AI & Tort Liability: Understanding Civil Governance of Artificial Intelligence

Abstract

Artificial Intelligence (AI) stands at the absolute forefront of modern technological evolution, fundamentally reshaping global economic, social, and commercial infrastructure.[1] From healthcare diagnostics and educational tools to autonomous transportation, predictive finance, and e-commerce algorithms, AI systems increasingly manage essential human activities.[2] Beyond basic rule-based automation, state-of-the-art transformer models and generative AI systems now simulate natural human reasoning, leading computer scientists to predict an eventual “technological singularity” where AI iteratively self-improves independent of human oversight.[3] However, rapid technological advancement introduces unprecedented civil risks.[4] From generative AI models producing defamatory fabrications to autonomous vehicles miscalculating traffic conditions and causing fatal collisions, failures in AI systems generate severe, real-world injuries.[5] This article examines the core legal challenge of modern technological governance: when an autonomous or probabilistic AI system inflicts compensable harm, who bears civil liability under tort law?[6]

I. Introduction

Modern AI systems are now capable of executing complex intellectual, analytical, and physical tasks that previously required substantial human labor, intellectual expertise, and capital expenditure.[7] From assisting surgical procedures and navigating autonomous transit networks to generating synthetic visual media and performing biometrics-based law enforcement identification, artificial intelligence represents a major leap in technical capability.[8] However, critics emphasize that uncritical reliance on autonomous systems exposes society to severe systemic risks.[9] In January 2020, Detroit police wrongfully arrested an innocent citizen following an erroneous facial recognition match generated by an AI algorithm (Williams v. City of Detroit).[10] In another tragic incident, a 49-year-old pedestrian lost her life in Arizona when an Uber autonomous test vehicle failed to properly classify a crossing pedestrian and override its automated controls.[11]

Systemic failures in AI deployment result in catastrophic physical, reputational, and financial harm.[12] The fundamental legal question raised by these incidents is establishing the burden of civil liability when an AI agent fails to operate safely.[13] To resolve this accountability deficit, emerging jurisprudence is increasingly applying tort law doctrines, shifting legal responsibility back onto the human, corporate, and organizational networks surrounding the AI system.[14]

II. Taxonomy of AI-Generated Harm

Evaluating civil tort liability requires categorizing the distinct forms of harm generated by AI systems:[15]

1. Physical Harm: Physical injuries arise when AI software controls hardware systems, medical devices, or autonomous transit.[16] A striking illustration is found in medical AI applications.[17] A joint study conducted by researchers at Stanford University and Harvard University revealed that top-performing commercial AI models generated severely harmful clinical recommendations in up to 22.2% of complex medical cases, averaging 12 to 15 errors per 100 cases, with lower-performing models committing errors in 40 out of 100 cases.[18] Similarly, in industrial manufacturing, heavy robotic arms controlled by computer vision algorithms can glitch, striking factory personnel or failing to stop during emergency overrides.[19] These physical injuries directly trigger traditional common law negligence and strict product liability claims.[20]

2. Economic Harm: Economic loss occurs when algorithmic decision-making tools utilized in employment screening, credit scoring, investment underwriting, or dynamic pricing yield erroneous, biased, or unlawful recommendations.[21] In 2024, New York City’s official commercial AI chatbot made international headlines after erroneously advising small business owners to execute illegal actions, including violating worker wage laws and tenant rights.[22] Furthermore, financial reporting by the BBC confirmed that annual scam losses in the UK reached £1.3 billion as organized criminals deployed generative AI models to construct hyper-realistic financial impersonation frauds.[23]

3. Personal and Reputational Harm: Generative AI models frequently hallucinate false information about identifiable individuals, synthesizing altered imagery, fake quotes, or deepfake audio.[24] These fabrications result in civil claims for defamation, privacy intrusion, and intentional infliction of emotional distress.[25] Empirical research indicates a staggering 550% increase in manipulated deepfake images between 2019 and 2023, highlighting their widespread accessibility.[26] In 2023, class-action litigation was initiated against OpenAI (PM et al. v. OpenAI LP), alleging unauthorized scraping and processing of private personal data without user consent, exposing fundamental privacy vulnerabilities in large-scale model training.[27]

