1. Introduction

Artificial General Intelligence (AGI) refers to a hypothetical or prospective form of artificial intelligence capable of performing a broad range of intellectual tasks, learning across different domains, and adapting to unfamiliar situations at a level comparable to or exceeding human capabilities. Unlike narrow AI, which is designed for particular tasks, a highly capable AGI system could potentially operate across healthcare, finance, transport, legal services, scientific research, industrial production, and autonomous decision-making.

As of October 2026, there is no universally accepted legal definition of AGI and no comprehensive, globally harmonized civil-liability regime specifically designed for it. Existing legal principles—such as negligence, product liability, contractual responsibility, vicarious liability, and liability for hazardous activities—remain the principal starting points for analyzing possible AGI-related harm. Recent comparative scholarship highlights the difficulties that advanced autonomy, opacity, and distributed control create for these traditional doctrines.

Cambridge Repository

+1

Civil law and comparative AGI liability frameworks examine how responsibility should be allocated when an advanced AI system causes personal injury, property damage, financial loss, privacy violations, discrimination, or other legally recognized harm.

The central question is:

Who should bear civil liability when an intelligent system produces harmful consequences that no single human participant fully predicted or controlled?

Possible defendants include:

The developer or model provider.

The manufacturer of a product incorporating AGI.

The organization deploying or operating the system.

A business that relies on its decisions.

A professional who uses its output without adequate verification.

A person who deliberately misuses the system.

Multiple parties whose conduct jointly contributes to the harm.

The challenge is not simply to determine whether an AI system made an error. It is to establish a legally recognized duty, breach or other basis of liability, causation, compensable damage, and an effective remedy.

2. Meaning and scope of AGI civil liability

Civil liability is the legal obligation to compensate another person for damage caused by conduct, a defective product, breach of contract, or another legally recognized event.

AGI liability extends this analysis to systems that may possess substantial operational autonomy, adapt to changing circumstances, use external tools, and make decisions across multiple domains.

For example, consider an advanced AI system used by a hospital network. The system recommends treatment, manages equipment, allocates resources, and communicates with other software. A harmful outcome could result from a defective model, inadequate clinical data, poor integration, negligent supervision, or an unsafe deployment decision.

Traditional legal analysis would ask:

Was the system or associated product defective?

Did a developer, deployer, manufacturer, or professional breach a duty of care?

Was the injury sufficiently connected to the alleged breach or defect?

Was the harm foreseeable or otherwise legally attributable?

Which party had the practical ability to prevent or mitigate the risk?

What compensation or other civil remedy is appropriate?

AGI complicates these questions because its behavior may be difficult to predict, its internal reasoning may not be fully explainable, and responsibility may be distributed among several organizations.

However, technical autonomy does not automatically establish legal personhood or remove the responsibility of the humans and organizations involved. Existing law generally directs liability toward legally recognized persons and entities, subject to the particular jurisdiction's rules.

3. Principal theories of AGI liability

A. Fault-based liability and negligence

Fault-based liability requires a claimant to establish the elements prescribed by the relevant legal system. In a typical negligence claim, these include a duty of care, breach, causation, and damage.

For AGI developers, a duty of care might require reasonable measures concerning:

Testing and evaluation before deployment.

Security against foreseeable misuse.

Monitoring of dangerous capabilities.

Accurate documentation of limitations.

Appropriate safeguards against unauthorized actions.

Timely correction of known vulnerabilities.

Warnings about material risks.

For deployers, relevant duties may include restricting system permissions, conducting risk assessments, maintaining human oversight where appropriate, and verifying high-consequence decisions.

The applicable standard should reflect the system's capabilities, intended use, deployment environment, and foreseeable consequences—not merely the fact that the system uses AI.

Example: If a company deploys an advanced autonomous trading system without appropriate transaction limits despite known risks of uncontrolled trading, a negligence claim may arise if that failure causes foreseeable losses.

A major difficulty is proving what precautions were reasonable when a system behaves in an unexpected manner. Courts may need expert evidence, technical records, testing data, and evidence of industry practices.

