In AI, hallucination is the generation of false, irrelevant, or nonsensical content by a model, often with high confidence. Hallucinations arise from model limitations, insufficient training data, or overgeneralization.