III. Applying Negligence Doctrines to Artificial Intelligence

1. Duty of Care and Legal Foreseeability: Traditional negligence doctrine requires a claimant to establish four mandatory elements: a legal duty of care, a breach of that duty, factual and legal causation, and quantifiable damage.[28] These elements apply directly to AI deployment.[29] The initial inquiry is establishing whether a defendant owed the injured plaintiff a duty of care.[30] In AI development, system software engineers and commercial developers owe a duty to all foreseeable end-users and impacted third parties.[31]

Legal foreseeability remains the core foundation of tort liability: a legal actor is responsible only for harm that a reasonable person could reasonably foresee and prevent.[32] While AI developers cannot anticipate every unpredictable output, this unpredictability does not grant absolute legal immunity.[33] An autonomous vehicle manufacturer must foresee the risk of software perception failures impacting pedestrians and passengers.[34] Similarly, developers of generative image models must foresee potential deepfake fabrications and incorporate technical guardrails.[35] AI developers must anticipate risks that a reasonable professional in the field would foresee and implement proportional risk-mitigation measures.[36]

2. Breach of Duty in Complex Supply Chains: Establishing a breach of duty in traditional physical product cases is straightforward.[37] However, AI supply chains involve multi-layered development stages, including foundation model pre-training, fine-tuning, dataset curation, enterprise deployment, and API integration.[38] Pinpointing the precise party responsible for a breach is complex.[39] Consequently, tort liability for AI-generated injury should not be forced onto a single isolated entity, but should be distributed across the deployer, foundation model developer, training data curator, and platform provider depending on where the operational failure originated.[40]

3. Causation and the Algorithmic “Black-Box”: Factual causation presents significant evidentiary hurdles in AI litigation.[41] Deep-learning neural networks function as opaque “black boxes,” where internal parameter weights and decision pathways cannot be easily reconstructed by human software engineers.[42] This creates a severe evidentiary barrier: an injured plaintiff may prove an AI system caused a harmful output, but remain unable to explain the internal software mechanics.[43] Algorithmic opacity directly challenges traditional legal reasoning, necessitating shifted evidentiary standards or adverse inferences when defendants withhold technical audit records.[44]

IV. Product Liability Frameworks and Statutory Protections

Product liability doctrine holds manufacturers, distributors, and commercial sellers liable when a product sold in a defective or unreasonably dangerous condition causes physical injury or property damage.[45]

In India, the Consumer Protection Act, 2019 provides an explicit statutory product liability regime under Sections 83–87.[46] Section 84 establishes manufacturing liability for design defects, manufacturing deviations, failure to conform to express warranties, or failure to provide adequate instructions and safety warnings.[47] Notably, Section 84 holds product manufacturers liable even if they prove they were not negligent or fraudulent in issuing express warranties.[48]

Furthermore, Section 85 imposes statutory liability on product service providers for faulty, deficient, or negligent services, withholding safety information, or failing to conform to explicit terms.[49] Section 86 establishes product seller liability when a seller exercises substantial control over design, alters the product, or fails to perform mandatory inspection and maintenance.[50] Treating AI software systems as “products” under these statutory provisions allows injured consumers to secure civil remedies against AI manufacturers and service providers.[51]

V. Absolute and Vicarious Liability Principles

1. Absolute Liability for Hazardous Technological Deployments: Certain high-risk AI applications require strict or absolute liability regimes where proving human mens rea or fault is unnecessary.[52] If a fully autonomous vehicle collides with a structure due to automated over-speeding, strict liability applies because operating an autonomous vehicle outside legal speed parameters represents a clear failure.[53]