B. Product liability

Product liability addresses harm caused by defective products. Depending on the jurisdiction and the applicable legislation, liability may be strict, fault-based, or subject to specific statutory conditions.

Potential defects in an AGI-enabled product include:

Unsafe design.

Defective integration with hardware or software.

Inadequate warnings.

Failure to provide necessary updates.

Security weaknesses that create foreseeable physical or financial risks.

A failure to meet applicable safety requirements.

The European Union's revised Product Liability Directive, Directive (EU) 2024/2853, expressly updates the product-liability framework to address modern digital products, including software and AI-related products. Its application is governed by the Directive's scope and implementation provisions.

Cambridge Repository

+1

The central question for AGI would be whether the product was defective under the applicable legal test, not simply whether the system produced an undesirable result.

C. Contractual liability

Contractual liability arises when a party fails to perform an obligation imposed by an agreement.

AGI-related contractual disputes may concern:

AI service-level commitments.

Accuracy or reliability representations.

Data protection obligations.

Cybersecurity warranties.

Failure to deliver agreed functionality.

Unauthorized automated transactions.

Failure to maintain or update a system.

Allocation of losses between providers and commercial users.

A provider may contractually promise a specified level of service without guaranteeing every output. Whether an output constitutes a breach depends on the contract's language, applicable law, mandatory protections, and the circumstances.

Contractual indemnities and liability caps can allocate risk between commercial parties, but they cannot necessarily exclude liability to third parties or override mandatory statutory protections.

D. Vicarious and organizational liability

Vicarious liability may apply where the law attributes one party's wrongful conduct to another, commonly within employment or agency relationships. Its precise scope varies by jurisdiction.

Separately, an organization may face direct liability for its own governance failures, including inadequate supervision, unsafe procurement, deficient training, or failure to respond to known risks.

The use of AGI does not automatically create vicarious liability. Courts must identify the applicable legal relationship and the relevant statutory or common-law test.

E. Strict liability for hazardous activities

Some legal systems impose liability for particular hazardous activities without requiring proof of ordinary negligence. Other systems impose strict or specially regulated liability in defined fields, such as certain environmental, industrial, or transport risks.

A sufficiently dangerous AGI application might prompt proposals for a similar regime, particularly where the potential damage is severe and victims would struggle to establish fault.

However, strict liability should not be assumed to apply to all AGI systems under existing law. It depends on the applicable statute, established doctrine, or future legislative reform.

F. Liability for intentional misuse

A third party might deliberately use AGI to generate fraudulent instructions, manipulate financial transactions, impersonate individuals, or conduct cyberattacks.

The perpetrator may be liable for the resulting harm. A developer or deployer might also face liability if the relevant law recognizes a duty to prevent a foreseeable form of misuse and the evidence establishes a sufficient causal connection.

The mere possibility that a system can be misused does not automatically make its developer liable for every subsequent criminal or wrongful act.

4. Comparative AGI liability frameworks across jurisdictions

A. European Union

The EU combines product-safety legislation, data-protection rules, consumer protection, and risk-based AI regulation. The AI Act, Regulation (EU) 2024/1689, establishes obligations according to the type and risk of AI use, while the revised Product Liability Directive addresses defects in modern digital products.

For advanced AI systems, the relevant questions include whether the system falls within a regulated category, whether mandatory safety obligations were breached, and whether the applicable product-liability rules permit compensation for the particular damage.

The former proposed AI Liability Directive should not be treated as enacted law. General tort law and applicable EU legislation remain important to the analysis.

B. United States

US AGI-related liability would generally be examined through state tort law, product-liability doctrines, contract law, consumer-protection statutes, and applicable federal legislation.

Negligence, manufacturing defects, defective design, inadequate warnings, and misrepresentation may be relevant, depending on the facts and state law. Federal law can also affect particular claims or remedies.

A central issue is whether an AI developer or deployer owed a legally recognized duty and whether the claimant can establish causation. Product-liability standards and defenses differ between states.

