arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2510.04320cs.CLcs.LG
收藏

读懂场景而非脚本:面向大语言模型的结果感知安全性

Read the Scene, Not the Script: Outcome-Aware Safety for LLMs

Rutgers University(新泽西罗格斯大学)

Rui Wu, Yihao Quan, Zeru Shi, Zhenting Wang, Yanshu Li, Ruixiang Tang

AI总结:

本研究针对大语言模型存在的“后果盲”安全失效问题,构建了CB-Bench基准开展评估,并提出CS-Chain-4k后果推理数据集进行微调,有效提升了模型抗越狱能力、减少了过度拒绝,为安全对齐提供了新方向。

摘要:

经过安全对齐的大语言模型(LLM)仍存在两种主要失效模式:易被越狱,或是过度拒绝包含敏感表面信号的无害输入。我们发现二者的共同成因是:当前模型对行动与结果之间的关联推理能力较弱,且过度依赖表面形式信号——即不编码后果的词汇或风格线索。我们将这种失效模式定义为“后果盲(Consequence-blindness)”。\n为研究后果盲问题,我们构建了名为CB-Bench的基准,涵盖四种风险场景,区分语义风险与结果风险是否对齐的情况,可在现有安全基准常忽略的匹配与不匹配条件下开展评估。主流模型始终无法区分这两类风险,表现出后果盲特征,说明该问题普遍且具有系统性。\n为缓解后果盲,我们提出了CS-Chain-4k——一个用于安全对齐的后果推理数据集。在CS-Chain-4k上微调的模型在对抗语义伪装越狱方面表现出明显提升,同时减少了对无害输入的过度拒绝,且在其他基准上保持了实用性与泛化性。这些结果明确了当前对齐方法的局限,确立了后果感知推理作为核心对齐目标的地位,并提供了更实用、可复现的评估路径。

英文摘要:

Safety-aligned Large Language Models (LLMs) still show two dominant failure modes: they are easily jailbroken, or they over-refuse harmless inputs that contain sensitive surface signals. We trace both to a common cause: current models reason weakly about links between actions and outcomes and over-rely on surface-form signals, lexical or stylistic cues that do not encode consequences. We define this failure mode as Consequence-blindness. To study consequence-blindness, we build a benchmark named CB-Bench covering four risk scenarios that vary whether semantic risk aligns with outcome risk, enabling evaluation under both matched and mismatched conditions which are often ignored by existing safety benchmarks. Mainstream models consistently fail to separate these risks and exhibit consequence-blindness, indicating that consequence-blindness is widespread and systematic. To mitigate consequence-blindness, we introduce CS-Chain-4k, a consequence-reasoning dataset for safety alignment. Models fine-tuned on CS-Chain-4k show clear gains against semantic-camouflage jailbreaks and reduce over-refusal on harmless inputs, while maintaining utility and generalization on other benchmarks. These results clarify the limits of current alignment, establish consequence-aware reasoning as a core alignment goal and provide a more practical and reproducible evaluation path.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