1. Introduction

Artificial Intelligence (AI) is transforming healthcare, transportation, banking, manufacturing, education, and other sectors. However, its increasing use raises significant questions of civil liability when an AI system causes personal injury, property damage, financial loss, privacy violations, or other legally recognized harm.

Traditional civil law generally assumes that damage can be attributed to a human being, a business, a manufacturer, or another legally responsible person. AI complicates this approach because an AI system may make predictions, generate outputs, or perform actions autonomously, sometimes in ways that its developers or users cannot fully anticipate.

AI liability refers to the legal responsibility of persons or entities for harm caused by the development, supply, deployment, or use of an AI system. Depending on the jurisdiction and circumstances, liability may arise under negligence, product liability, contract law, vicarious liability, consumer protection, data protection, or other civil law principles.

A fundamental distinction is that AI is generally not treated as an independent legal person capable of bearing civil liability in its own name. Liability is ordinarily assessed against the natural or legal persons involved in creating, supplying, operating, or using the system.

This discussion compares India, the United States, the United Kingdom, the European Union, and selected European civil-law jurisdictions, and examines more than six judicial decisions relevant to AI liability. An important qualification is that most of the decisions discussed below concern autonomous machines, software, algorithms, or related technologies rather than generative AI specifically. They provide principles that courts may apply to AI-related disputes, but they should not all be described as direct AI liability precedents.

2. Meaning and scope of civil liability in AI

Civil liability is the legal obligation to compensate for injury, loss, or damage, or to provide another civil remedy where the requirements of the applicable law are satisfied.

In the AI context, a claim may arise when:

A medical AI system recommends an unsafe diagnosis or treatment.

An autonomous vehicle causes an accident.

An algorithm discriminates against a loan applicant or job candidate.

A generative AI system produces defamatory or misleading content.

An AI-enabled product malfunctions and causes physical injury.

A company uses personal data unlawfully to train or operate an AI model.

An automated trading system causes financial loss through negligent design or deployment.

The central legal questions are: What duty was owed? Was that duty breached? Did the breach cause legally recognized damage? Who controlled or contributed to the risk? Which legal rules govern the claim?

Principal forms of civil liability

1. Negligence

A developer, manufacturer, hospital, employer, or AI user may be liable for failing to exercise reasonable care in designing, testing, validating, supervising, or using an AI system.

2. Product liability

A manufacturer or other responsible supplier may be liable when a defective AI-enabled product causes legally recognized harm, subject to the relevant jurisdiction's statutory requirements.

3. Vicarious liability

An employer or principal may be responsible for wrongful acts committed by an employee or agent within the applicable scope of employment or agency rules. This does not automatically make a business liable for every AI output.

4. Contractual liability

A party may be liable for breaching contractual promises about an AI system's performance, security, accuracy, service availability, or compliance, subject to contract terms and applicable law.

5. Statutory and data-related liability

Consumer protection, privacy, discrimination, and sector-specific legislation may create duties, remedies, penalties, or compensation rights independently of—or alongside—ordinary negligence principles.

3. Why traditional civil law faces difficulties with AI

A. Identification of the responsible party

An AI system may involve a model developer, data provider, hardware manufacturer, software integrator, distributor, business operator, and end user. Harm may result from their combined conduct.

For example, a hospital may use a third-party diagnostic AI tool. The model developer may have supplied an inadequately validated system, the hospital may have used it outside its intended purpose, and the clinician may have ignored warning signs. The court must determine which acts or omissions were legally causative.

B. The black-box problem

Some AI systems, especially complex machine-learning models, do not provide a simple explanation for individual outputs. A claimant may struggle to establish what went wrong, whether the model was defective, and whether the defendant's conduct caused the injury.

However, a lack of explainability does not automatically establish liability or excuse a defendant from responsibility. Courts must apply the relevant evidentiary and substantive rules.

