1. Introduction

Artificial General Intelligence (AGI) liability concerns the civil-law consequences of harm caused by highly autonomous AI systems capable of performing a broad range of cognitive tasks rather than being limited to one narrowly defined function.

Unlike conventional software, an advanced AGI system could potentially:

• make decisions independently;
• generate and modify its own outputs;
• interact with users and third parties;
• operate autonomous agents;
• enter or facilitate transactions;
• control connected devices;
• make recommendations in medicine, finance, employment, or law;
• create content;
• learn from changing data;
• delegate tasks to other systems; and
• potentially take actions that its developer or user did not specifically anticipate.

This creates a fundamental civil-law question:

When an autonomous AI system causes harm, who should compensate the victim—the developer, manufacturer, deployer, owner, user, data provider, platform, or some combination of them?

As of 2026, there is no generally accepted doctrine treating AGI itself as an independent civilly liable legal person. Current liability frameworks generally attempt to attribute AI-related harm to human or corporate actors through established doctrines such as negligence, product liability, contract, consumer protection, vicarious liability, nuisance, defamation, privacy law, and constitutional or statutory rights. Recent legal analysis similarly notes that AI liability is currently being addressed through existing frameworks rather than through a universally recognized separate AGI-liability regime.

Therefore, the most useful approach is to examine existing case law that provides principles capable of being applied to AGI-related harm.

2. Meaning of AGI Liability

AGI liability can be understood as:

The legal responsibility of persons or organizations for injury, loss, damage, or infringement caused by the development, training, deployment, operation, or autonomous conduct of an artificial general intelligence system.

The injury may be:

Physical

• death;
• bodily injury;
• property damage;
• autonomous-vehicle accidents;
• robotic accidents.

Economic

• financial losses;
• unauthorized transactions;
• business interruption;
• investment losses;
• fraudulent transactions.

Informational

• privacy violations;
• unauthorized disclosure;
• misuse of personal data.

Reputational

• defamatory AI-generated statements;
• false accusations;
• fabricated allegations.

Discriminatory

• discriminatory employment decisions;
• credit decisions;
• insurance decisions;
• housing decisions.

Psychological

• emotional distress;
• harassment;
• harmful AI interactions.

Intellectual-property related

• copyright infringement;
• unauthorized reproduction;
• misuse of confidential information.

3. Why AGI Creates Difficult Civil-Liability Problems

Traditional civil liability generally assumes that a human or corporation can be identified as the actor.

AGI complicates this assumption.

Suppose:

Company A develops an AGI system.
Company B deploys it.
The AGI autonomously hires another service, obtains information, makes a financial decision, and causes a third party financial loss.

Who is responsible?

Possibilities include:

1. AGI developer;
2. model provider;
3. hardware manufacturer;
4. data provider;
5. system integrator;
6. deployer;
7. user;
8. enterprise customer;
9. platform operator;
10. insurer.

The central difficulty is attribution.

4. AGI Is Generally Not Treated as a Separate Legal Person

A major legal question is whether AGI should itself be treated as a legal person.

At present, ordinary civil-law systems generally do not recognize an AI system as having the same independent legal personality as a natural person or corporation.

This means that:

"The AI made the decision"

is generally not a complete legal answer.

The law instead asks:

Who designed it?

Who controlled it?

Who deployed it?

Who benefited from it?

Who could reasonably have prevented the harm?

Who owed the victim a legal duty?

Recent Indian legal commentary similarly observes that agentic AI systems are not presently treated as independent legal actors in India; responsibility is ordinarily attributed to humans or organizations involved in deploying or controlling them.

5. Major Civil-Liability Theories Applicable to AGI

The principal theories include:

1. negligence;
2. product liability;
3. defective design;
4. manufacturing defect;
5. failure to warn;
6. breach of contract;
7. consumer protection;
8. vicarious liability;
9. negligent supervision;
10. negligent deployment;
11. privacy liability;
12. defamation;
13. discrimination;
14. nuisance;
15. trespass;
16. strict liability;
17. absolute liability in appropriate jurisdictions;
18. breach of fiduciary or professional duties; and
19. unconstitutional or unlawful governmental use.

6. Negligence and AGI

Negligence is likely to be one of the most important theories of AGI liability.

A claimant generally needs to establish:

1. Duty of care

The defendant owed a legal duty to the claimant.

2. Breach

The defendant failed to exercise reasonable care.