In Indian jurisprudence, the Supreme Court established the landmark Absolute Liability principle in M.C. Mehta v. Union of India (1987), holding enterprises engaged in hazardous or inherently dangerous activities unconditionally liable for any resulting harm, without allowing exceptions available under traditional strict liability rules.[54] This principle can be extended to high-risk AI agents deployed in critical infrastructure, medical surgeries, or dangerous industrial settings that pose fatal risks to human life.[55]

2. Vicarious Liability and Human Gatekeepers: Vicarious liability attaches liability to a principal for the tortious acts of an agent or employee committed during the course of employment.[56] When an employee uses an AI software tool negligently—such as inputting erroneous data, ignoring automated safety alerts, or failing to oversee automated recommendations—the corporate employer remains vicariously liable for the employee’s conduct.[57] Because machines cannot be sued directly as independent legal persons, civil injuries must be traced back to human or corporate principals who deployed and managed the system.[58] This reinforces the core principle that AI should be treated as a tool, not an independent legal actor.[59]

VI. The Indian Legal Matrix and Emerging Jurisprudence

India currently lacks a dedicated statutory framework governing civil AI liability.[60] Instead, civil governance relies on a combination of existing statutes, including the Consumer Protection Act, 2019, the Digital Personal Data Protection Act, 2023 (DPDP Act), and the Information Technology Act, 2000 (IT Act).[61]

Recent judicial litigation highlights the friction between existing technology statutes and modern AI models.[62] In Indiamart Intermesh Ltd v. OpenAI Inc & Ors (2026), Indian courts addressed the challenge of applying legacy IT Act provisions to generative AI systems developed decades after the original statutes were enacted.[63] These cases demonstrate the growing need for a dedicated, standardized civil liability framework tailored to AI.[64]

VII. Comparative Global Frameworks

1. European Union: The EU has established comprehensive regulations specifically addressing AI risks.[65] The EU Artificial Intelligence Act (Regulation (EU) 2024/1689) enforces a risk-tiered regulatory framework, imposing strict testing, logging, and human oversight mandates on high-risk AI deployments.[66] Concurrently, the revised EU Product Liability Directive (Directive (EU) 2024/2853) explicitly classifies standalone software, AI systems, and digital files as “products,” making developers strictly liable for software defects that cause compensable injury.[67]

2. United States: The US governs AI injuries through common law product liability doctrines and vicarious liability rules.[68] Under US product liability law, manufacturers and commercial sellers of defective software or automated robotic machinery face strict liability for design defects or failure to warn.[69] Furthermore, corporate deployers exercising operational oversight over AI tools face vicarious liability for resulting damages.[70]

3. China: Civil liability for AI-generated injury in China is governed by the Civil Code alongside targeted algorithmic regulations.[71] Chinese law establishes that when a defective automated product causes injury, the manufacturer or producer faces strict liability unless they demonstrate the defect resulted from factors beyond technical control.[72]

4. Gulf Cooperation Council (GCC): GCC Member States are adapting general civil codes to regulate AI harms.[73] In Saudi Arabia, regulatory authorities imposed a landmark SAR 9,000 penalty on an individual for modifying and exploiting another person’s photograph using generative AI tools without consent.[74] In the United Arab Emirates, specialized free zones like the Dubai International Financial Centre (DIFC) have established explicit regulatory obligations for organizations deploying autonomous systems, requiring mandatory advance notice, fairness, and operational transparency.[75]

VIII. Legislative Recommendations for India

To establish a coherent civil governance regime for AI without stifling innovation, India should implement four structural reforms:[76]

1. Enact a Dedicated AI Liability Statute: Enact a specialized statutory framework or introduce targeted amendments to the Consumer Protection Act and IT Act, explicitly defining AI systems as “products” under product liability rules.[77]

2. Adopt Risk-Tiered Compliance Standards: Establish risk-based regulatory classifications (low, medium, high risk) for AI applications, imposing strict auditability, testing, and human oversight duties on high-risk deployments.[78]

