Risks · Established · Beginner

AI Hallucination

Also known as: Confabulation

When a generative model produces fluent output that is factually wrong or entirely invented while sounding confident.

What AI Hallucination is

Hallucination is a structural consequence of next-token prediction: the model optimises for plausible continuations, and plausibility is not truth. Invented citations, fake case law and imagined API methods are typical.

How it works

Mitigations include grounding answers in retrieved sources, requiring citations, constraining outputs to schemas, lowering temperature, using verification passes, and designing interfaces that make checking easy.

Why it matters

It is the primary blocker for deploying generative AI in medicine, law, finance and anywhere an error carries real cost.

Common uses

→Risk assessment for AI features
→Designing verification workflows
→Selecting grounding architecture

Watch for

✓Cannot currently be eliminated
✓Fluency makes errors hard to spot
✓Users over-trust confident output

Continue exploring

Retrieval-Augmented Generation Grounding AI Citations AI Guardrails AI Literacy