C. United Kingdom

The UK framework combines negligence, contract, consumer protection, data protection, and product-safety law. The Consumer Protection Act 1987 may be relevant where an AI-enabled product falls within its statutory scope and causes qualifying damage.

A claimant may need to establish a duty of care, breach, factual causation, and legal causation under negligence principles. The nature of the product and the relationship between developer, supplier, and user may determine which cause of action is available.

The framework does not automatically impose strict liability on every provider of autonomous AI.

D. China

China's Civil Code provides a foundation for tort and product-liability claims, alongside legislation and regulations addressing data, cybersecurity, and AI-related services.

Depending on the circumstances, claims may involve defective products, infringement of personal rights, unlawful data processing, or negligent conduct. The applicable rules depend on the harm, the parties involved, and the specific legal regime.

China illustrates a model in which general civil-law principles operate alongside more specialized technology regulation.

E. India

India currently relies primarily on general principles of tort, contract, consumer protection, product liability, information technology, and other applicable legislation rather than a comprehensive AGI-specific civil-liability statute.

The Consumer Protection Act, 2019, contains product-liability provisions. The Indian Contract Act, 1872, may govern contractual claims, while tort principles may apply to negligent acts causing legally recognized harm.

A claimant would need to identify the responsible legal person or entity and establish the elements of the relevant cause of action. The precise allocation of responsibility between developer, deployer, and user remains fact-dependent.

5. At least eight important case laws relevant to AGI liability

Important distinction: There are not yet eight reported judicial decisions establishing a dedicated legal doctrine of AGI civil liability. The following cases are therefore divided between established negligence and product-liability precedents, decisions directly involving algorithmic or generative AI, and a major Indian strict-liability authority. They offer legal analogies, not proof that a court has already recognized a universal AGI-liability rule.

Case 1: Donoghue v. Stevenson (1932)

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

Facts: A consumer alleged illness after drinking ginger beer that contained the remains of a decomposed snail. She had no direct contract with the manufacturer.

Legal issue: Could a manufacturer owe a duty of care to a consumer who did not purchase the product directly from it?

Decision and principle: The House of Lords recognized a manufacturer's duty to take reasonable care toward consumers who could foreseeably be injured by a defective product.

Relevance to AGI: The case provides a foundational analogy for claims against AGI developers or manufacturers. A person harmed by an AI-enabled product may not have a contract with the developer, yet negligence principles may provide a route to compensation if the relevant duty, breach, causation, and damage are established.

The analogy is not automatic: courts must determine whether the particular defendant owed the claimant a duty under the applicable law.

Case 2: 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 motor vehicle allegedly contained a defective wheel that broke and injured its purchaser. The vehicle had been manufactured by Buick, although the purchaser had not bought it directly from Buick.

Legal issue: Could a manufacturer be liable in negligence to an ultimate consumer despite the absence of a direct contract?

Decision and principle: The court recognized manufacturer liability where a negligently made product was reasonably certain to endanger users and was supplied without adequate inspection.

Relevance to AGI: The case supports examining whether an AGI developer or manufacturer took reasonable care in designing, testing, and distributing a system capable of causing foreseeable harm.

It also illustrates why the absence of direct contractual privity does not necessarily defeat a tort claim.

Case 3: 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 allegedly exploded and injured a restaurant employee. The case concerned the manufacturer's responsibility for a potentially dangerous product.

Legal issue: How should responsibility for defective products be allocated when the injured person cannot readily prove the precise manufacturing error?

Decision and principle: The court's majority upheld the result for the injured plaintiff under the applicable doctrine of res ipsa loquitur. Justice Roger Traynor's influential concurring opinion argued for strict product liability.

Relevance to AGI: The case illustrates a persistent problem in AI litigation: victims may know that a system caused injury but lack access to the technical information needed to identify the precise failure.

A future AGI framework might respond through evidentiary presumptions or strict liability in defined circumstances. However, Escola itself did not decide an AGI or modern AI case, and evidentiary presumptions depend on the jurisdiction's law.