C. Causation

A claimant ordinarily must establish the legally required causal connection between a defendant's conduct and the damage. In an AI dispute, the relevant chain may include training data, model design, deployment configuration, user instructions, human supervision, and subsequent decisions.

D. Standard of care

Traditional negligence law asks what a reasonably careful person or organization would have done in the circumstances. AI complicates this assessment because risks may be probabilistic, change after deployment, and depend on the operational environment.

Relevant precautions may include pre-deployment testing, bias assessment, cybersecurity measures, human oversight, monitoring, warnings, and timely correction of known defects.

E. Proof and access to evidence

Important evidence may be held by AI developers or operators, including training documentation, system logs, model versions, risk assessments, and incident reports. Disclosure rules, expert evidence, confidentiality, and trade-secret protections can affect a claimant's ability to prove the case.

F. Rapid technological change

A system may change through updates, fine-tuning, integration with other software, or continuous learning. Liability analysis therefore needs to consider not only the original design but also maintenance, post-deployment modifications, and the conduct of the parties operating the system.

4. Comparative AI liability frameworks

4.1 India

India does not have a single, comprehensive statute that establishes a complete civil liability regime specifically for AI. AI-related disputes may be addressed through existing tort principles, consumer protection legislation, contract law, product liability rules, data protection law, and sector-specific regulation.

A. Law of torts and negligence

A claimant may seek compensation by establishing the applicable elements of negligence, ordinarily including:

A duty of care owed by the defendant.

Breach of that duty.

A causal connection between the breach and the harm.

Legally recognized damage.

For example, if a hospital negligently relies on an inadequately validated AI diagnostic system and a patient suffers injury, potential defendants could include the hospital, the clinician, and, depending on the facts and applicable law, the system's manufacturer or supplier.

The mere fact that an AI output is incorrect does not automatically prove negligence. The court must examine the applicable standard of care, the parties' roles, foreseeable risks, warnings, and the circumstances of deployment.

B. Consumer Protection Act, 2019

The Act provides consumer remedies and a product liability framework. Relevant provisions include Sections 2(34) and 2(35), concerning product liability and product liability actions, and Sections 82–87, which address product liability and the responsibility of product manufacturers, product service providers, and product sellers.

These provisions may become relevant where a defective AI-enabled product or a deficient AI-related service causes legally actionable harm to a consumer. Whether a particular standalone software service falls within a specific statutory category depends on the facts and the statutory definitions.

C. Information technology and data protection

The Information Technology Act, 2000, and applicable rules may be relevant to certain technology-related wrongs, including unauthorized access, data misuse, or other conduct within the legislation's scope.

The Digital Personal Data Protection Act, 2023, is also relevant to AI systems that process digital personal data. Its applicability, operative provisions, and implementation status must be assessed for the relevant date and circumstances. A data protection violation does not automatically establish every element of a separate tort claim.

D. Contract law

Under the Indian Contract Act, 1872, contractual disputes may arise from warranties, representations, service-level commitments, security obligations, or failures to deliver the agreed functionality. Remedies depend on the contract, the nature of the breach, and the applicable legal requirements.

E. Absolute liability

In appropriate circumstances, Indian environmental and industrial liability principles may offer useful comparisons. However, the doctrine of absolute liability developed for hazardous industrial activities should not be automatically extended to every AI system.

Assessment: India's existing legal principles provide potential routes to redress, but difficult questions remain regarding algorithmic opacity, multi-party responsibility, technical evidence, and allocation of loss.

4.2 United States

The United States has no single nationwide AI civil liability statute governing all AI-related injuries. Liability is shaped by state tort law, product liability doctrine, contract law, consumer protection statutes, federal civil rights legislation, privacy laws, and sector-specific requirements.

A. Negligence

Developers and deployers may face negligence claims where they fail to take reasonable precautions against foreseeable harm. Relevant considerations may include testing, foreseeable misuse, monitoring, and the provision of adequate warnings.

