An AI hallucination happens when a model generates information that is not actually supported by the input, its tools, or reliable sources. The answer may look polished even when the facts are wrong.
In practice
Hallucinations happen because language models generate likely text, not guaranteed truth. They can also appear when a question asks for obscure facts, current events, exact numbers, private data, or sources the model cannot access.
What to watch
The safest pattern is to ground important answers in documents, databases, citations, tool outputs, or human approval. Confidence in tone is not proof of accuracy.