Case 4: Greenman v. Yuba Power Products, Inc. (1963)

Citation: 59 Cal. 2d 57, 377 P.2d 897, Supreme Court of California, United States.

Facts: A consumer was injured while using a power tool that allegedly malfunctioned during ordinary use.

Legal issue: Whether a manufacturer could be held strictly liable for a defective product without requiring the injured consumer to prove negligence.

Decision and principle: The California Supreme Court established an important foundation for strict products liability, holding that a manufacturer could be liable when a product placed on the market proved defective and caused injury during intended or reasonably foreseeable use.

Relevance to AGI: The decision supports the argument that, where the applicable statute or doctrine permits, victims should not always have to reconstruct a developer's internal decision-making to obtain compensation for damage caused by a defective AI-enabled product.

Nevertheless, whether a standalone AI service qualifies as a product, and whether strict liability applies, must be determined under the relevant jurisdiction's law.

Case 5: State v. Loomis (2016)

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

Facts: Eric Loomis challenged the use of a risk assessment generated by the COMPAS algorithm during criminal sentencing. He argued that the use of an opaque proprietary algorithm raised fairness and due-process concerns.

Legal issue: Whether reliance on an algorithmic risk assessment in sentencing violated the defendant's constitutional rights.

Decision and principle: The court upheld the sentence, subject to cautions about the limitations of COMPAS and the appropriate use of risk assessments. It emphasized that such tools should not determine sentencing outcomes by themselves.

Relevance to AGI: Although this was a criminal sentencing case rather than a civil damages action, it is significant for the governance of high-impact automated decisions.

It illustrates the importance of human oversight, awareness of model limitations, and safeguards against treating algorithmic outputs as conclusive. Those concerns may become central to civil claims arising from AGI-assisted medical, financial, employment, or administrative decisions.

Case 6: Mata v. Avianca, Inc. (2023)

Citation: 678 F. Supp. 3d 443, United States District Court for the Southern District of New York.

Facts: Lawyers representing a plaintiff submitted legal authorities generated by ChatGPT, including fabricated judicial decisions, in federal litigation. The court investigated the submissions and the lawyers' verification practices.

Legal issue: Whether lawyers could avoid professional responsibility for inaccurate legal submissions by relying on generative AI.

Decision and principle: The court imposed monetary sanctions on the responsible lawyers and law firm for their conduct, emphasizing that attorneys remain accountable for their representations to the court.

Relevance to AGI: This decision illustrates an important principle of human accountability: using an AI system does not ordinarily relieve a professional of duties imposed by law and professional rules.

In a future AGI-related negligence claim, a court might similarly examine whether a professional or organization unreasonably relied on an AI output instead of performing required verification. The precise legal test would depend on the circumstances.

Case 7: Mobley v. Workday, Inc. (2024–2025 proceedings)

Court: United States District Court for the Northern District of California.

Facts: A job applicant alleged that Workday's AI-assisted employment screening tools discriminated against him and other applicants on protected grounds.

Legal issue: Whether a technology vendor providing employment-screening tools could potentially be treated as an agent of employers or otherwise face liability under federal employment-discrimination law.

Decision and principle: In 2024, the court allowed the case to proceed on an agency-based theory at the pleading stage. In 2025, it granted preliminary collective-action certification for certain claims. These procedural rulings did not establish final liability for discrimination.

Relevance to AGI: The proceedings demonstrate that liability may extend beyond the organization making the final decision to a technology provider whose system materially shapes that decision, where the applicable legal elements are met.

For AGI systems used in hiring, credit, insurance, or access to services, the roles of developer, vendor, employer, and final decision-maker may need to be examined separately.

Case 8: Thomson Reuters Enterprise Centre GmbH v. Ross Intelligence Inc. (2025)

Citation: No. 1:20-cv-613, United States District Court for the District of Delaware, summary-judgment decision dated 11 February 2025.

Facts: Thomson Reuters brought a copyright action alleging that Ross Intelligence used protected Westlaw materials in developing a competing AI-assisted legal research product.

