

Model 5: Statutory liability and compensation schemes

Legislation may establish specific liability rules, presumptions, insurance requirements, or compensation mechanisms for particular technologies or activities.

Examples include motor-vehicle liability, workplace injuries, medical-device regulation, and consumer product safety. Such mechanisms can sometimes provide remedies without requiring the claimant to prove every element of ordinary negligence.

Model 6: Shared or proportionate responsibility

AI-related harm may result from multiple contributing failures: defective software, inadequate training data, negligent deployment, poor maintenance, and failure to respond to warnings.

Depending on the jurisdiction, the law may allocate responsibility through joint and several liability, contribution claims, comparative fault, or other apportionment rules.

This model is particularly important where the injured person cannot reasonably identify a single technical failure as the sole cause of harm.

4. Comparative legal frameworks

A. India

Indian AI-related tort liability draws on general civil-law principles, judicial precedent, constitutional law, consumer protection, product liability, and sector-specific legislation.

Important legal foundations include:

Negligence: The claimant must establish the applicable duty of care, breach, causation, and damage.

Consumer Protection Act, 2019: Its product-liability provisions may apply to qualifying claims involving defective products and services.

Constitutional remedies: Articles 14 and 21 may be relevant to unlawful state action affecting equality, privacy, or personal liberty.

Information Technology Act, 2000: Relevant provisions may apply to specified electronic and cyber-related conduct.

Data-protection legislation: Applicable personal-data rules may create obligations and remedies for unlawful processing.

Motor-vehicle and medical regulation: Special statutory rules may govern particular AI-enabled applications.

India does not have a single, comprehensive AI-specific tort-liability statute. Courts therefore need to assess the facts against existing doctrines and any applicable statutory provisions.

B. European Union

The EU combines general civil liability, data-protection law, product-safety rules, and AI-specific regulation.

The AI Act, Regulation (EU) 2024/1689, establishes obligations for specified AI systems according to their risk and intended use. It is not, by itself, a comprehensive compensation code for every AI-related injury.

The revised EU Product Liability Directive, Directive (EU) 2024/2853, expressly accommodates software, including certain AI systems, within the product-liability framework. It also contains provisions addressing disclosure of evidence and certain evidentiary difficulties in complex cases. Its application depends on the Directive's transposition and temporal rules.

Traditional non-contractual liability remains important. Whether a claimant must prove fault, whether presumptions apply, and whether a particular defendant is liable depend on the applicable EU and national rules.

C. United Kingdom

The UK framework includes negligence, product liability, data protection, consumer law, and statutory rules governing particular activities.

The Consumer Protection Act 1987 provides a strict-liability framework for qualifying damage caused by defective products. Negligence remains relevant where the claimant alleges a failure to exercise reasonable care.

The legal analysis must distinguish a product defect from negligent deployment or supervision. An AI provider, manufacturer, and business user may have different duties and defences.

D. United States

US AI-related tort claims are shaped by state tort law, product-liability doctrines, federal legislation, and sector-specific regulation.

Negligence, strict product liability, failure-to-warn claims, and vicarious liability may all be relevant. However, the elements and defences differ between states.

A further challenge is federal pre-emption: certain federal statutory or regulatory schemes may limit or displace particular state-law claims. The availability of a tort remedy therefore depends on the nature of the claim and the governing jurisdiction.

E. Comparative overview

5. Case laws: at least eight important authorities

The following cases establish principles relevant to AI-related tort liability. Most predate modern generative AI and autonomous systems. They should therefore be used as doctrinal foundations or analogies, not mischaracterized as direct rulings on modern AI liability.

Case 1. Donoghue v. Stevenson (1932)

Citation: [1932] AC 562, House of Lords, United Kingdom.

Facts: A consumer alleged illness after consuming ginger beer containing a decomposed snail. The bottle had been opaque, preventing inspection of its contents before consumption.

Legal principle: The House of Lords recognized the modern negligence principle that a manufacturer may owe a duty of care to the ultimate consumer, even without a direct contract between them.

Relevance to AI liability: An AI developer or manufacturer may owe duties to foreseeable users or other persons affected by a product, depending on the circumstances and applicable law. The absence of a direct contract does not necessarily eliminate a tort claim.

Civil-law significance: The case supplies a foundational approach to duty of care, foreseeability, and responsibility for harm caused by products. It does not establish that every AI developer owes a duty to every person affected by an algorithm.

Case 2. Caparo Industries plc v. Dickman (1990)

Citation: [1990] 2 AC 605, House of Lords, United Kingdom.

Facts: Caparo relied on audited company accounts when acquiring shares and sought damages arising from the alleged inadequacy of the audit.

Legal principle: The House of Lords addressed the circumstances in which a duty of care arises, considering foreseeability, proximity, and whether imposing a duty would be fair, just, and reasonable.

Relevance to AI liability: An organization may foresee that its AI output will influence important decisions, but foreseeability alone does not automatically establish a duty to every affected party. The relationship between the parties and the purpose for which the system was supplied can be decisive.