B. Product liability

Depending on state law, claims may involve defective design, manufacturing defects, or failure to warn. Courts must determine whether software or an AI-enabled service qualifies as a product under the applicable legal doctrine and whether the claimant can establish the required defect and causation.

C. Algorithmic discrimination

Employment, lending, housing, and other automated decisions may engage applicable anti-discrimination statutes. An employer cannot necessarily avoid legal responsibility simply by outsourcing a decision to an algorithm.

D. Contractual and consumer claims

A company may face claims based on misleading representations, contractual non-performance, or unfair business practices, where the relevant law and facts support such claims.

Assessment: The American framework is flexible and allows existing legal doctrines to be applied to new technology, but outcomes can differ considerably across states and factual contexts.

4.3 United Kingdom

The United Kingdom principally relies on established negligence principles, the Consumer Protection Act 1987 for qualifying defective products, contract law, and applicable statutory duties.

A. Negligence

A claimant generally needs to establish a duty of care, breach, causation, and actionable damage. AI-related disputes can therefore turn on whether a defendant reasonably assessed the risks of deploying a system in a particular setting.

B. Product liability

The Consumer Protection Act 1987 implements a statutory regime for damage caused by defective products. Whether a particular AI system, software component, or digital service falls within the relevant statutory definition must be assessed carefully.

C. Professional responsibility

Doctors, lawyers, engineers, and other professionals remain subject to applicable professional duties when they use AI. Reliance on automated recommendations does not, by itself, eliminate their responsibilities.

D. Human oversight

Where an AI system supports a professional or organizational decision, courts may examine the extent of human review, the reasonableness of reliance, and whether foreseeable risks were addressed.

Assessment: The UK model is adaptable because of its established common-law negligence principles, but AI-specific questions concerning evidence, software defects, and distributed responsibility remain fact-sensitive.

4.4 European Union

The European Union has developed a more explicit regulatory framework for AI safety and defective products. Two legal instruments must be distinguished: the EU AI Act and the revised Product Liability Directive.

A. EU AI Act

Regulation (EU) 2024/1689 establishes a risk-based regulatory framework for AI. It imposes obligations on relevant providers, deployers, and other actors according to the applicable provisions and the system's classification.

Depending on the system and the relevant obligations, requirements may include risk management, documentation, logging, transparency, human oversight, and monitoring.

However, regulatory non-compliance and civil liability are not identical. A breach may be relevant evidence or trigger regulatory consequences, but compensation requires a valid cause of action and satisfaction of the applicable legal conditions.

B. Revised Product Liability Directive

Directive (EU) 2024/2853 expressly includes software, including AI systems, within its product liability framework. It retains a no-fault liability model for qualifying defective products, subject to its requirements concerning defect, damage, causation, liable economic operators, and applicable defenses. The Directive entered into force on 8 December 2024, with a transposition deadline of 9 December 2026. It applies to products placed on the market or put into service after that date under its transitional provisions.

EUR-Lex

+1

This is especially significant because it expressly addresses software and AI in a modernized product liability framework.

C. Proposed AI Liability Directive

The proposed AI Liability Directive was intended to address certain difficulties in fault-based claims involving AI, including access to evidence and proof of causation. The proposal was withdrawn from the European Commission's legislative agenda in 2025. It should therefore not be described as an enacted, operative EU liability directive.

Comparative AI

+1

Assessment: The EU has strengthened product liability rules and adopted a dedicated AI regulatory regime, but fault-based civil liability continues to depend significantly on applicable national law.

4.5 Comparative summary

The comparison shows that the key difference is not whether civil liability exists, but how each jurisdiction allocates risk, treats software defects, regulates AI deployment, and enables injured parties to prove their claims.

5. Judicial decisions relevant to AI liability

The following decisions provide a combination of direct algorithmic or AI-related jurisprudence and foundational principles applicable by analogy. They should be distinguished from judgments that have actually held an AI developer liable for personal injury: the latter category remains much narrower.

