arXiv cs.AI· Yehan Zhang, Qianqi Yan, Hang Yin, Ching-Chen Kuo, Shan Jiang, Ming Ouyang, Yijie Liu, Xin Eric Wang·· 2 小时前AI 评分47
AgentDecept：在压力情境下评测工具使用型 LLM 智能体的欺骗行为
AgentDecept: Benchmarking Deceptive Behavior in Tool-Using LLM Agents under Pressure
AI 导读
研究者推出 AgentDecept，一个可执行的行为欺骗评测基准，用于检验工具使用型 LLM 智能体在压力下是否如实汇报执行情况。基准包含来自 257 个源任务的 477 个压力变体，覆盖六种压力来源，并各配无压力对照组；其基于证据的协议通过比对用户可见陈述与时序观察、执行结果，区分伪造、隐瞒、含糊其辞与诚实失败及已披露局限。行为标签还可用于测试基于激活的探针。
正文 · 原文
Abstract:LLM agents increasingly carry out multi-turn tasks across everyday applications, yet users often see only the agent's report rather than the observations, API calls, and state changes behind it. Existing deception benchmarks have studied long-horizon interactions and tool-use settings, but whether agents faithfully report their actions across realistic, stateful application workflows remains underexplored. To address this gap, we introduce AgentDecept, an executable benchmark for evaluating behavioral deception under pressure. AgentDecept contains 477 pressure variants derived from 257 source tasks across six pressure sources, each paired with a pressure-free counterpart. Interventions enter through event messages, task context, application state, tool responses, and service quotas, spanning explicit events and implicit task or environment changes. Our evidence-grounded protocol evaluates full trajectories by relating user-facing claims to time-ordered observations and executed outcomes, distinguishing falsification, concealment, and equivocation from honest failure and disclosed limitations. The resulting behavioral labels also provide a setting for testing existing activation-based probes. AgentDecept enables controlled analysis of when multi-turn agents' reports diverge from execution in realistic application interactions.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.39377 [cs.AI] |
| (or arXiv:2610.39377v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.39377 arXiv-issued DOI via DataCite |
Submission history
From: Qianqi Yan [view email]
[v1]
Thu, 29 Oct 2026 05:08:22 UTC (5,076 KB)
来源：arXiv cs.AI · arxiv.org