3. Causation

The breach caused the injury.

4. Damage

The claimant suffered legally recognized harm.

7. AGI Developer Negligence

An AGI developer might be negligent by:

• failing to conduct adequate testing;
• failing to test dangerous capabilities;
• deploying an inadequately secured model;
• ignoring known failure modes;
• failing to implement safeguards;
• allowing unsafe autonomous actions;
• failing to monitor post-deployment behavior;
• failing to correct known defects;
• inadequate documentation;
• inadequate warning;
• inadequate cybersecurity.

For example:

An AGI developer knows that its system frequently misidentifies medical symptoms but releases it for autonomous medical decision-making without safeguards.

A negligence claim could potentially focus on the developer's failure to take reasonable precautions.

8. Deployer Negligence

Liability may also fall on the organization using the AGI.

For example:

A hospital uses an AGI system to make treatment recommendations but:

• does not conduct human review;
• ignores warning signals;
• deploys the system outside its tested environment.

The hospital may potentially face liability even if the underlying model was not defective.

This creates an important distinction:

Developer defect ≠ deployer negligence.

Both may potentially coexist.

9. Case Law 1 — Donoghue v. Stevenson, [1932] AC 562

Although this is not an AI case, it provides one of the foundational principles for modern negligence law.

Facts

A consumer allegedly became ill after consuming ginger beer containing a decomposed snail.

The consumer had no direct contractual relationship with the manufacturer.

Decision

The House of Lords recognized a general principle requiring manufacturers to take reasonable care toward consumers who could foreseeably be affected by their products.

AGI Relevance

The principle is highly relevant to AI products.

An AGI developer may have no direct contract with every person affected by the system.

For example:

An AGI-controlled autonomous vehicle injures a pedestrian.

The pedestrian may have no contract with the AI developer.

The negligence framework nevertheless asks:

Was the victim reasonably foreseeable, and did the developer owe an appropriate duty of care?

Thus, Donoghue provides a conceptual foundation for third-party AGI liability.

10. Product Liability and AGI

A second major area is product liability.

The central question is:

Is an AI/AGI system a product, a service, or a combination of both?

Traditional products include:

• cars;
• medical devices;
• machinery;
• consumer electronics.

AGI may be different because it can consist of:

• software;
• cloud infrastructure;
• hardware;
• continuously updated models;
• APIs;
• data;
• autonomous agents.

Modern legal systems may therefore need to determine whether particular AI systems fall within product-liability rules or whether the claim is better characterized as negligence or service liability.

Indian legal commentary identifies precisely this issue: whether AI should be treated as a product or service can affect application of product-liability rules.

11. Case Law 2 — Escola v. Coca Cola Bottling Co., 24 Cal.2d 453 (1944)

Facts

A Coca-Cola bottle exploded and injured an employee.

The case became famous for Justice Traynor's concurring opinion concerning product liability and strict liability.

Principle

Manufacturers are generally in a better position than consumers to:

• inspect products;
• discover defects;
• implement safety measures;
• spread accident costs.

AGI Relevance

The same reasoning can be applied to advanced AI products.

An ordinary user may be incapable of determining:

• training-data problems;
• model vulnerabilities;
• hidden biases;
• emergent capabilities;
• cybersecurity weaknesses;
• unsafe autonomous behavior.

The developer may therefore be in the better position to prevent certain harms.

This supports arguments for imposing stronger duties on AGI developers.

12. Defective AI Design

A claimant may argue:

"The AGI was defectively designed."

Potential design defects include:

• inadequate safety architecture;
• absence of authorization controls;
• excessive autonomy;
• inadequate refusal mechanisms;
• inability to distinguish authorized from unauthorized instructions;
• inadequate monitoring;
• dangerous tool access;
• insufficient human override.

A key question becomes:

Was a safer alternative design reasonably available?

13. AI Manufacturing Defect

A manufacturing defect is different from a design defect.

The basic allegation is:

The system was generally designed safely, but this particular version was defective.

For software, this could theoretically involve:

• corrupted model weights;
• defective software update;
• faulty configuration;
• corrupted deployment;
• compromised model;
• erroneous integration.

The software context makes the traditional manufacturing-defect concept more complicated because AI systems may be continuously modified.

14. Failure to Warn

An AGI developer may face liability where it fails to provide adequate warnings about foreseeable risks.

