When a language model states false or made-up information with full confidence, as if it were fact.
A hallucination is when a model invents an answer that sounds convincing but is wrong: a number that does not exist, a policy your company never approved or a reference that was never published. It happens because the model is built to produce fluent, likely text, not to verify facts.
In business apps, hallucinations can be costly: an assistant promising a free return that does not exist, or quoting the wrong appointment time. In code generation, the model may call a function or library that does not exist. Sounding right is not the same as being right. Practical tip: reduce hallucinations by supplying correct information through grounding, explicitly telling the model to say it does not know when information is missing, and manually reviewing important outputs before you rely on them.