Legal issue: Whether the use of copyrighted legal materials to develop a competing AI system qualified as fair use under US copyright law.

Decision and principle: The court ruled against Ross on its fair-use defense concerning the uses at issue. The decision addressed specific copyright questions and should not be treated as a general determination of liability for all AI training.

Relevance to AGI: The case illustrates that advanced AI development does not eliminate existing intellectual-property obligations. An AGI developer could face civil claims if the development or deployment process infringes protected rights, subject to the applicable law and facts.

The decision concerns copyright rather than physical injury or general AGI safety, but it is important when considering civil responsibility across the AI development lifecycle.

Case 9: M.C. Mehta v. Union of India (Oleum Gas Leak case, 1987)

Citation: (1987) 1 SCC 395, Supreme Court of India.

Facts: A hazardous gas leak from an industrial enterprise in Delhi caused public concern and exposed people to risk. The litigation raised questions about the responsibility of enterprises engaged in inherently hazardous activities.

Legal issue: Whether Indian law should impose a heightened form of liability on enterprises engaged in hazardous or inherently dangerous activities.

Decision and principle: The Supreme Court articulated the doctrine of absolute liability for enterprises engaged in hazardous or inherently dangerous activities, with a standard more stringent than the traditional rule in Rylands v. Fletcher.

Relevance to AGI: This decision may inform academic proposals for special liability regimes covering exceptionally dangerous autonomous systems. If an AGI system were deployed in an activity legally classified as hazardous, questions could arise about whether existing statutory or doctrinal rules apply.

However, the case does not establish that all AI developers or AGI operators are absolutely liable. Its application would require careful analysis of the relevant doctrine, the nature of the activity, and the legal basis of the claim.

6. Comparative analysis of the principal liability models

The cases illustrate that AGI liability is likely to develop through a combination of established legal doctrines rather than a single universal rule.

No single model is sufficient for every situation. A defective autonomous vehicle may raise product-liability and negligence issues, whereas discriminatory AGI-assisted recruitment may primarily raise employment and equality-law questions.

7. The principal legal challenges of AGI liability

A. The problem of causation

AGI-related harm may result from several interacting factors: model design, training data, third-party tools, user instructions, software updates, and organizational decisions.

A claimant must ordinarily establish a sufficient causal connection between the defendant's conduct or product and the damage. The precise test differs by jurisdiction.

Potential reforms include requiring the preservation of audit logs, improving access to technical evidence, and permitting carefully defined evidentiary presumptions where a defendant's breach has made causation unusually difficult to establish.

B. The black-box problem

A model's internal operation may be difficult to explain, even to its developers. This creates obstacles when a claimant must identify a specific defect or negligent act.

Possible safeguards include:

Mandatory records of material system decisions.

Independent technical audits for high-risk deployments.

Preservation of relevant model versions and system logs.

Court-supervised access to confidential technical evidence.

Proportionate disclosure obligations and protection of trade secrets.

These mechanisms should improve access to evidence without automatically presuming that every harmful output establishes negligence or a product defect.

C. Multiple defendants and responsibility allocation

A single AGI system may involve a model developer, hardware manufacturer, cloud provider, integrator, deployer, and end user.

Courts must distinguish each party's contribution. A developer may be responsible for a defective design, while a deployer may be responsible for unsafe permissions or failure to follow required safeguards.

Joint or several liability, contribution between defendants, and indemnity depend on the governing law. A future legislative framework could clarify how responsibility is distributed where several failures materially contribute to one injury.

D. Foreseeability and autonomous behavior

A central question is whether the relevant harm was reasonably foreseeable at the time of development or deployment.

The more capable and adaptable an AI system becomes, the harder it may be to identify every possible behavior in advance. Nevertheless, unforeseeability should not automatically excuse failures to test, monitor, or mitigate known categories of risk.

A capability-sensitive standard of care could consider:

The system's level of autonomy and permissions.

The severity of potential harm.

The availability of practical safeguards.

