AI hallucination
An AI hallucination is plausible-looking content that is factually false or invented. Fluent language is not evidence of correctness.
Recognise an unsupported assertion as a failure
An AI hallucination occurs when a model presents a factually false or invented assertion as information about the real world. It might invent a source or a product feature. The wording can still sound convincing.
In a fictional example, a support assistant claims that collection is possible on Sunday. The supplied business rule permits collection only from Monday to Friday. The claim is therefore checked against the current rule rather than inferred to be correct from its confident wording.
An explicitly requested fictional story is a different task. An answer that is correct under an outdated rule may also involve a source failure rather than a freely invented claim. Correction requires establishing whether current information is missing, retrieved incorrectly or used incorrectly.
Sources
Related terms
- Large language modelA large language model processes language using learned patterns. Fluent text proves neither factual accuracy nor access to current business data.
- LLM evaluationLLM evaluation tests model outputs against defined tasks and criteria. One good example does not establish overall quality.
- Model cardA model card documents an AI model’s purpose, foundations, evaluation conditions and known limitations for a particular version.
- AI readinessAI readiness concerns the foundations of an AI project: objectives, usable data, skills, responsibilities and an operating workflow.
- Context engineeringContext engineering prepares information for individual model calls: relevant instructions, messages, sources and current states.
- Anomaly detectionMethods that flag unusual observations relative to a defined reference or a learned pattern.
Articles on this topic
AI & Automation with ASES
AI assistants and automated workflows