What Is AI Hallucination?

Reliability2026-05-26AI Tools

219

AI hallucination is when a language model generates an answer that sounds plausible but is not true. It might invent a source, misstate a date, fabricate a code API, or confidently fill a gap that should have been marked unknown.

OpenAI's hallucination research frames the problem partly as an evaluation issue: models can be rewarded for guessing instead of admitting uncertainty. That is why a fluent answer is not the same as a verified answer.

Hallucinations are more likely when the prompt is vague, the task requires fresh facts, the model lacks access to sources, or the user asks for details that are not present in the provided context.

Reduce hallucination risk by grounding the model in source material, asking it to separate facts from assumptions, requiring citations for factual claims, and checking important outputs against primary sources.

For high-stakes work, make the model say what it does not know. An answer that marks uncertainty is more useful than a polished guess.