AI hallucination
An answer from an AI system that sounds convincing but is invented or wrong.
Glossary · Automation & AI
A hallucination is when a language model states a fact, quotation, reference or number that does not exist. The answer looks just as confident as a correct one, so the error cannot be spotted from the text alone.
It is not a fault that can be repaired but a consequence of how the model works: it produces plausible text, it does not check facts. The risk can be reduced, by giving the model sources, narrowing the task or requiring it to say what an answer is based on, but not removed.
So when introducing an AI solution it is worth deciding in advance what kind of mistake is acceptable and who checks the result. A summary for internal use and an answer to a customer about a price are tasks of very different risk.