arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2602.01425cs.AIcs.LG
收藏

一个探针无法捕捉所有:迈向有针对性的欺骗检测

One Probe Won't Catch Them All: Towards Targeted Deception Detection

LASR Labs(LASR实验室);UK AI Security Institute(英国人工智能安全研究所)

Vikram Natarajan, Devina Jain, Shivam Arora, Satvik Golechha, Joseph Bloom

AI总结:

针对线性探针在欺骗检测中的异质性,提出根据具体欺骗类型匹配探针可显著提升性能(AUC提升0.108),建议组织定义威胁模型并部署相应探针。

摘要:

线性探针是一种有前景的监测AI系统欺骗行为的方法。先前工作表明,在对比指令对和简单数据集上训练的线性分类器可以达到良好性能。然而,这些探针即使在简单场景中也表现出显著失败,包括虚假相关性和对非欺骗响应的误报。在本文中,我们证明欺骗检测本质上是异质的:虽然单个通用探针实现了适度的改进(+0.032 AUC),但事后最优分析显示,当探针与特定欺骗类型匹配时,潜力显著更高(+0.108 AUC),并且合成验证实验表明,当欺骗类型事先已知时,这一上限是先验可实现的。我们的发现表明,指令对捕捉的是欺骗意图而非内容特定模式,这解释了为什么提示选择主导探针性能(占70.6%的方差)。鉴于这种异质性,我们得出结论,组织应定义其特定威胁模型并部署适当匹配的探针,而不是寻求通用的欺骗检测器。

英文摘要:

Linear probes are a promising approach for monitoring AI systems for deceptive behaviour. Previous work has shown that a linear classifier trained on a contrastive instruction pair and a simple dataset can achieve good performance. However, these probes exhibit notable failures even in straightforward scenarios, including spurious correlations and false positives on non-deceptive responses. In this paper, we demonstrate that deception detection is inherently heterogeneous: while a single universal probe achieves modest improvements (+0.032 AUC), post-hoc oracle analysis reveals substantially higher potential (+0.108 AUC) when probes are matched to specific deception types, and synthetic validation experiments suggest this ceiling is achievable a priori when the deception type is known in advance. Our findings reveal that instruction pairs capture deceptive intent rather than content-specific patterns, explaining why prompt choice dominates probe performance (70.6% of variance). Given this heterogeneity, we conclude that organizations should define their specific threat models and deploy appropriately matched probes rather than seeking a universal deception detector.

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。

↑