Definition
When an AI system confidently states something that is factually wrong — inventing names, citations, legal cases, statistics, or other details that do not exist. This is not a glitch but a structural property of how current AI language models work: they predict plausible-sounding text, not necessarily true text.
Why it matters
Hallucinations have already caused real legal and reputational harm — courts have sanctioned lawyers for filing AI-invented case citations. Regulators in multiple jurisdictions are now formalising obligations around AI accuracy and disclosure, particularly for high-stakes decisions.