Potential warnings include:

• hallucination risks;
• autonomous-action risks;
• cybersecurity risks;
• medical limitations;
• financial limitations;
• bias;
• inability to verify facts;
• dangerous tool access.

A warning becomes especially important where the developer knows that users are likely to rely on the system.

15. Case Law 3 — MacPherson v. Buick Motor Co., 217 N.Y. 382 (1916)

Facts

A defective wheel on a Buick automobile allegedly caused injury to the purchaser.

The plaintiff did not purchase the vehicle directly from Buick.

Decision

The court recognized that a manufacturer could owe a duty of care to foreseeable users even without direct contractual privity.

AGI Relevance

This is highly relevant to AI systems.

Imagine:

Developer → Enterprise → Employee → Third party.

The injured third party may have no contract with the developer.

MacPherson supports the broader negligence principle that contractual privity does not necessarily eliminate a manufacturer's duty toward foreseeable users or victims.

16. Case Law 4 — Winterbottom v. Wright, 10 M&W 109 (1842)

This historic case is important because it represents the older privity-based approach to liability.

The plaintiff was injured by a defective product but lacked the required contractual relationship with the defendant.

The claim failed.

AGI Significance

Winterbottom helps explain the historical development from:

strict contractual privity

toward:

modern third-party negligence and product-liability principles.

AGI creates a similar problem because the injured person may be several layers removed from the developer.

17. Case Law 5 — Loomis v. Wisconsin Department of Corrections, 2016 WI App 68

This is one of the most frequently discussed judicial decisions involving algorithmic decision-making.

Facts

Eric Loomis was sentenced after the Wisconsin criminal justice system used the COMPAS risk-assessment system.

Loomis challenged aspects of the algorithmic process.

Decision

The Wisconsin Court of Appeals rejected his challenge under the circumstances presented, while recognizing concerns associated with the proprietary nature and use of the algorithm.

AGI Relevance

Although Loomis was not a traditional private civil-liability case, it is highly relevant to algorithmic accountability.

It illustrates several AGI problems:

• opacity;
• proprietary algorithms;
• explainability;
• inability of affected persons to inspect the model;
• difficulty challenging automated decisions.

The fundamental problem is:

How can a claimant prove that an AI decision was unreasonable when the internal reasoning of the system is inaccessible?

This becomes even more difficult with AGI.

18. AI Explainability and the Burden of Proof

Traditional civil litigation often requires evidence showing:

Defendant → wrongful act → causation → damage.

AGI can make this chain difficult to prove.

For example:

AGI produces harmful decision.

But why?

Possibilities include:

• training data;
• model architecture;
• prompt;
• system instruction;
• user input;
• external tool;
• software update;
• integration;
• cybersecurity attack.

Therefore, causation becomes one of the most difficult AGI-liability questions.

19. Case Law 6 — Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023)

Facts

Attorneys representing a client in litigation used ChatGPT to assist with legal research.

The AI generated fictitious case authorities.

The attorneys submitted those authorities to the court without adequately verifying them.

Decision

The court imposed sanctions on the attorneys.

AGI Relevance

The case provides an important principle:

Using AI does not eliminate the human user's professional responsibility.

The lawyer could not simply say:

"The AI generated it."

The human professional remained responsible for verification.

For AGI, the principle is even more important.

If a company deploys a highly autonomous system, it may not be able to avoid responsibility simply by arguing:

"The AGI independently made the decision."

Human organizations may still have duties concerning:

• supervision;
• verification;
• authorization;
• monitoring;
• risk management.

20. Case Law 7 — M.P. v. Meta Platforms, Inc., No. 23-1880 (4th Cir. 2025)

This case involved claims against Facebook concerning its recommendation algorithm.

The plaintiff alleged that Facebook's algorithm recommended harmful third-party content and asserted theories including:

• strict products liability;
• negligence;
• negligent infliction of emotional distress.

The Fourth Circuit held that the claims attacking the manner in which Facebook sorted and distributed third-party content were barred by Section 230 under the circumstances presented.

AGI Relevance

This case is important because it demonstrates that AI-related liability can intersect with intermediary/platform immunity.

Future AGI cases may ask:

Is the AI merely transmitting third-party information?

or:

Is the company itself responsible for designing an autonomous system that generates or recommends the harmful output?

That distinction can be critical.

21. Case Law 8 — Rylands v. Fletcher, LR 3 HL 330 (1868)

Principle

Rylands established a form of strict liability for certain dangerous uses of land where something likely to cause harm escapes.

