Definition
Hallucination occurs when an AI model generates information that sounds plausible but is factually incorrect, fabricated, or not grounded in the provided context. In AI agents, hallucination is particularly dangerous because agents can take real-world actions based on false information.
Example
An AI agent confidently provides a citation to a research paper that doesn't exist, or generates a plausible-looking API endpoint URL that returns a 404 error.
Why it matters
Hallucination is the single biggest risk in deploying AI agents. Techniques like RAG, grounding, and human-in-the-loop checkpoints help mitigate it. Stanford research found that hallucination rates range from 3-27% depending on the model and task.