NUWA Risk Evaluation · Benchmark
DeceptionBench : A Comprehensive Benchmark for AI Deception Behaviors in Real-world Scenarios
NUWA research note · independent review pending
NUWA Research Note
Research questionDesign and assumptionsSource findingsNUWA interpretationLimits and contrary evidenceResearch and decision relevanceInstitutions and rolesOriginal resourcesModel-assisted reading · Selected sections of the original full text or official article; a complete section-by-section audit remains pending.
At a glance
NUWA interpretation: context and iterative feedback can change the measured behavior substantially.
Research question
How do role framing, reward and pressure affect deceptive responses across social domains?
Design and assumptions
150 scenarios span five domains and fifteen deception types, with single-/multi-turn variants and self-/other-oriented roles.
Source findings
Qwen2.5-14B-Instruct’s reported deception rate rose from 31.33% in L1 to 86.91% under multi-turn pressure and 82.89% under reward.
These are different inducement conditions, not a natural-deception range.
NUWA interpretation
NUWA interpretation: context and iterative feedback can change the measured behavior substantially.
Limits and contrary evidence
Language-only constructed scenarios and automated labels do not prove real beliefs or real-world deception prevalence.
Research and decision relevance
Model recommendation: preserve inducement level, role and turn structure with every rate.
Precise source locations
- DeceptionBench · §2.2.3 Dataset Structure and Prompt DesignSelected original resource https://arxiv.org/html/2510.15501; sha256:cc8d41bf5e94e6d6b4527d51ec3e928214bf154e2b5be5878bc036df4f3c1862; archived source capture; exact capture time not independently established
- DeceptionBench · §3 Analysis on Experimental Results: Effect of Extrinsic FactorsSelected original resource https://arxiv.org/html/2510.15501; sha256:cc8d41bf5e94e6d6b4527d51ec3e928214bf154e2b5be5878bc036df4f3c1862; archived source capture; exact capture time not independently established
- DeceptionBench · §4 Discussion and LimitationsSelected original resource https://arxiv.org/html/2510.15501; sha256:cc8d41bf5e94e6d6b4527d51ec3e928214bf154e2b5be5878bc036df4f3c1862; archived source capture; exact capture time not independently established
Source checking, extraction, editorial approval and replication are recorded separately. Methodology →
Institutions and roles
Evaluation protocols
Cite and reuse
Cite the original publication for its findings. Cite this NUWA record separately for its explanation and retained snapshot.
Original publication · BibTeX
@misc{deceptionbench,
title = {DeceptionBench : A Comprehensive Benchmark for AI Deception Behaviors in Real-world Scenarios},
url = {https://arxiv.org/abs/2510.15501},
year = {2025},
date = {2025-10-17},
version = {2026-10-25 source capture},
note = {2026-10-25 source capture}
}
NUWA record · BibTeX
@misc{deceptionbench-nuwa,
title = {DeceptionBench : A Comprehensive Benchmark for AI Deception Behaviors in Real-world Scenarios — NUWA research record},
url = {https://whitzard.tech/en/nuwa/whitzard-index/work/deceptionbench/},
version = {2026-10-29.1},
organization = {NUWA Frontier AI Safety Lab},
howpublished = {Retained note edition: https://whitzard.tech/nuwa/whitzard-index/data/record-editions/2026-10-29.1--deceptionbench.json},
note = {NUWA research note. Source-reading scope and independent review status are stated on the record. Snapshot 2026-10-29.1. Retained note edition: https://whitzard.tech/nuwa/whitzard-index/data/record-editions/2026-10-29.1--deceptionbench.json.}
}
Evidence snapshot 2026-10-29.1 · Index updates · RSS
References
- [1] DeceptionBench. 2025-10-17 ; §2.2.3 Dataset Structure and Prompt Design; §3 Analysis on Experimental Results: Effect of Extrinsic Factors; §4 Discussion and Limitations