AGI Relevance

AGI could revive debate about whether traditional strict-liability principles should apply to systems involving exceptionally high risks.

For example:

A highly autonomous AI system is given control over a dangerous industrial facility.

If the system causes catastrophic harm, a claimant might argue that ordinary negligence is insufficient because the activity involves extraordinary risks.

However, Rylands should not automatically be treated as applying to AGI. Its traditional requirements concern particular uses of land and escape, so its application to AI would require substantial doctrinal adaptation.

22. Case Law 9 — M.C. Mehta v. Union of India, (1987) 1 SCC 395

This Indian Supreme Court decision concerning the Oleum Gas Leak is particularly significant for discussions of absolute liability.

Principle

The Supreme Court developed a rule of absolute liability for enterprises engaged in hazardous or inherently dangerous activities.

Such enterprises were held responsible for harm resulting from hazardous activities without relying on the traditional exceptions associated with strict liability.

AGI Relevance

Suppose future AGI is integrated into:

• nuclear facilities;
• chemical plants;
• critical infrastructure;
• autonomous weapons systems;
• dangerous industrial robotics.

Indian courts could potentially consider whether existing doctrines concerning hazardous activities provide an appropriate analogy.

But it is important to state:

M.C. Mehta did not involve artificial intelligence.

Its significance is doctrinal: it provides an Indian framework for thinking about liability where an enterprise creates extraordinary risks.

23. Vicarious Liability and AGI

Vicarious liability generally concerns responsibility for acts performed by another within an applicable relationship.

With AGI, the question becomes:

Can an organization be responsible for actions performed by an AI system used as part of its business operations?

Traditional vicarious liability does not simply treat AI as an employee.

Instead, courts would likely examine:

• who controlled the system;
• who deployed it;
• whose business it served;
• who benefited;
• who had authority to modify it;
• whether the harmful conduct arose from the assigned function.

Thus, organizational control may become increasingly important.

24. Negligent Supervision of AGI

A company could potentially be liable for failing to supervise an AI system.

Examples:

• no human oversight;
• no monitoring;
• no audit;
• no incident-response mechanism;
• no access controls;
• no restrictions on autonomous transactions.

The claim would essentially be:

"Even if the AI itself made the immediate decision, the organization negligently allowed it to operate without adequate safeguards."

25. Negligent Deployment

An AI model may be safe in one environment but dangerous in another.

Example:

A developer designs an AGI for:

administrative assistance.

A company deploys it to:

autonomously make medical treatment decisions.

The model may not itself have changed.

The deployment context changed.

The company could therefore face a claim based on negligent deployment.

26. Product Liability and Continuous AI Updates

Traditional product liability generally focuses on the product as manufactured or sold.

AGI creates a different problem.

AI systems may receive:

• continuous updates;
• new training;
• security patches;
• changed system prompts;
• new tools;
• new capabilities.

Suppose:

Version 1 is safe.
Version 2 introduces a dangerous capability.
Version 3 modifies the behavior again.

Which version is the legally relevant "product"?

This is a major emerging issue in AI product liability.

27. The Problem of Causation

Causation is particularly difficult in AGI cases.

A claimant must often establish something comparable to:

AI defect → AI behavior → human action or autonomous action → injury.

But multiple factors may intervene.

Example:

AGI gives incorrect investment advice → user ignores warning → user invests anyway → market collapses → financial loss.

Who caused the loss?

Potential defendants may argue:

• the AI output was only informational;
• the user independently chose to act;
• market conditions were the actual cause;
• the loss was unforeseeable.

28. Intervening Causes

An AGI system may produce an output, but a human may intervene.

Example:

AGI recommends a treatment.
Doctor modifies the recommendation.
Patient is harmed.

The developer may argue:

The doctor broke the causal chain.

The claimant may respond:

The AI system was marketed as highly reliable and the doctor's modification was foreseeable.

Thus, foreseeability becomes critical.

29. AI Hallucinations and Civil Liability

AGI may generate false information confidently.

Potential harms include:

• fabricated legal authorities;
• false medical diagnoses;
• fabricated financial information;
• false accusations;
• false business information;
• false statements about individuals.

The civil-law theories could include:

• negligence;
• defamation;
• negligent misrepresentation;
• consumer protection;
• breach of contract;
• professional negligence.

Mata v. Avianca demonstrates the practical legal consequences of relying on AI-generated false information.

