Foundations

Hallucination

When an AI states something false with full confidence — invented facts, figures or sources that sound completely plausible.

Hallucination is the AI failure mode every executive must internalise: the model produces fluent, confident, wrong answers. It may invent a statistic, misattribute a quote, or cite a legal case that does not exist. It is not lying — it has no concept of truth, only of plausible language.

Executives should care because tone is no signal. A hallucinated answer reads exactly as confidently as a correct one, so intuition ('it sounded sure') fails completely. The risk concentrates in numbers, names, dates, citations and niche facts — precisely the details busy people skim. Organisations that skip verification eventually ship a confident error to a client, a regulator or the public.

Concrete example: lawyers in several jurisdictions have been sanctioned after filing briefs containing AI-invented case citations — each a professional who trusted fluency over verification.

The countermeasures are process, not technology: require sources for consequential claims, actually check a sample of them, prefer RAG setups that answer from your documents, and classify use cases so high-stakes outputs always get human review.