AI hallucination

A hallucination is an answer generated by a language model that looks perfectly credible yet is false: an invented price, an imagined availability date, a service area asserted with confidence. It isn't a bug in the strict sense: the model doesn't "know" anything, it predicts the most plausible continuation of a text — and plausibility isn't truth. The more pointed the question, the higher the risk.

In a business context, the stakes are frontal: an AI promising a 24-hour callout when you're booked for a week creates a furious customer, not a won one. The protections are well known and stack: a closed knowledge base bounding the possible answers, confidence thresholds below which the agent stops asserting, and human escalation on any sensitive topic. A front-desk AI without these guardrails isn't an assistant — it's a liability.