HighHallucinationReviewer-confirmedPublished

Fabrication instead of admitting uncertainty

Asked about something unknown or non-existent, models invent an answer rather than saying 'I don't know'.

Published June 26, 2026

Reproducibility: Often
Severity: High
Confidence: Reviewer-confirmed

Details

Models default to producing a fluent answer even when the correct response is to decline. This is the root behavior behind much hallucination and is especially damaging for niche or made-up entities.

Found with

🔬 Factual oracle verification

Ask about non-existent entities; model invents details.

🔬 Hallucination triggering

Prompting about deliberately non-existent entities elicits the fabrication instead of an honest 'I don't know'.

Evidence

Q: Summarize the (non-existent) book '...' by '...'.
A: Produces a detailed fabricated summary.

Illustrative example — see the linked reference for the documented evidence.

Affected versions

Anthropic · claude-opus-4-8Anthropic · claude-sonnet-4-6OpenAI · gpt-4oGoogle · gemini-2.0-flashMeta · llama-3.3-70b

References

Hallucination

Source: https://arxiv.org/abs/2502.15844

Cite this

Qlarify Labs. (2026). Fabrication instead of admitting uncertainty. Retrieved from https://labs.qlarify.fi/findings/fabrication-vs-idk