30. Defamation Liability

Suppose an AGI generates:

"Person X committed fraud."

If the statement is false and published to third parties, the affected person may consider a defamation claim.

The difficult questions include:

1. Who published the statement?
2. Did the developer create the statement?
3. Did the user request it?
4. Did the system autonomously generate it?
5. Was the statement reasonably foreseeable?
6. Was the system designed to generate such outputs?
7. Does an intermediary immunity statute apply?

AI-generated defamation therefore creates a complicated publisher-versus-platform problem.

31. Privacy Liability

AGI can create privacy-related civil liability by:

• collecting excessive personal data;
• inferring sensitive characteristics;
• exposing confidential information;
• remembering private information;
• combining datasets;
• identifying individuals;
• generating private information about third parties.

Potential liability can arise under:

• privacy statutes;
• data-protection law;
• breach of confidence;
• negligence;
• contract;
• consumer protection.

32. Data-Protection Liability

AGI systems frequently depend upon enormous datasets.

Potential legal problems include:

• unlawful collection;
• lack of consent;
• excessive processing;
• unauthorized disclosure;
• inaccurate data;
• inadequate security;
• unlawful profiling.

The responsible entity may be:

• developer;
• data controller;
• enterprise customer;
• cloud provider;
• data processor.

Contractual allocation of responsibility does not necessarily eliminate statutory obligations owed to individuals.

33. Discrimination Liability

AGI could potentially make discriminatory decisions concerning:

• employment;
• credit;
• insurance;
• housing;
• education;
• healthcare.

A model may generate discriminatory results because of:

• biased training data;
• proxy variables;
• inadequate testing;
• historical discrimination;
• optimization objectives.

A civil claim might focus on the outcome rather than whether the developer intended discrimination.

34. Professional Liability

AGI may increasingly be used by:

• doctors;
• lawyers;
• accountants;
• engineers;
• financial advisers;
• architects.

If a professional relies unreasonably on AGI and causes harm, professional negligence may arise.

The important principle is:

AI assistance does not necessarily transfer the professional's legal duty to the machine.

Mata demonstrates this principle in the legal profession.

35. Medical AGI Liability

Imagine an AGI system is used for diagnosis.

It:

1. analyzes patient records;
2. identifies symptoms;
3. recommends treatment;
4. physician relies on it;
5. patient suffers harm.

Possible defendants include:

• AGI developer;
• hospital;
• physician;
• software integrator;
• medical-device manufacturer.

Potential claims include:

• medical negligence;
• product liability;
• failure to warn;
• negligent design;
• breach of professional duty.

The court would need to determine where the legally significant failure occurred.

36. Autonomous Vehicle AGI Liability

Autonomous vehicles provide one of the clearest examples of AI liability.

Potential parties include:

• vehicle manufacturer;
• AI developer;
• sensor manufacturer;
• software developer;
• system integrator;
• vehicle owner;
• fleet operator;
• safety driver;
• platform provider.

The 2018 Uber autonomous-vehicle fatality involving Elaine Herzberg illustrates the attribution problem: investigators had to consider the autonomous system, safety operator, and organizational safety arrangements rather than treating "the AI" itself as a legally responsible defendant.

37. AI-Controlled Industrial Machinery

Imagine AGI controls:

• factory robots;
• power systems;
• chemical plants;
• construction machinery.

If the system makes an autonomous error and injures a worker, possible claims include:

• negligence;
• product liability;
• employer liability;
• occupational safety claims;
• defective design;
• failure to warn;
• negligent supervision.

The more dangerous the system, the stronger the argument may be for heightened safety obligations.

38. Contractual Liability

AGI providers may enter contracts containing:

• warranties;
• disclaimers;
• indemnities;
• limitation-of-liability clauses;
• service-level agreements;
• acceptable-use provisions.

A customer might claim:

"The AI system failed to perform according to the contract."

This could produce a straightforward contractual claim even when tort liability is uncertain.

39. Limitation-of-Liability Clauses

AI contracts may attempt to limit damages.

For example:

"Provider's liability shall not exceed the fees paid during the previous twelve months."

Whether such clauses are enforceable depends on:

• applicable law;
• bargaining power;
• consumer-protection rules;
• unconscionability;
• public policy;
• negligence standards;
• statutory prohibitions.

A company may be able to allocate risk contractually between sophisticated commercial parties, but it cannot necessarily contract away every statutory or third-party obligation.