The defendant's knowledge and technical resources.

The system's intended environment.

The feasibility of human intervention.

E. The legal-personhood question

One proposal is to recognize advanced AI systems as separate legal persons. This approach raises significant difficulties concerning assets, insurance, responsibility, control, and the ability to satisfy judgments.

Granting AI a legal identity would not, by itself, ensure that injured people receive compensation. A system without sufficient assets could become a liability shield rather than an effective defendant.

For that reason, a more practical starting point is to retain human and corporate responsibility while considering specialized compensation mechanisms for defined high-risk applications.

8. A proposed comparative framework for AGI liability

A balanced framework could combine preventive duties, conventional civil remedies, and special mechanisms for unusually dangerous systems.

Layer 1: Preventive obligations

Require risk assessments, appropriate testing, cybersecurity, documentation, monitoring, and safeguards proportionate to the system's capabilities and potential harm.

Layer 2: Civil liability

Apply negligence, product liability, contract, consumer protection, privacy, discrimination, and intellectual-property law according to the facts and jurisdiction.

Layer 3: Evidentiary safeguards

Provide appropriate access to technical records, expert evidence, and audit trails, with proportionate presumptions where legislatively justified.

Layer 4: Compensation and risk financing

Consider mandatory insurance, financial-responsibility requirements, or compensation funds for narrowly defined high-risk activities where ordinary remedies are inadequate.

This is a proposed policy framework, not a statement that all four layers are currently mandatory in any one jurisdiction.

9. Recommendations for civil-law reform

Comparative AGI liability frameworks could be strengthened through the following reforms.

First, establish clear duties for each participant. Legislation should distinguish the obligations of developers, manufacturers, professional users, commercial deployers, and operators.

Second, preserve access to compensation. Victims should not be required to prove technically inaccessible facts when a carefully designed statutory evidentiary mechanism could fairly address the imbalance.

Third, tailor liability to risk. Ordinary low-risk AI applications should not automatically face the same insurance and liability requirements as systems capable of causing catastrophic physical or economic harm.

Fourth, regulate the full system lifecycle. Duties should address development, testing, deployment, modification, monitoring, incident reporting, and retirement.

Fifth, preserve judicial independence. Courts should be able to examine causation, evidence, procedural fairness, and applicable defenses without being bound by a provider's own assessment that a system is safe.

Sixth, coordinate internationally. Because AGI systems and their supply chains cross borders, states should work toward compatible rules on jurisdiction, evidence preservation, liability allocation, and recognition of judgments.

Seventh, avoid automatic immunity or automatic liability. The mere use of AI should neither excuse a defendant from ordinary legal duties nor establish liability without satisfying the applicable legal requirements.

10. Conclusion

Civil law and comparative AGI liability frameworks concern one of the most significant prospective challenges for modern private law: how to allocate responsibility when increasingly capable autonomous systems cause harm.

The nine cases discussed above provide relevant foundations. Donoghue v. Stevenson and MacPherson v. Buick establish important principles of manufacturer responsibility; Escola and Greenman inform product-liability analysis; State v. Loomis addresses algorithmic decision-making safeguards; Mata v. Avianca illustrates continuing human professional responsibility; Mobley v. Workday examines potential liability for AI-assisted employment screening; Thomson Reuters v. Ross Intelligence addresses intellectual-property obligations; and M.C. Mehta v. Union of India offers a significant Indian analogy concerning hazardous activities.

These authorities do not establish a complete AGI-specific liability doctrine. Instead, they demonstrate how existing principles may apply to emerging technologies and where new legislation could be necessary.

The strongest approach is likely to combine risk-sensitive preventive duties, clearly allocated responsibility, effective access to evidence, proportionate liability rules, and reliable compensation mechanisms. Such a framework would seek to protect injured persons without imposing indiscriminate liability on every participant in the development or use of advanced AI.

Academic note: This is a comparative legal overview, not jurisdiction-specific legal advice. The cases should be checked against their full judgments and subsequent treatment before being relied upon in formal proceedings.