Case 1. Donoghue v. Stevenson 19321932 AC 562

Jurisdiction: United Kingdom · House of Lords

Facts: A consumer allegedly became ill after consuming ginger beer containing a decomposed snail. The product had been manufactured and sold in circumstances where the consumer had no direct contract with the manufacturer.

Legal issue: Could a manufacturer owe a duty of care to an ultimate consumer despite the absence of a direct contractual relationship?

Decision and principle: The House of Lords recognized a manufacturer's duty of care toward the ultimate consumer in the relevant circumstances. The decision became a foundation of modern negligence law and the neighbour principle.

Relevance to AI liability: The case helps explain why an AI developer or manufacturer may owe duties to people who are foreseeably affected by an AI-enabled product, even if those people never contracted directly with the developer.

For example, a person injured by a defective AI-enabled medical device might seek to establish that the manufacturer owed a duty to foreseeable users or patients.

Limitation: The case did not concern AI, and it does not establish that every AI developer owes a duty to every person affected by an output. Duty, proximity, foreseeability, and other jurisdiction-specific requirements must still be considered.

Case 2. Caparo Industries plc v. Dickman 19901990 2 AC 605

Jurisdiction: United Kingdom · House of Lords

Facts: Caparo Industries relied on audited financial information in circumstances involving a company's financial position and subsequently acquired shares. The litigation concerned the scope of the auditors' duty of care.

Legal issue: When should the law recognize a duty of care for economic loss arising from information or statements supplied by another party?

Decision and principle: The judgment is associated with the familiar framework of foreseeability of harm, proximity between the parties, and whether it is fair, just, and reasonable to impose a duty in the circumstances.

Relevance to AI liability: AI outputs are frequently used to make decisions about credit, investments, employment, and commercial transactions. Caparo helps illustrate why foreseeable reliance on an AI-generated report does not automatically create a duty of care to every possible recipient.

A court may need to consider the intended audience, the purpose for which the output was generated, the relationship between the parties, and the foreseeability of the relevant loss.

Limitation: The Caparo framework should not be applied mechanically to every negligence claim. Its precise role depends on the jurisdiction and type of duty alleged.

Case 3. Jacob Mathew v. State of Punjab (2005) 6 SCC 1

Jurisdiction: India · Supreme Court of India

Facts: The case concerned allegations of medical negligence following the treatment of a patient. The Supreme Court considered the standards governing professional negligence and the distinction between civil and criminal negligence.

Legal issue: What level of care is expected from a medical professional, and what degree of negligence is required for criminal liability?

Decision and principle: The Court discussed professional standards of care and emphasized that criminal negligence requires a higher degree of negligence than ordinary civil negligence. A professional is not automatically negligent merely because treatment does not succeed or another professional might have adopted a different approach.

Relevance to AI liability: This principle is important when medical professionals use AI-assisted diagnostic or treatment systems. A court may need to assess whether the professional acted reasonably in selecting, interpreting, and relying on the system.

The decision also cautions against equating an incorrect AI-assisted outcome with automatic legal liability.

Limitation: This judgment is not an AI liability ruling. Its relevance lies in the general principles of professional negligence, which must be applied with reference to the actual conduct, professional standards, and causal facts.

Case 4. M.C. Mehta v. Union of India (1987) 1 SCC 395

Jurisdiction: India · Supreme Court of India

Facts: The litigation arose from the escape of oleum gas from an industrial establishment in Delhi and the resulting public safety concerns.

Legal issue: What liability should attach to an enterprise conducting inherently hazardous or dangerous activities when harm results from those activities?

Decision and principle: The Supreme Court articulated the doctrine of absolute liability for enterprises engaged in hazardous or inherently dangerous activities. This doctrine is distinct from the English rule in Rylands v. Fletcher and does not simply reproduce its exceptions.

Relevance to AI liability: The decision provides a comparative example of a legal system developing a stringent liability principle in response to serious risks associated with industrial activity.

