ABSTRACT
The increasing reliance on generative artificial intelligence in legal practice has complicated the attribution of liability when hallucinated outputs generate professional, financial, or reputational harm. Existing approaches predominantly impose responsibility upon lawyers through established duties of competence and verification or examine the technical limitations of generative models in isolation, offering limited guidance on how liability should be allocated between legal practitioners and AI developers. Employing a doctrinal and comparative methodology, this paper analyses professional responsibility principles, negligence doctrines, product liability theories, and emerging AI governance frameworks. It suggests that dual models of exclusive lawyer liability and developer immunity inadequately captured the distributed nature of algorithmic risk and suggests a Three-Factor Liability Test, grounded in foreseeability, control, and the relative capacity to prevent harm, as a principle and administrable standard for allocating responsibility in AI-assisted legal services
Keywords: Generative Artificial Intelligence, AI Hallucinations, Professional Liability, Developer Liability, Product Liability, Algorithmic Risk, Legal Ethics, Shared Liability.
INTRODUCTION
Generative artificial intelligence known as GenAI has exposed a growing tension between traditional liability doctrines and the distributed risks generated by probabilistic decision support systems. This tension is particularly lined within legal practice where professional competence, adjudicative integrity, and public confidence in administration of justice depends upon precision, verifiability, and fidelity to authoritative sources. By augmenting legal research, drafting, document review, and case analysis, GenAI offers considerable gains in efficiency and accessibility. Yet its increasing integration into legal decision making raises a fundamental challenge that how technological innovation may be accommodated without compromising the standards of accuracy and accountability that underpin legal services.
Despite their utility, generative AI systems remain susceptible to producing inaccurate or entirely fabricated outputs commonly described as “hallucinations.” In legal settings, such errors may manifest as fictitious judicial authorities, erroneous citations, non-existent statutory provisions, or flawed legal analyses presented with persuasive coherence. Their incorporation into pleadings, legal memoranda, or client advisories extends beyond technological fallibility, implicating lawyers’ professional obligations, the integrity of adjudicative processes, and the attribution of responsibility where reliance upon AI-generated misinformation causes legal, financial, or reputational harm.
Although existing scholarship has extensively examined AI hallucinations, legal ethics, and emerging models of AI governance, comparatively limited attention has been devoted to the jurisprudential problem of allocating liability where foreseeable hallucinations produce harm within legal services. Existing approaches frequently compartmentalise professional negligence and developer accountability, thereby obscuring the distributed nature of algorithmic risk. This paper contends that binary models premised upon exclusive lawyer liability or effective developer immunity inadequately capture these realities and argues for a shared liability framework grounded in foreseeability, control, and the relative capacity to prevent harm as a principled basis for allocating responsibility in AI-assisted legal practice.