AI Hallucination
An AI hallucination is generated output that presents false or unsupported information as if it were accurate. Examples include invented references or claims not supported by the supplied material. NIST uses the term confabulation for related errors. The language does not imply that an AI has a human perceptual experience or deliberately intends to lie.
Term statusEmerging Concept
A term in research or professional use whose meaning still varies across fields. About these labels.
Evidence for the claims on this page: Technical risk framework
NIST describes confabulation as a risk. Error rates and definitions depend on the model, task, and evaluation method.
How to read this assessmentWhat it is
AI hallucination is a common term for generated content that is factually incorrect or unsupported by the relevant evidence or input. NIST uses confabulation for confidently presented erroneous or false content. (National Institute of Standards and Technology, 2024)
In everyday language
An AI system can produce a plausible answer that includes invented facts or references. Confident wording is not evidence that the answer was checked.
What it isn't
The term does not imply a human perceptual experience or deliberate lying. Requested fiction is also different from an invented claim presented as fact.
An example
Hypothetical example: a chatbot supplies a convincing quotation and a book page number, but checking the book shows that the quotation is absent.
Key features
- Concerns the content of a generated output
- Can include fabricated citations or details
- May require source checks to detect
How it relates to synthetic imprinting
An unsupported output could become information that someone later remembers or relies on. Whether that produces lasting change is a separate empirical question.
Research context
NIST's Generative AI Profile identifies confabulation as a risk. Definitions and evaluation methods differ, so a study should specify whether it tests factual accuracy, support from sources, or consistency with supplied material. (National Institute of Standards and Technology, 2024)
Related terms
Sources and what they support
- Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile — National Institute of Standards and Technology (2024). NIST AI 600-1.
Technical report. Describes generative-AI risks, including confabulation: confidently presented content that is erroneous, false, or inconsistent with relevant inputs.
Cite this definition
SyntheticImprinting.com Editorial Project. (2026-09-27). AI Hallucination. Version 1.4. SyntheticImprinting.com. https://syntheticimprinting.com/terms/ai-hallucination/
APA: SyntheticImprinting.com. (2026). AI Hallucination. https://syntheticimprinting.com/terms/ai-hallucination/
MLA: “AI Hallucination.” SyntheticImprinting.com, 2026, https://syntheticimprinting.com/terms/ai-hallucination/. Accessed November 12, 2026.
Chicago: SyntheticImprinting.com. “AI Hallucination.” Last modified 2026-09-27. https://syntheticimprinting.com/terms/ai-hallucination/.
Revision history
- Version 1.4 · Current
Added claim-level citations and a separate assessment of evidence support. The opening definition is unchanged.
Read this version’s definition
An AI hallucination is generated output that presents false or unsupported information as if it were accurate. Examples include invented references or claims not supported by the supplied material. NIST uses the term confabulation for related errors. The language does not imply that an AI has a human perceptual experience or deliberately intends to lie.
- Version 1.3
Expanded the opening definition; added explicit boundaries, source context, related-term cards, and linked editorial attribution.
Read the previous definition
An AI hallucination is generated output that presents false or unsupported information as if it were accurate. Examples include invented references or claims not supported by the supplied material. NIST uses the term confabulation for related errors. The language does not imply that an AI has a human perceptual experience or deliberately intends to lie.
- Version 1.0
Definition recorded in the previous release. Earlier edits are not reconstructed here.
Read the previous definition
An AI hallucination is generated output that presents false or unsupported information as if it were accurate.
What the status means
Emerging Concept
A term in research or professional use whose meaning still varies across fields.
How evidence labels workHelp improve this entry
Suggest a correction to AI Hallucination. Your email app opens with the page address and version included. Add the passage, your suggested change, and a source if you have one.