It may inform academic discussion about whether certain exceptionally dangerous AI applications warrant specialized liability or compensation arrangements.

Limitation: The doctrine is not a general rule that all AI developers are absolutely liable whenever an AI system causes harm. Its application is linked to the relevant Indian legal principles and the nature of the activity.

Case 5. State v. Loomis, 881 N.W.2d 749 (Wis. 2016)

Jurisdiction: United States · Wisconsin Supreme Court

Facts: Eric Loomis challenged the use of a COMPAS risk-assessment report during criminal sentencing. COMPAS is an algorithmic tool used to estimate recidivism risk.

Legal issue: Could a sentencing court consider an algorithmic risk assessment, and what limitations should apply to its use?

Decision and principle: The Wisconsin Supreme Court upheld the sentencing court's use of the report under the circumstances, while emphasizing limitations and caution regarding reliance on the risk assessment. The court addressed concerns including the proprietary nature of the tool, the ability to scrutinize its methodology, and the risk of inappropriate reliance on a risk score.

Relevance to AI liability: Although this was a criminal sentencing case rather than a civil damages action, it is highly relevant to algorithmic accountability. It demonstrates the legal importance of understanding a system's limitations, communicating uncertainty, and ensuring that automated assessments do not replace appropriate human judgment.

In a civil claim involving an AI credit-scoring, employment, or insurance system, analogous questions may arise about transparency, fairness, and the extent to which a decision-maker reasonably relied on the output.

Limitation: Loomis did not establish a general civil cause of action against AI developers. Nor did it hold that every algorithmic risk score is legally impermissible.

Case 6. SCHUFA Holding AG, Case C-634/21, Court of Justice of the European Union, judgment of 7 December 2023

Jurisdiction: European Union · CJEU

Facts: The dispute concerned credit-scoring practices and the processing of personal data in circumstances where a score could play a significant role in a third party's decision about an individual.

Legal issue: Under the EU General Data Protection Regulation (GDPR), when can the creation of a credit score itself constitute automated individual decision-making with legal or similarly significant effects?

Decision and principle: The Court interpreted Article 22 GDPR in a way that prevents the protection afforded by the provision from being circumvented simply because a scoring agency generates the score while another entity formally makes the ultimate decision. Where the score plays a determining role in the decision, the scoring activity may fall within the provision's scope, subject to the relevant legal conditions.

Relevance to AI liability: The case is important for automated decision-making in credit, insurance, recruitment, and similar fields. It shows that responsibility cannot always be avoided by separating the generation of an algorithmic score from the final decision.

For civil claims, the judgment can be relevant to determining which actors are subject to data protection obligations. Compensation, however, depends on the applicable legal basis and its requirements.

Limitation: The judgment concerns the GDPR and automated decision-making, not a universal tort rule imposing strict liability for algorithmic errors.

Case 7. Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023)

Jurisdiction: United States · U.S. District Court, Southern District of New York

Facts: Lawyers representing a plaintiff in litigation against Avianca submitted legal authorities that included fictitious judicial decisions generated using ChatGPT. The fabricated authorities were presented in court filings.

Legal issue: Could lawyers avoid responsibility for false legal citations by claiming that they had relied on a generative AI system?

Decision and principle: The court imposed sanctions on the lawyers involved. It emphasized their obligations to verify the accuracy of authorities submitted to the court and found their handling of the fabricated citations inconsistent with those professional responsibilities.

Relevance to AI liability: The case illustrates an essential principle: using AI does not automatically transfer responsibility for an output from the human user to the system's developer.

A lawyer, business, or professional who relies on generative AI may still be responsible for taking reasonable steps to verify high-stakes information. Depending on the circumstances, unverified AI output may lead to procedural sanctions, professional consequences, contractual disputes, or other legal claims.

Limitation: This was principally a sanctions decision concerning litigation conduct, not a civil damages judgment against the AI developer. It does not settle the question of when a model provider may be liable for an inaccurate output.

