AI Hallucinations as a Form of Harm: An Analysis under WHO Guidance on Ethics and Governance of Artificial Intelligence for Health

Generative artificial intelligence systems, including large language models (LLMs), are increasingly used in health-related contexts. A documented feature of these systems is the production of hallucinations, outputs that are factually incorrect, fabricated, or misleading. While the World Health Organization (WHO) has described hallucinations as a "risk and challenge," this paper argues that under the WHO’s ethical framework, certain health-related hallucinations should be understood as forms of harm. Drawing on WHO Principle 5.2, requiring protection from mental and physical harm, this paper examines how health-related hallucinations generate psychological distress, enable harmful decision-making, and undermine patient trust. The reclassification of hallucinations from risk to harm carries significance for the regulatory obligations of AI developers, deployers, and policymakers.