CyberGuard dictionary / AI Security
Hallucination
An AI hallucination is a generated statement presented as if factual even though it is unsupported, incorrect, or invented.
In simple terms
A fluent answer can include a citation, number, or event that the model did not verify. Human review and source checking matter when an error has consequences.
Technical explanation
Language models predict plausible output from context; they do not automatically validate each factual claim. Retrieval, grounding, and evaluation can reduce errors but do not eliminate them, especially when source material is poor.
Fluent wording is not evidence that a claim is true. High-impact uses need source checks, uncertainty handling, and clear limits on actions based solely on generated text.
Examples in practice
A security assistant invents a vendor advisory number and recommends a patch that does not exist.
A security assistant cites a non-existent advisory. An analyst checks the vendor's official advisory list before scheduling a patch and flags the fabricated reference.
Why it matters
Unsupported claims can misdirect incident response, purchasing, or customer advice.
VERIFY & LEARN MORE
Sources and further reading
- airc.nist.govhttps://airc.nist.gov/airmf-resources/airmf/↗
- tsapps.nist.govhttps://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=958388↗
Published 2026-10-15 · Last updated 2026-10-15 · CyberGuard editorial content. Source links provide context; they do not imply endorsement.