Civil-law significance: The case assists courts in identifying the scope of potential liability for AI-generated assessments, forecasts, and recommendations.

Case 3. MacPherson v. Buick Motor Co. (1916)

Citation: 217 N.Y. 382, 111 N.E. 1050, Court of Appeals of New York, United States.

Facts: A purchaser was injured when a wheel of a motor vehicle collapsed. The vehicle's manufacturer had not itself manufactured the defective wheel.

Legal principle: The court recognized a manufacturer's duty of care to foreseeable users where a product, if negligently made, could endanger life or limb.

Relevance to AI liability: A manufacturer may not escape potential responsibility merely because a harmful component was supplied by another company. In AI-enabled products, the relevant components might include sensors, software, training-data pipelines, or control systems.

Civil-law significance: The case is useful for assessing responsibility across supply chains, although modern claims also depend on product-liability statutes and the specific role of each supplier.

Case 4. Escola v. Coca-Cola Bottling Co. (1944)

Citation: 24 Cal. 2d 453, 150 P.2d 436, Supreme Court of California, United States.

Facts: A bottle of Coca-Cola broke and injured a restaurant employee. The case involved the inference of negligence from the circumstances of the accident.

Legal principle: The majority addressed the evidentiary doctrine of res ipsa loquitur. In a famous concurrence, Justice Roger Traynor advocated strict liability for defective products.

Relevance to AI liability: Where a system failure occurs under circumstances that ordinarily would not happen if reasonable precautions had been taken, the available evidence may support an inference of negligence under the applicable law. Justice Traynor's concurrence also influenced the development of strict product liability.

However, a complex AI failure does not automatically justify such an inference; the claimant must satisfy the governing evidentiary requirements.

Civil-law significance: The case highlights two approaches to product-related harm: fault-based responsibility and strict liability for qualifying defects.

Case 5. A v. National Blood Authority (2001)

Citation: [2001] 3 All ER 289, High Court of England and Wales.

Facts: The claimants suffered harm after receiving blood contaminated with hepatitis C. The litigation concerned liability under the Consumer Protection Act 1987.

Legal principle: The court considered whether the blood supplied was defective under the statutory product-liability framework and rejected the argument that the relevant products met the level of safety persons were entitled to expect.

Relevance to AI liability: This authority illustrates that a product can be legally defective even when the supplier did not act negligently in the conventional sense. An AI-enabled medical device or other qualifying product may similarly raise strict-liability questions if it fails to provide the safety persons are entitled to expect.

Civil-law significance: The case is useful for understanding the distinction between negligence and statutory product liability. It does not establish that an inaccurate AI prediction automatically makes a product defective.

Case 6. Boston Scientific Medizintechnik GmbH v. AOK Sachsen-Anhalt and B. v. AOK Sachsen-Anhalt (2015)

Citation: Joined Cases C-503/13 and C-504/13, Court of Justice of the European Union.

Facts: The disputes involved implanted medical devices in which potential defects could pose serious health risks, even though an actual defect had not necessarily been demonstrated in each individual device.

Legal principle: The Court held, in the context of the applicable medical-device product-liability framework, that a potential defect in a group of products presenting an unusually high risk could support classification as defective. Necessary replacement costs could qualify as damage in the circumstances considered.

Relevance to AI liability: AI-enabled medical devices may involve risks that emerge from model architecture, software updates, or a particular class of devices. The case helps explain why product-safety assessment may need to consider systemic risk rather than only a single isolated malfunction.

Civil-law significance: The decision illustrates the interaction between product safety, risk evidence, and strict liability in the EU. Its reasoning is tied to the applicable product-liability framework and should not be extended automatically to every software service.

Case 7. State v. Loomis (2016)

Citation: 881 N.W.2d 749, Supreme Court of Wisconsin, United States.

Facts: A defendant challenged the use of the COMPAS algorithmic risk-assessment tool during criminal sentencing.

Legal principle: The court permitted the tool's use in the circumstances before it, while recognizing limitations and the need for caution in interpreting its output.

Relevance to AI liability: The case illustrates how a decision-maker may over-rely on an opaque algorithmic score. In civil disputes, comparable issues may arise where an employer, lender, healthcare provider, or public authority relies on a model without understanding its limitations.

Civil-law significance: It supports careful consideration of human oversight and algorithmic reliability. It is not a tort-liability judgment and does not determine when a private AI developer is liable for damages.

Case 8. Loomis v. Wisconsin is not the only relevant authority: NJCM and Others v. State of the Netherlands (SyRI case, 2020)

Citation: ECLI:NL:RBDHA:2020:865, District Court of The Hague.

Facts: Civil-society organizations challenged the Dutch government's use of the System Risk Indication (SyRI), a system used to identify possible fraud in public-benefit and related contexts.

Legal principle: The court found the relevant SyRI legislation incompatible with Article 8 of the European Convention on Human Rights because the framework did not adequately balance the public interest with the right to respect for private life.

