Hallucination
An AI hallucination is output from an AI system, especially a generative AI model, that sounds plausible but is made up or not supported by its data or sources. Examples include invented facts, citations, or figures. Hallucinations become a problem when users are not trained to spot and correct them.
Generative models produce the most likely next words or pixels based on patterns in their training data. They do not check facts against a source by default, so when a question falls outside what they know, they can produce a confident answer that is wrong.
For example, a chat assistant asked about a machine's maintenance interval might invent a number that is not in the manual. Grounding the model in the actual manual with retrieval-augmented generation, and having a person review high-stakes answers, reduces that risk.
A hallucination is different from a simple error in a traditional model, such as a vision model misclassifying a part. Hallucinations are fabricated content presented as fact, which makes them harder to notice.