A generative model can answer a bereavement-fare question, summarize a patient chart, or draft a court brief. In each case, its text reads as confident whether or not it’s correct. In regulated workflows, hallucinations are confident errors in places where accuracy carries legal, regulatory, or financial weight.