When an AI model states something false or invented with the same confidence as a fact, often in fluent, plausible language.
Hallucinations occur when an LLM favors fluent, statistically likely text over factual accuracy. Enterprise decisions cannot rest on that kind of output, so mitigation combines grounding through RAG, Model Evaluation pipelines, and confidence thresholds that hold back low-confidence answers.
See also Private AI, Digital Transformation and Enterprise Architecture.