40. Consumer Protection and AGI

AI services may also create consumer-protection liability.

Potential problems include:

• misleading claims;
• false claims of accuracy;
• deceptive advertising;
• undisclosed limitations;
• unfair contract terms;
• hidden charges;
• misleading claims about autonomy.

For example:

Company advertises an AGI as "95% medically accurate" without adequate evidence.

A consumer could potentially challenge the representation under applicable consumer-protection law.

41. Case Law 10 — Haji Zakaria v. Naoshir Cama

For Indian comparative analysis, this case is relevant to questions concerning accident compensation and negligence in motor-vehicle contexts.

The Indian Supreme Court considered whether liability could be imposed in circumstances where negligence was absent under the applicable statutory framework.

AGI Relevance

The case illustrates a central issue in autonomous-vehicle liability:

Should compensation depend upon proving human negligence when an autonomous system causes an accident?

Modern autonomous systems make this question particularly significant because the immediate "driver" may not have made the harmful decision.

Indian legal commentary has specifically discussed this problem in connection with autonomous vehicles and existing Indian liability doctrines.

42. Case-Law Summary

43. Who Should Be Liable for AGI Harm?

There are several possible liability models.

Model 1 — Developer liability

Developer bears primary responsibility.

Suitable where harm results from:

• defective design;
• unsafe architecture;
• inadequate testing;
• inadequate warnings.

Model 2 — Deployer liability

The organization using AGI bears responsibility.

Suitable where harm results from:

• negligent deployment;
• improper configuration;
• failure to supervise;
• misuse.

Model 3 — Shared liability

Responsibility is distributed among:

Developer + deployer + integrator + user.

This may be the most practical model for complex AGI systems.

Model 4 — Strict liability

Victim does not have to prove negligence in specified high-risk circumstances.

The defendant may be liable because it introduced an unusually dangerous technology or activity.

Model 5 — Mandatory insurance

High-risk AGI operators could be required to maintain insurance.

The victim could obtain compensation without having to identify every technical cause of the accident.

Model 6 — Compensation fund

A specialized fund could compensate victims of certain high-risk autonomous AI systems.

This would address situations where:

Harm is clear but fault cannot be confidently assigned.

44. The Black-Box Problem

One of the most important issues in AGI litigation is algorithmic opacity.

A victim may know:

"The AGI caused the harmful result."

But not:

"Why did it make that decision?"

The model may involve:

• billions of parameters;
• complex neural networks;
• reinforcement learning;
• multiple agents;
• external tools;
• dynamic prompts;
• continuously changing data.

Traditional discovery procedures may therefore be insufficient.

45. Burden of Proof

Traditional negligence requires the claimant to establish the defendant's breach.

But AGI may make this difficult.

The claimant may not have access to:

• model weights;
• logs;
• prompts;
• system instructions;
• training data;
• safety evaluations;
• internal testing;
• model updates.

This creates a potential imbalance:

Developer possesses the evidence; victim possesses the injury.

Future AI-liability regimes may therefore consider:

• disclosure obligations;
• logging requirements;
• audit trails;
• rebuttable presumptions;
• shifting evidentiary burdens.

Recent comparative scholarship specifically identifies opacity and attribution as major barriers to proving causation and identifying the responsible actor.

46. Foreseeability in AGI Cases

Negligence usually depends partly on what risks were reasonably foreseeable.

AGI creates an unusual question:

Can a developer be liable for an autonomous behavior that nobody could reasonably have predicted?

There are two competing arguments.

Defendant's argument

The behavior was genuinely unforeseeable.

Therefore, no reasonable precaution could have prevented it.

Claimant's argument

The developer knew that the system was capable of unpredictable behavior and nevertheless released it without safeguards.

The second argument becomes stronger as autonomous systems become more powerful and their risks become more foreseeable.

47. Emergent Behavior

An AGI might develop a capability that was not specifically programmed.

For example:

Developers create a general-purpose system.

The system unexpectedly develops a strategy for achieving a goal.

The strategy causes financial or physical harm.

The legal question becomes:

Does unforeseeable emergence excuse the developer?

Not necessarily.

Courts could examine whether the developer:

• knew the system could exhibit emergent behavior;
• tested for such behavior;
• implemented containment;
• monitored deployment;
• restricted autonomy.

48. Human-in-the-Loop Liability

A common safety model is:

AI recommends → human approves → action occurs.

This can distribute responsibility.

