Talking AI — Hallucination: when AI answers wrong with confidence

A hallucination is a false answer an AI model delivers with the same confidence as a correct one: an invented reference, a plausible but wrong figure, a misquoted standard. OpenAI showed why in September 2025: models “guess” rather than admit uncertainty, because they are scored like a multiple-choice test where guessing pays and “I don't know” earns zero. So the question to ask a vendor is not “what is your accuracy rate?” but “what does the system do when it isn't sure: does it answer anyway, or does it abstain?”

Before deploying AI on any process that commits the company, require a demonstration of abstention behavior and measure the error rate on a real sample, reviewed by a domain expert, before allowing any output to leave without human validation.

Source: OpenAI, “Why Language Models Hallucinate,” September 2025.