Glossary / AI & data

AI hallucination

A hallucination is an AI model stating something false with complete confidence — an invented statistic, a nonexistent court case, a plausible-sounding wrong answer. It happens because language models predict likely text, not verified truth. Production systems contain it with grounding (RAG with citations), constrained outputs, and human review where stakes are high; any vendor claiming zero hallucinations is themselves hallucinating.