Relevance to AI liability: Although the case concerned human rights and public-law review rather than a conventional damages action, it illustrates how opaque risk-profiling systems can cause legally significant harm. A related civil claim would still need to establish a recognized cause of action and satisfy its applicable requirements.

Civil-law significance: The decision demonstrates that automated risk assessment is subject to legal constraints, including proportionality and privacy protections. It does not create a general rule of strict civil liability for AI systems.

6. Essential elements of an AI-related tort claim

Although the details vary by jurisdiction and cause of action, a typical negligence claim requires the claimant to establish the following elements.

1. Duty of care

The defendant owed the claimant a legally recognized duty. For example, a manufacturer may owe duties concerning the safety of an AI-enabled product.

2. Breach of duty

The defendant failed to meet the applicable standard of care, such as by neglecting required testing, ignoring known defects, or failing to implement appropriate safeguards.

3. Causation

The claimant must establish the legally required connection between the breach or defect and the harm. This can be difficult when several systems and actors contributed to an incident.

4. Legally recoverable damage

The claimant must demonstrate harm recognized by the applicable cause of action, such as physical injury, property damage, or qualifying financial loss.

These are the usual elements of negligence, not universal requirements for every liability model. Strict product liability, statutory compensation, and privacy-related claims may have different elements.

7. The problem of causation and proof

Causation is one of the most difficult issues in AI-related litigation.

Suppose an autonomous vehicle causes an accident. The injury might result from a defective sensor, an unsuitable model, inaccurate training data, negligent maintenance, poor road conditions, or a human operator's conduct.

A claimant may need evidence concerning:

The version of the AI system operating at the time.

The data inputs and system logs.

The system's design and known limitations.

Testing, validation, and safety documentation.

Software updates and maintenance history.

Human interventions and warnings.

The relationship between the alleged defect and the injury.

Where evidence is held exclusively by a manufacturer or service provider, disclosure procedures and applicable evidentiary presumptions may become particularly important.

The revised EU Product Liability Directive is significant in this context because it addresses disclosure and evidentiary difficulties in qualifying product-liability claims. However, the availability of disclosure, presumptions, or burden-shifting depends on the applicable rules and circumstances.

A court should not assume that the defendant is liable merely because an AI system was involved. Equally, technical complexity should not make a legitimate claim practically impossible to pursue.

8. Liability of different AI participants

These categories are not mutually exclusive. More than one party may be liable, but responsibility must be established under the governing law.

9. Remedies and defences

Remedies

Depending on the cause of action and jurisdiction, a successful claimant may obtain:

Compensation for physical injury, property damage, or other recognized losses.

Medical expenses and qualifying future-loss compensation.

Repair or replacement costs under applicable product-liability rules.

Injunctive relief or other orders to prevent continuing unlawful conduct.

Statutory remedies for qualifying privacy or consumer-protection violations.

Contribution or indemnity between responsible parties.

Punitive or exemplary damages are available only where the relevant legal system and cause of action permit them.

Potential defences

A defendant may argue that no duty was owed, the system was not defective, the standard of care was satisfied, the alleged breach did not cause the harm, or the damage was not legally recoverable.

Other possible issues include contributory negligence, assumption of risk where recognized, statutory defences, limitation periods, and compliance with applicable product-liability rules.

Compliance with a regulation may be relevant evidence, but it does not necessarily eliminate every tort claim.

10. Recommendations for an effective comparative model

A balanced AI tort-liability framework should:

Clarify responsibility: Identify the respective duties of developers, manufacturers, deployers, and operators.

Improve traceability: Preserve relevant records of model versions, testing, inputs, and incidents.

Facilitate proportionate disclosure: Enable claimants to obtain relevant technical evidence through lawful procedures.

Apply suitable liability standards: Distinguish negligence from strict product liability and statutory compensation.

Address complex causation: Consider appropriate evidentiary presumptions where legislation permits and the facts justify them.

Preserve insurance and compensation mechanisms: Use mandatory insurance or other schemes where appropriate for high-risk activities.

Avoid automatic immunity: Do not permit organizations to escape responsibility merely because a decision was generated by an algorithm.

Avoid automatic liability: Do not make every developer, manufacturer, or deployer liable for every AI error without applying the relevant legal test.

11. Conclusion

Comparative AI-related tort liability requires courts and legislators to adapt established principles to increasingly complex technological systems.

The eight cases discussed provide important foundations. Donoghue v. Stevenson, Caparo, and MacPherson illuminate duty of care and responsibility toward foreseeable users. Escola and A v. National Blood Authority help distinguish negligence from product liability. Boston Scientific addresses systemic product risk, while State v. Loomis and the SyRI judgment illustrate concerns about opaque automated assessments and the need for safeguards.

These cases do not establish a universal AI-liability rule. Their significance lies in showing how existing legal principles can address new technological risks while respecting the differences between national legal systems.

The central conclusion is that AI-related tort law must balance effective compensation for injured persons with fair allocation of responsibility. Developers and deployers should be accountable for legally actionable failures, but liability must remain grounded in the applicable law, the evidence, and the relationship between the defendant's conduct and the harm suffered.