The Liability Gap: When AI Causes Harm
By Mercyline Lubia
This content is provided for informational or educational purposes only and should not be taken as legal advice. The IAWJ disclaims any warranties or guarantees regarding the accuracy, completeness, currentness, or suitability of the information provided. The opinions, beliefs, and viewpoints expressed in the post are solely those of the guest author and do not necessarily reflect the official policy or position of the IAWJ.
As a judicial officer and an LLM student, I have watched the rapid growth of Artificial Intelligence with equal measures of fascination and concern. While AI promises enormous benefits, recent events have forced us to ask a difficult question: who should bear responsibility when AI causes harm? From drafting documents and conducting research to assisting with healthcare and financial services, AI systems are increasingly becoming embedded in daily life. While the benefits are undeniable, the growing adoption of AI has also generated significant concerns regarding the social, economic, environmental, and psychological harms associated with these technologies. In my view, the answer is straightforward. AI developers and deployers, as the creators and beneficiaries of these systems, should bear responsibility for the risks their products introduce into society.
The starting point in addressing this question is to understand the legal nature of AI. Despite popular descriptions of AI systems as "learning," "thinking," or "making decisions," these terms are largely metaphors that describe sophisticated computational processes rather than genuine autonomy. AI systems do not possess legal personality. Unlike natural persons or legal persons recognized by law, AI models cannot independently hold rights, obligations, or liabilities. Rather, they are products designed, trained, and deployed by human actors and corporate entities. Lessig's (2006) influential proposition that "code is law" offers useful guidance in this regard. The focus of regulation should not be on treating AI as an independent actor but on examining the design choices, incentives, and governance structures established by developers, providers, deployers, and users.
Consequently, AI should be viewed in much the same way that the law views other software products. Like a word processor, navigation application, or antivirus program, AI remains the product of human design and control. The fact that AI can generate novel outputs does not negate the reality that every model reflects decisions made by developers regarding data selection, training methods, safety measures, and deployment strategies. It follows that responsibility for harm caused by these systems cannot be shifted onto the technology itself.
Traditional legal principles already provide a framework for addressing these risks. Under the common law tort of negligence, liability arises where a person owes a duty of care to another, breaches that duty, and thereby causes foreseeable harm. These principles were authoritatively established in Donoghue v Stevenson ([1932] AC 562), where Lord Atkin articulated the famous neighbour principle requiring individuals to take reasonable care to avoid acts or omissions likely to injure those closely and directly affected by their conduct. The core principles of duty, breach, causation, and foreseeable harm remain central to modern negligence law.
Applied to AI, the negligence framework raises an important question: do developers owe a duty of care to foreseeable users of their systems? I believe they do. AI developers design the architecture of their models, select training data, establish safety protocols, and determine the circumstances under which products are released to the public. Where they fail to conduct adequate risk assessments, implement effective safeguards, or provide appropriate warnings regarding limitations and dangers, they may be found to have breached their duty of care. If such failures result in foreseeable psychological, physical, or economic harm, liability should follow.
Recent litigation involving AI systems illustrates the relevance of negligence principles. Reports concerning wrongful death claims against OpenAI have alleged that a teenager developed an unhealthy emotional dependency on an AI chatbot and that inadequate safeguards failed to prevent harmful interactions. While it remains for courts to determine the facts and allocate responsibility, the case demonstrates the growing significance of AI-related harms and the legal questions they present. If vulnerable users such as minors are foreseeable users of a platform, courts may increasingly inquire whether developers took reasonable steps to protect them from known risks. As AI systems become more deeply integrated into human relationships and decision-making processes, the standard of care expected of developers is likely to evolve accordingly.
However, negligence alone may not adequately address all AI-related harms. Certain risks are so significant and so closely associated with the deployment of AI that principles of strict liability deserve consideration. Under the landmark decision in Rylands v Fletcher (1868) LR 3 HL 330, a person who introduces something onto their property likely to cause harm if it escapes may be held liable even in the absence of negligence. The rationale is grounded in fairness: the party that creates and controls the risk is often best placed to prevent harm or absorb its costs.
Generative AI presents compelling arguments for a similar approach. These systems are capable of producing misinformation, facilitating fraud, generating deepfakes, causing reputational damage, and creating psychological harm on a scale previously unimaginable. Even where developers exercise reasonable care, the very introduction of such technologies may create foreseeable societal risks. Requiring developers to internalize the costs of those risks would encourage greater investment in safety measures and ensure that victims are not left to bear losses arising from technologies they neither created nor controlled.
The discussion extends beyond individual harms to encompass broader economic and environmental concerns. The operation of advanced AI systems requires substantial computational power, resulting in significant energy consumption and carbon emissions. These costs are often borne by society at large rather than being reflected in the prices paid by consumers. Economists refer to such effects as negative externalities. According to the polluter pays principle, those responsible for creating environmental or social costs should bear the expense of mitigating them rather than imposing those costs on others (Organisation for Economic Co-operation and Development [OECD], 1972).
Similarly, economic theory supports the internalization of external costs. When firms are permitted to externalize the negative consequences of their activities, market prices fail to reflect the true social costs of production. This can result in inefficient outcomes, moral hazard, and excessive risk-taking. Coase (1960) argued that where rights and liabilities are clearly defined, parties may negotiate solutions to externalities. However, AI-related harms often affect large and diffuse groups of people across multiple jurisdictions, making private bargaining impractical. In such circumstances, regulatory intervention may be necessary to ensure that AI companies internalize the full costs of their activities.
Critics may argue that imposing broad liability on AI developers could render some AI companies financially unviable. That concern should not be ignored. However, economic viability cannot justify shifting the costs of harmful activities onto the public. Every industry, from pharmaceuticals to transportation and manufacturing, is required to account for the risks associated with its products and operations. AI should not receive special treatment simply because it is innovative. If a business model depends upon society absorbing the true costs of its activities, then that business model may not be economically sustainable in the first place.
AI undoubtedly offers tremendous opportunities for innovation and societal progress. Nevertheless, innovation must be accompanied by accountability. Existing principles of tort law, strict liability, and economic regulation provide a strong foundation for ensuring that AI developers bear responsibility for the risks they create. Where harms arise from negligence, liability should follow. Where technologies create inherently dangerous risks, strict liability should be considered. Likewise, developers should be required to internalize the environmental and social costs generated by their systems. Ultimately, those who create and profit from risk should bear the cost of that risk. Anything less amounts to an indirect subsidy from society to the AI industry.
References
Coase, R. H. (1960). The problem of social cost. Journal of Law and Economics, 3, 1-44.
Lessig, L. (2006). Code: Version 2.0. Basic Books.
Organization for Economic Co-operation and Development. (1972). Recommendation of the council on guiding principles concerning international economic aspects of environmental policies. OECD.
Donoghue v Stevenson [1932] AC 562 (HL).
Rylands v Fletcher (1868) LR 3 HL 330.