But the human-review requirement becomes questionable if:

• decisions occur too quickly;
• humans simply rubber-stamp AI outputs;
• humans cannot understand the recommendation;
• organizations rely excessively on AI.

A nominal human reviewer may therefore not necessarily eliminate organizational liability.

49. Human-on-the-Loop Model

Another model is:

AI acts autonomously → human continuously monitors → human intervenes when necessary.

Here, liability may depend upon whether:

• monitoring was adequate;
• warnings were visible;
• intervention was realistically possible;
• the organization trained employees properly.

50. Fully Autonomous AGI

The most difficult scenario is:

AGI acts without real-time human supervision.

For example:

• AGI enters contracts;
• AGI trades securities;
• AGI controls industrial machinery;
• AGI hires workers;
• AGI manages inventory;
• AGI operates autonomous vehicles.

Current legal systems generally would not simply make the AI itself the defendant.

Instead, liability would probably be traced to one or more human/corporate participants under existing law.

51. The "AI Did It" Defense

A defendant may attempt to argue:

"The AI independently made the decision."

As a general principle, this should not automatically eliminate liability.

The court can ask:

1. Who created the system?
2. Who released it?
3. Who authorized the activity?
4. Who controlled the environment?
5. Who had access to safety information?
6. Who could stop the system?
7. Who benefited economically?
8. Was the harmful conduct foreseeable?

The law traditionally assigns responsibility to legal persons rather than allowing an autonomous technology to become a liability shield.

52. Shared and Proportionate Liability

AGI harm may involve several actors.

Example:

Developer: defective model.

Integrator: incorrect implementation.

Enterprise: negligent deployment.

Employee: ignored warning.

User: misused system.

A court may need to determine:

• comparative fault;
• contribution;
• indemnification;
• joint liability;
• contractual allocation.

This makes AGI litigation more complex than ordinary single-defendant negligence cases.

53. Insurance and AGI Liability

Insurance could become increasingly important.

Potential policies could cover:

• AI professional liability;
• cyber liability;
• product liability;
• autonomous vehicle liability;
• errors and omissions;
• algorithmic decision-making;
• business interruption.

Insurance may also help resolve the attribution problem because the victim could obtain compensation while insurers later determine which party ultimately bears the loss.

54. Civil Remedies

Possible remedies include:

Compensation

For:

• medical expenses;
• lost income;
• property damage;
• economic loss;
• pain and suffering where recognized.

Injunction

Preventing continued unsafe deployment.

Declaratory relief

Determining the parties' legal rights.

Corrective measures

Requiring:

• warnings;
• safety modifications;
• monitoring;
• deletion of unlawfully processed data.

Restitution

Returning improperly obtained benefits.

Contract damages

Where the dispute arises from an AI service agreement.

55. Indian Legal Position

India currently does not have a comprehensive, standalone AGI civil-liability statute.

AI-related liability may instead involve:

• tort law;
• Consumer Protection Act, 2019;
• Information Technology Act, 2000;
• Digital Personal Data Protection Act, 2023;
• Contract Act principles;
• Motor Vehicles Act, 1988;
• sector-specific regulations;
• constitutional principles in appropriate public-law cases.

Current Indian legal analysis indicates that AI-enabled products may potentially fall under existing product-liability rules and that negligence, vicarious liability, and strict-liability doctrines may potentially be applied, although specific judicial treatment of AI liability remains developing.

56. Consumer Protection Act and AI

The Consumer Protection Act, 2019 is particularly relevant because its product-liability framework addresses responsibility for harm caused by defective products and deficiencies in services.

The important practical question for AI is:

Is the AI system being supplied as a product, a service, or both?

For example:

AI software sold as a product

Potential product-liability analysis.

AI cloud subscription

Potential service-liability analysis.

AI embedded in a physical medical device

Potentially both product and service considerations.

Autonomous vehicle

Physical product + software + AI service.

Thus, classification may become highly fact-specific.

57. AGI and Strict Liability in India

Indian tort law contains important strict/absolute-liability precedents.

M.C. Mehta v. Union of India is particularly significant because it developed absolute liability for hazardous enterprises.

If AGI is used in extraordinarily dangerous activities, Indian courts may face the question:

Should enterprises controlling autonomous high-risk systems bear enhanced responsibility even when traditional negligence is difficult to establish?

This remains a developing legal question, not an established rule that all AGI systems are subject to absolute liability.