Case 8. Dinerstein v. Google, LLC, 73 F.4th 502 (7th Cir. 2023)

Jurisdiction: United States · U.S. Court of Appeals for the Seventh Circuit

Facts: The case involved allegations concerning the sharing of patient medical information with Google in connection with a healthcare technology project. The litigation raised issues involving privacy, consent, and the handling of sensitive information.

Legal issue: Did the plaintiff establish a legally sufficient basis for the asserted claims arising from the use and sharing of medical information?

Decision and principle: The Seventh Circuit affirmed dismissal of the claims at issue, applying the relevant legal requirements to the alleged disclosure and use of medical data.

Relevance to AI liability: AI systems often depend on large datasets, including health and other sensitive information. The case provides a useful illustration of why the collection, transfer, and use of data may generate legal disputes independent of whether an AI system produces a technically accurate output.

In an AI-related dispute, an organization may face questions about authorization, privacy duties, consent, and the lawful basis for processing data.

Limitation: This was not a ruling establishing AI-specific liability, and it should not be characterized as a general determination that the use of patient data for AI is lawful or unlawful. The outcome depends on the particular facts and applicable legislation.

6. Additional case law and supporting principles

The following decisions further explain how traditional civil liability principles can be applied to AI-related disputes.

Case 9. Bolam v. Friern Hospital Management Committee 19571957 1 WLR 582

Jurisdiction: United Kingdom · Professional negligence

Facts: The case concerned a patient who suffered injury during electroconvulsive therapy and alleged that the hospital had failed to provide an appropriate standard of care.

Legal principle: The decision is associated with the Bolam test: a medical professional's conduct may be assessed by reference to a practice accepted as proper by a responsible body of professional opinion, subject to subsequent judicial developments and qualifications.

Relevance to AI: If a physician uses an AI diagnostic tool, the legal question is not simply whether the AI's output was wrong. It is whether the physician's conduct met the applicable standard of professional care, taking into account the circumstances, available evidence, system limitations, and the need for independent clinical judgment.

Limitation: The decision does not permit professionals to justify unreasonable conduct merely by pointing to an industry practice. The appropriate standard is jurisdiction-dependent, and the decision has been qualified by later authorities.

Case 10. Bolitho v. City and Hackney Health Authority 19981998 AC 232

Jurisdiction: United Kingdom · House of Lords

Facts: The case involved a child's respiratory emergency and the alleged failure of a doctor to attend and provide appropriate treatment.

Legal principle: The House of Lords confirmed that professional opinion relied upon in negligence proceedings must withstand logical analysis. A court is not invariably required to accept a professional practice merely because a responsible group of practitioners supports it.

Relevance to AI: An organization cannot necessarily defend its reliance on an AI system merely by arguing that the system is widely used in the industry. The court may need to examine whether the decision to rely on it was logically defensible, properly informed, and appropriate to the risks.

For example, if an AI system recommends withholding urgent treatment, a hospital's reliance on that recommendation may require scrutiny of the system's validation, limitations, and the clinical evidence available.

Limitation: The case concerns professional medical negligence, not the liability of AI developers. Its importance is analogical: technological adoption does not eliminate judicial scrutiny of the reasonableness of a decision.

7. Comparative analysis of the cases

The decisions above reveal several principles relevant to the development of civil liability for AI.

These authorities do not establish one uniform global rule. Instead, they demonstrate how established doctrines can address different components of AI-related harm.

8. Who should be liable when AI causes harm?

A central debate in comparative AI law is how responsibility should be distributed among the participants in the AI supply chain.

Consider an autonomous vehicle that strikes a pedestrian. The accident may involve a defective perception model, faulty sensors, inadequate integration, poor maintenance, improper deployment, or a driver's failure to intervene.

The legally responsible party will depend on the actual cause of the accident and the law governing the claim.

AI developer or model provider

Potential responsibility for negligent development, inadequate testing, misleading documentation, or failure to address known defects, where the relevant legal elements are established.

