What is it?
Hallucination
When a model states something that is false or made up but presents it confidently as fact. In agent systems this is dangerous because a hallucinated fact can get stored in memory and treated as truth later, or a hallucinated citation can make a report look credible while being wrong.
From the AI agent glossary
As the Planning and Reasoning glossary puts it
Hallucination is when a model states something false with confidence, essentially making it up. It happens because a model's core skill is producing plausible-sounding text, and plausible is not the same as true. If it does not know a fact, it may generate a convincing-looking invention rather than admit uncertainty.
Hallucination is a central motivation for tool use and grounding (Chapter 2): if the model can look a fact up rather than guess, it has far less need to invent one.
Where Hallucination is explained
- Skills, MCP, and Deterministic ComputationBuilding Coding Agents and Harnesses · Deterministic computation and the anti-hallucination contract
- How Memory FailsAgentic Memory · Validity failures: wrong or outdated facts
- Governing Evolving Memory (the SSGM Framework)Agentic Memory · The shift the paper is responding to
- Deep Research Agents and How They Are TrainedBuilding Coding Agents and Harnesses · Fact-checking: the research version of "do not trust yourself"
- ReAct, Reasoning and Acting TogetherPlanning and Reasoning · Two failure modes ReAct fixes