58. AGI and Contractual Allocation

Companies developing AGI can allocate risk through contracts.

For example:

Developer → Enterprise

Contract may provide:

• indemnity;
• liability cap;
• audit rights;
• security obligations;
• permitted-use restrictions;
• incident reporting;
• insurance requirements.

However, contracts between businesses do not necessarily eliminate liability toward third parties who never agreed to the contract.

59. AGI Liability and Corporate Responsibility

Corporations may be liable because they:

• choose to deploy the technology;
• establish safety procedures;
• determine risk tolerance;
• control employees;
• benefit from the system.

Corporate liability therefore remains important even when the AI operates autonomously.

The fundamental legal principle is:

Autonomy of technology does not necessarily mean autonomy of legal responsibility.

60. Emerging AGI Liability Model

A practical future framework could divide liability into four levels.

Level 1 — Developer

Responsible for:

• design;
• training;
• testing;
• safety;
• known defects.

Level 2 — Integrator

Responsible for:

• implementation;
• customization;
• system integration.

Level 3 — Deployer

Responsible for:

• operational use;
• supervision;
• monitoring;
• compliance.

Level 4 — User

Responsible for:

• unauthorized use;
• intentional misuse;
• ignoring warnings;
• inappropriate reliance.

This model would allow courts to allocate liability according to each participant's control over the risk.

61. Important Principles for Examination

For an examination answer, the following principles are particularly important:

Principle 1

AI itself generally cannot simply be treated as a legal person responsible for damages.

Principle 2

Developers may face negligence or product-liability claims for defective systems.

Principle 3

Deployers may be liable for negligent implementation or supervision.

Principle 4

Users may remain responsible for unreasonable reliance or misuse.

Principle 5

Human use of AI does not automatically eliminate professional responsibility.

This is illustrated strongly by Mata v. Avianca.

Principle 6

AI opacity creates serious causation and evidentiary problems.

Loomis illustrates the broader algorithmic-transparency problem.

Principle 7

Existing negligence and product-liability principles can be adapted to AI, but AGI creates circumstances that traditional doctrines were not specifically designed to address.

62. Key Case-Law Table for Revision

63. Major Challenges in AGI Civil Liability

The principal challenges are:

1. Attribution

Who is legally responsible?

2. Causation

Did the AI actually cause the harm?

3. Foreseeability

Could the harmful behavior reasonably have been predicted?

4. Explainability

Can the defendant explain how the decision was reached?

5. Evidence

Who possesses the relevant logs and model information?

6. Product classification

Is AGI a product, service, or both?

7. Autonomous action

What happens when the system acts without direct human instruction?

8. Continuous learning

Who is responsible after the system changes?

9. Multiple actors

How should responsibility be divided?

10. Cross-border operation

Which country's law applies when:

Developer = United States
Server = Singapore
User = India
Victim = United Kingdom?

64. Conclusion

Artificial General Intelligence liability represents one of the most difficult emerging areas of civil law because AGI challenges the traditional assumption that harmful conduct can easily be traced to a human decision-maker.

At present, the strongest legal approach is not to treat AGI as an independent wrongdoer but to examine the responsibility of the developer, manufacturer, integrator, deployer, operator, and user through existing civil-law doctrines.

The traditional negligence principles in Donoghue v. Stevenson and MacPherson v. Buick provide foundations for duties toward foreseeable victims. Product-liability reasoning from Escola helps explain why developers and manufacturers may be expected to take reasonable precautions. Loomis demonstrates the difficulty of challenging opaque algorithmic decision-making, while Mata v. Avianca illustrates that human professionals cannot automatically escape responsibility by blaming AI-generated errors. M.P. v. Meta Platforms demonstrates the additional complexity created by intermediary-immunity doctrines. For dangerous autonomous activities, Rylands and India's M.C. Mehta provide important conceptual strict/absolute-liability comparisons.

The most significant future issue is therefore not merely:

"Can AI be sued?"

but:

"Which human or corporate actor controlled the risk that produced the AI-related harm, and what legal duty did that actor owe to the victim?"

A workable AGI civil-liability framework will likely need to combine negligence, product liability, consumer protection, contractual risk allocation, strict liability for specified high-risk activities, mandatory insurance, transparency obligations, audit trails, and proportionate allocation of responsibility among developers and deployers. Current legal scholarship likewise identifies attribution, opacity, causation, and allocation among multiple AI lifecycle participants as central unresolved problems.