Manufacturer or system integrator

Potential responsibility for defective hardware, unsafe integration of software and machinery, or inadequate warnings and safeguards.

Business or institutional deployer

Potential responsibility for using a system outside its intended purpose, inadequate staff training, improper configuration, or insufficient monitoring.

Professional or end user

Potential responsibility for unreasonable reliance, failure to verify consequential outputs, disregard of warnings, or misuse of the system.

Data provider or other intermediary

Potential responsibility where a legally actionable failure involving data quality, disclosure, security, or contractual obligations contributed to the harm.

These categories are not mutually exclusive. Several parties may be responsible, but joint liability, contribution, indemnification, and allocation of damages are governed by the applicable legal system.

Importantly, participation in the AI supply chain alone does not establish liability. A claimant must identify an applicable legal basis and satisfy its requirements.

9. Burden of proof and the black-box problem

One of the most significant challenges in AI litigation is proving what happened inside a complex system.

In an ordinary product defect case, a claimant may be able to identify a broken component or a manufacturing error. With a machine-learning system, the problem may arise from training data, a model limitation, an unexpected input, an integration failure, or a change introduced through an update.

The claimant may not have access to the relevant records.

Possible legal and procedural responses

1. Disclosure of technical evidence

Courts may use applicable disclosure or discovery rules to obtain relevant records, including system logs, model documentation, testing results, and incident reports.

2. Expert evidence

Technical experts may assist in determining whether a system was defective, whether reasonable precautions were taken, and whether the alleged failure caused the damage.

3. Evidentiary presumptions

Some legal frameworks permit presumptions or adjustments to proof requirements under specified circumstances. Such measures must come from applicable law; a court cannot simply presume liability because an AI system is difficult to understand.

4. Record-keeping requirements

Maintaining accurate records of model versions, updates, human interventions, and system outputs can help identify the cause of an incident and allocate responsibility fairly.

5. Regulatory compliance evidence

Compliance with applicable AI safety and governance requirements may be relevant to the assessment of reasonable care. It does not necessarily eliminate civil liability, and non-compliance does not automatically establish every element of a damages claim.

The revised EU Product Liability Directive expressly addresses the evidentiary difficulties that arise in technically complex product liability cases. However, its provisions operate within the Directive's scope and applicable national implementation. They should not be generalized to every AI dispute worldwide.

10. Strict liability versus fault-based liability

Two major approaches dominate the debate over AI-related compensation.

Fault-based liability

Negligence and related fault-based claims

The claimant generally must establish the relevant duty or obligation, a breach, causation, and legally recognized damage.

Advantages

Encourages responsible design and deployment.

Links responsibility to blameworthy conduct where fault is required.

Allows courts to consider the circumstances and roles of different parties.

Challenges

Technical complexity can make fault difficult to prove.

Evidence may be controlled by the defendant.

Multiple parties may contribute to a single incident.

Strict or no-fault liability

Liability for qualifying defective products

Where a statutory strict liability regime applies, the claimant need not prove negligence, but must still establish the required elements, typically including a defect, legally recognized damage, and causation.

Advantages

Reduces the need to prove negligent conduct.

Allocates certain product-related risks to responsible economic operators.

Can improve access to compensation for qualifying injuries.

Challenges

The claimant must still prove the statutory requirements.

Not every AI service or output necessarily falls within a product liability regime.

Defenses, exclusions, and limits on recoverable damage may apply.

Strict liability does not mean automatic liability for every harmful AI outcome. Equally, fault-based liability does not require proof that a human consciously intended the harm.

The most appropriate model may depend on the type of AI system, the severity and foreseeability of the risks, the defendant's control over the system, and the availability of insurance.

11. Illustrative application: AI-assisted medical negligence

Assume a hospital deploys an AI system to identify patients at risk of internal bleeding. The system incorrectly assigns a low-risk score to a patient, and the hospital delays treatment. The patient suffers serious injury.

