Hallucination is the term for when an AI model generates information that sounds convincing and grammatically correct – but is simply not true. The model "makes up" a fact, citation, date, or name that doesn't exist.
Why does it happen? LLMs generate text statistically – predicting which word most likely follows the previous one. Sometimes this statistical logic produces grammatically correct but factually wrong statements.
Examples: the model cites an article that doesn't exist; claims a law has been in effect since year X when it's actually year Y; describes a product with features it doesn't have.
How to manage it in practice? Don't give AI agents tasks where they'd need to invent facts – connect them to real sources (RAG). For critical decisions, always include human review. Use models with lower hallucination rates (Claude consistently outperforms most competitors in benchmarks).
Hallucinations aren't a reason to avoid AI – they're a reason to use it correctly.