AI Hallucinations

An AI hallucination is when a language model produces confident but false or fabricated information: an invented policy clause, a wrong leave balance, or a citation to a document that does not exist. It happens because the model generates what is statistically plausible rather than what is verified, so when it lacks the real answer it can still produce a fluent wrong one.

This is the number-one risk for an enterprise assistant: a confident wrong answer about leave, pay, or policy is worse than "I don't know." The controls are architectural rather than the model "trying harder": grounding (answer only from approved sources and say so when they do not cover the question), guardrails, low temperature for factual use cases, and a maker-checker design where a second model validates the first model's output against rules and thresholds before it reaches the user.