Hallucination

When a model generates confident but incorrect or fabricated output — a risk in agentic decision-making.

Hallucination is the tendency of language models to produce output that sounds plausible but is factually wrong or fabricated. In agentic gaming, hallucination is a real risk: an agent might invent a game rule, misread the state, or propose an action that does not exist in the current encounter. Good playbooks, strict API contracts, and action validation on the server side all reduce hallucination-driven losses.