Saved in:
Bibliographic Details
Main Authors: Zheng, Baihui, Zheng, Boren, Cao, Kerui, Tan, Yingshui, Liu, Zhendong, Wang, Weixun, Liu, Jiaheng, Yang, Jian, Su, Wenbo, Zhu, Xiaoyong, Zheng, Bo, Zhang, Kaifu
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2505.19690
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918034757648384
author Zheng, Baihui
Zheng, Boren
Cao, Kerui
Tan, Yingshui
Liu, Zhendong
Wang, Weixun
Liu, Jiaheng
Yang, Jian
Su, Wenbo
Zhu, Xiaoyong
Zheng, Bo
Zhang, Kaifu
author_facet Zheng, Baihui
Zheng, Boren
Cao, Kerui
Tan, Yingshui
Liu, Zhendong
Wang, Weixun
Liu, Jiaheng
Yang, Jian
Su, Wenbo
Zhu, Xiaoyong
Zheng, Bo
Zhang, Kaifu
contents Despite the remarkable proficiency of \textit{Large Reasoning Models} (LRMs) in handling complex reasoning tasks, their reliability in safety-critical scenarios remains uncertain. Existing evaluations primarily assess response-level safety, neglecting a critical issue we identify as \textbf{\textit{Superficial Safety Alignment} (SSA)} -- a phenomenon where models produce superficially safe outputs while internal reasoning processes fail to genuinely detect and mitigate underlying risks, resulting in inconsistent safety behaviors across multiple sampling attempts. To systematically investigate SSA, we introduce \textbf{Beyond Safe Answers (BSA)} bench, a novel benchmark comprising 2,000 challenging instances organized into three distinct SSA scenario types and spanning nine risk categories, each meticulously annotated with risk rationales. Evaluations of 19 state-of-the-art LRMs demonstrate the difficulty of this benchmark, with top-performing models achieving only 38.0\% accuracy in correctly identifying risk rationales. We further explore the efficacy of safety rules, specialized fine-tuning on safety reasoning data, and diverse decoding strategies in mitigating SSA. Our work provides a comprehensive assessment tool for evaluating and improving safety reasoning fidelity in LRMs, advancing the development of genuinely risk-aware and reliably safe AI systems.
format Preprint
id arxiv_https___arxiv_org_abs_2505_19690
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Beyond Safe Answers: A Benchmark for Evaluating True Risk Awareness in Large Reasoning Models
Zheng, Baihui
Zheng, Boren
Cao, Kerui
Tan, Yingshui
Liu, Zhendong
Wang, Weixun
Liu, Jiaheng
Yang, Jian
Su, Wenbo
Zhu, Xiaoyong
Zheng, Bo
Zhang, Kaifu
Artificial Intelligence
Despite the remarkable proficiency of \textit{Large Reasoning Models} (LRMs) in handling complex reasoning tasks, their reliability in safety-critical scenarios remains uncertain. Existing evaluations primarily assess response-level safety, neglecting a critical issue we identify as \textbf{\textit{Superficial Safety Alignment} (SSA)} -- a phenomenon where models produce superficially safe outputs while internal reasoning processes fail to genuinely detect and mitigate underlying risks, resulting in inconsistent safety behaviors across multiple sampling attempts. To systematically investigate SSA, we introduce \textbf{Beyond Safe Answers (BSA)} bench, a novel benchmark comprising 2,000 challenging instances organized into three distinct SSA scenario types and spanning nine risk categories, each meticulously annotated with risk rationales. Evaluations of 19 state-of-the-art LRMs demonstrate the difficulty of this benchmark, with top-performing models achieving only 38.0\% accuracy in correctly identifying risk rationales. We further explore the efficacy of safety rules, specialized fine-tuning on safety reasoning data, and diverse decoding strategies in mitigating SSA. Our work provides a comprehensive assessment tool for evaluating and improving safety reasoning fidelity in LRMs, advancing the development of genuinely risk-aware and reliably safe AI systems.
title Beyond Safe Answers: A Benchmark for Evaluating True Risk Awareness in Large Reasoning Models
topic Artificial Intelligence
url https://arxiv.org/abs/2505.19690