The model confidently stating things that aren't true — inventing a citation, a person, a number. Not lying (it has no intent); it's producing plausible-sounding text, and plausible isn't the same as true.
Why it matters: the single biggest risk in putting AI in front of customers. A confident wrong answer is far more dangerous than an honest "I don't know" — you cannot take an AI feature to real users until you know how often, and how loudly, it's wrong.