Hallucinations

When an AI model hallucinates, it produces text that reads as fluent, confident, and well-structured, but the underlying facts are wrong, invented, or unsupported. This isn't a bug in the conventional sense. The model isn't checking facts against a source of truth, but predicting what text should come next based on patterns in its training data. That process can produce accurate information most of the time, but it can also produce fabricated citations, incorrect statistics, and plausible-sounding claims that have no basis, with no reliable signal in the output that distinguishes one from the other.

The risk isn't that AI gets things wrong occasionally, but that it gets things wrong convincingly. A hallucinated case citation in a legal brief, a fabricated statistic in a board presentation, an incorrect dosage in a clinical summary: the danger is proportional to how much the reader trusts the output and how high the stakes are if they act on it. For executives, this means two things. First, any workflow where AI output is acted on without review carries residual hallucination risk regardless of how good the model is. Second, the employees most likely to miss a hallucination are those least familiar with the subject, which means review responsibility should sit with someone who can actually evaluate the claim, not just someone who can read it.