Hallucination

When an AI language model generates plausible-sounding but factually incorrect or entirely fabricated information — with no reliable signal to the user that it is wrong. The model does not 'know' it is hallucinating; it generates the most statistically likely next token given its training. Hallucinations are a core risk in any AI deployment involving factual claims, legal analysis, medical information, financial guidance, or customer-facing communications. Governance controls for hallucination typically include human review requirements, RAG-based grounding, and output verification steps.

A chatbot confidently cites a legal case that does not exist. This is a hallucination — the model generated a plausible-sounding citation with no factual basis.