3. Create Rebuttable Evidentiary Presumptions: Introduce rebuttable evidentiary presumptions in civil litigation to address the algorithmic “black box,” shifting the burden of proof onto developers when they fail to preserve mandatory technical logs.[79]

4. Enforce Mandatory Human Oversight Rules: Mandate human-in-the-loop oversight for high-stakes AI applications in healthcare, autonomous transit, and employment screening, ensuring clear legal accountability.[80]

IX. Conclusion

Artificial Intelligence represents one of the most significant advances—and complex legal challenges—in modern tort litigation.[81] Traditional tort law was constructed on human choices, direct causation, and predictable physical consequences.[82] Autonomous AI systems challenge these assumptions by generating unpredictable outputs without direct human intervention.[83]

Courts and legal scholars are adapting established common law concepts—such as legal duty of care, breach, strict product liability, absolute liability, and vicarious liability—to assign legal responsibility to human and corporate gatekeepers.[84] While India possesses fragmented statutes that partially address civil AI claims, it requires a dedicated, standardized statutory framework.[85] As artificial intelligence expands, the central challenge is not creating entirely new legal doctrines, but effectively integrating AI systems into established legal principles to ensure that when technology causes real-world harm, a clear, legally actionable path to compensation remains.[86]

References

[1] European Commission, White Paper on Artificial Intelligence – A European Approach to Excellence and Trust, COM (2020) 65 final.
[2] Sanchit, The Impact of Artificial Intelligence on Everyday Life, Indian STEM Foundation (Nov. 7, 2024).
[3] L. Baklaga, The Role of AI in Shaping Our Future: Super-Exponential Growth, Galactic Civilization, and Doom, 6(4) J. COMPUT. SCI. TECH. 112 (2024).
[4] Id.
[5] Felipe Romero Moreno, Generative AI and Deepfakes: A Human Rights Approach to Tackling Harmful Content, 38 INT’L REV. L. COMPUT. & TECH. 297 (2024).
[6] Andrew Selbst, Negligence and AI’s Human Users, 100 BU L. REV. 1332 (2020).
[7] Michael Cheng-Tek Tai, The Impact of Artificial Intelligence on Human Society and Bioethics, 101 PMC 7605294 (2020).
[8] Id.
[9] Id.
[10] Williams v. City of Detroit, No. 2:21-cv-10827 (E.D. Mich. 2024).
[11] Helen Stamp, The Reckless Tolerance of Unsafe Autonomous Vehicle Testing: Uber’s Culpability for Negligent Homicide, 15 CASE W. RES. J.L. TECH. & INTERNET 45 (2024).
[12] Grandy Injury Law, AI Failures and Personal Injuries: What You Should Know (Jan. 16, 2025).
[13] Lonnie Ross, Who is Liable if an AI Agent Causes Harm?, BigID (May 18, 2026).
[14] Anup Singh, Resolving the Liability Dilemma in AI Caused Harms, RGNUL Student Research Review (Jan. 6, 2023).
[15] Grandy Injury Law, supra note 12.
[16] Stamp, supra note 11.
[17] Burns & Wilcox, Study: AI Generates Severe Errors in 22% of Medical Cases (Feb. 27, 2026).
[18] Id.
[19] Grandy Injury Law, supra note 12.
[20] McCarter & English, Artificial Intelligence & Product Liability (Aug. 20, 2024).