A court might examine the claim through the following sequence.

Identify the duty

Did the hospital or clinician owe the patient a duty to provide reasonable medical care? Did the manufacturer or supplier owe any relevant duty under the applicable law?

Identify the alleged breach

Was the system inadequately tested? Were known limitations ignored? Was the AI used outside its validated setting? Were clinical warning signs overlooked?

Establish causation

Did the AI-related failure actually contribute to the delayed treatment and injury? Would the injury likely have occurred even if reasonable precautions had been taken?

Identify the legally responsible parties

The hospital, clinician, system manufacturer, or another supplier may be implicated, depending on their conduct and the applicable legal rules.

Assess damages and defenses

The court would determine the recoverable loss, the effect of any relevant defenses, and whether responsibility should be apportioned among multiple parties.

This example illustrates why the mere presence of AI is not enough to determine liability. The decisive questions concern the applicable legal duties, the conduct of each participant, the causal chain, and the nature of the damage.

12. Reforms needed for a more effective AI liability system

A balanced civil liability framework should protect people harmed by AI without making innovation legally impossible.

The following reforms are worth considering:

1. Clear allocation of responsibilities: Define the duties of developers, manufacturers, deployers, and professional users according to their actual roles and control over the system.

2. Improved access to evidence: Establish practical mechanisms for obtaining relevant technical documentation and incident records while protecting legitimate confidentiality interests.

3. Risk-based standards of care: Require stronger safeguards where AI is used in high-stakes areas such as healthcare, transport, employment, credit, and public administration.

4. Meaningful human oversight: Clarify when human review is required and what constitutes adequate review in consequential decisions.

5. Appropriate insurance arrangements: Encourage insurance or compensation mechanisms for applications capable of causing serious harm, where justified by risk and feasibility.

6. Clear treatment of software updates: Determine when responsibility attaches to the original developer, the party modifying the system, or the entity deploying an updated version.

7. Effective remedies for data-related harm: Ensure that existing privacy and data protection rights can be enforced where AI systems unlawfully process personal information.

8. International coordination: Improve consistency across jurisdictions so that the allocation of responsibility does not depend unnecessarily on gaps between national legal systems.

9. Protection against overbroad liability: Preserve appropriate defenses and distinguish unavoidable errors from defective products, negligent conduct, and misuse.

13. Conclusion

Civil liability remains a central mechanism for addressing harm caused by AI. Although AI systems may operate autonomously in a technical sense, the law generally assesses responsibility through the conduct and obligations of the people and organizations that develop, supply, deploy, and use them.

The comparative position can be summarized as follows:

India relies on existing tort principles, consumer protection, contract law, data protection, and relevant sectoral legislation.

The United States uses a diverse combination of state tort law, product liability, contractual remedies, and federal and state statutory protections.

The United Kingdom applies established negligence and product liability principles to the circumstances of AI-related disputes.

The European Union combines AI-specific regulatory obligations with an updated product liability framework, while fault-based civil liability remains substantially dependent on applicable national law.

The cases discussed—particularly Donoghue v. Stevenson, Jacob Mathew, M.C. Mehta, State v. Loomis, SCHUFA, and Mata v. Avianca—demonstrate that many foundational legal principles remain relevant even as the technology changes.

Nevertheless, it is important to distinguish established precedent from legal analogy. These cases do not collectively establish a universal rule of AI liability, and most do not directly decide whether an AI developer must compensate someone injured by an AI system.

The central principle is that technological autonomy should not create a legal accountability vacuum. A sound framework should make compensation realistically accessible where the law recognizes liability, while allocating responsibility according to fault, product defects, statutory duties, causation, and the actual roles of the parties involved.

Academic note: This is a comparative legal overview, not jurisdiction-specific legal advice. For litigation or formal academic submission, verify the current statutory text, procedural rules, and subsequent judicial treatment of each authority in the relevant jurisdiction.