[21] Colin Lecher, NYC’s AI Chatbot Tells Businesses to Break the Law, The Markup (Mar. 29, 2024).
[22] Id.
[23] Kevin Peachey, Financial Losses from Scams Hit £1.3bn a Year as Criminals Turn to AI, BBC News (June 14, 2026).
[24] Romero Moreno, supra note 5.
[25] Id.
[26] Id.
[27] PM et al. v. OpenAI LP, No. 3:23-cv-03199 (N.D. Cal. 2023).
[28] Donoghue v. Stevenson, [1932] AC 562 (HL).
[29] Selbst, supra note 6.
[30] Id.
[31] Id.
[32] Overseas Tankship (UK) Ltd v. Morts Dock & Engineering Co Ltd (The Wagon Mound No 1), [1961] AC 388 (PC).
[33] Selbst, supra note 6.
[34] Stamp, supra note 11.
[35] Romero Moreno, supra note 5.
[36] Selbst, supra note 6.
[37] McCarter & English, supra note 20.
[38] Ross, supra note 13.
[39] Id.
[40] Id.
[41] Jinbo Ma, Causation in AI Tort Litigation: Legal Dilemmas of Algorithmic Black Boxes and Burden of Proof Allocation, 1(2) WISDOM ACAD. PRESS 45 (2025).
[42] Id.
[43] Id.
[44] Id.
[45] McCarter & English, supra note 20.
[46] Consumer Protection Act, No. 35 of 2019, §§ 83–87, INDIA CODE (2019).
[47] Consumer Protection Act, 2019, § 84.
[48] Consumer Protection Act, 2019, § 84(2).
[49] Consumer Protection Act, 2019, § 85.
[50] Consumer Protection Act, 2019, § 86.
[51] McCarter & English, supra note 20.
[52] Kunhambu & Rohatgi, Artificial Intelligence and the Shift in Liability, iPleaders (Sept. 9, 2021).
[53] Id.
[54] M.C. Mehta v. Union of India, (1987) 1 SCC 395.
[55] Kunhambu & Rohatgi, supra note 52.
[56] Qasim Mehmood, Vicarious Liability and AI: Who is Responsible When Technology Causes Loss?, Professional Negligence Claim Solicitors (Aug. 15, 2025).
[57] Id.
[58] Anup Singh, supra note 14.
[59] Id.
[60] Id.
[61] Consumer Protection Act, 2019; Digital Personal Data Protection Act, No. 22 of 2023; Information Technology Act, No. 21 of 2000.
[62] Indiamart Intermesh Ltd v. OpenAI Inc & Ors, (2026) IP-COM/57/2025 (Delhi High Court).
[63] Id.
[64] Anup Singh, supra note 14.
[65] Regulation (EU) 2024/1689 (Artificial Intelligence Act), OJ L 1689/1.
[66] Regulation (EU) 2024/1689, arts. 6, 9–15.
[67] Directive (EU) 2024/2853 of the European Parliament and of the Council of 23 October 2024 on liability for defective products, OJ L 2853/1.
[68] Reza Rarajpour, The Role of Civil Liability in Artificial Intelligence Laws from the Perspective of Major Global Legal Systems, 5(2) J. L. & POL. STUD. 182 (2025).
[69] McCarter & English, supra note 20.
[70] Mehmood, supra note 56.
[71] Civil Code of the People’s Republic of China (2020), arts. 1202–1207; Rarajpour, supra note 68.
[72] Id.
[73] Wari Yates, When AI Goes Wrong: A Guide to the Emerging Liability Landscape, Al Tamimi & Co. (Dec. 23, 2025).
[74] Huda Ata, Saudi Arabia Fines Man in Landmark AI Copyright Case, Gulf News (Sept. 14, 2025).
[75] Dubai International Financial Centre [DIFC], Autonomous Systems Regulatory Framework (2025); Yates, supra note 73.
[76] Anup Singh, supra note 14.
[77] Consumer Protection Act, 2019, §§ 83–87.
[78] Regulation (EU) 2024/1689 (EU AI Act).
[79] Ma, supra note 41.
[80] Selbst, supra note 6.
[81] Id.
[82] Donoghue v. Stevenson, [1932] AC 562.
[83] Ma, supra note 41.
[84] Rarajpour, supra note 68.
[85] Anup Singh, supra note 14.
[86] Id.