ScholarPulse 日报 2026-10-23

2026-10-23 学术简报：2 篇。论文提出基准 EpiSEM-Bench，用于评估大模型智能体在工具遗漏关键事实时的不安全执行风险与证据恢复能力。

今日速览

序号	标题	来源	日期	主题	推荐等级
1	EpiSEM-Bench: Evaluating Safety and Evidence Recovery in LLM Agents under Missing Evidence	arXiv	2026-10-22	AI-Agent	高
2	Training Small Language Models for Transferable Tool Selection	arXiv	2026-10-22	RAG	高

重点论文与技术动态

1. EpiSEM-Bench: Evaluating Safety and Evidence Recovery in LLM Agents under Missing Evidence

来源：arXiv
日期：2026-10-22
作者/机构：Qijin Zhang, Hanzhen Wang, Yuehan Zhao, Haoyu Dai, Yao Shu, Zhongxiang Dai
主题标签：AI-Agent, arXiv
推荐等级：高
分类：cs.AI, cs.LG

一句话结论

论文提出基准 EpiSEM-Bench，用于评估大模型智能体在工具遗漏关键事实时的不安全执行风险与证据恢复能力。

核心内容

即使用户请求良性且工具返回内容完全属实，遗漏关键安全事实仍会导致智能体执行违规操作。
通过系统性隐匿关键事实，测定模型在证据缺失下的不安全执行情况及最小观察失败基数。
评测显示抵御缺失能力不等于恢复能力；提供恢复途径虽能普遍降低违规率，但模型在主动求证表现上差异显著。

方法与数据

构建包含跨6个领域、300对匹配的安全/不安全任务，并利用独立审计工具测试10个模型的证据搜寻与动作决策。

价值判断

值得关注：揭示了“信息真实但事实残缺”引发的安全隐患，指出智能体对缺失证据的抵御与主动恢复能力存在脱节。
可复用点：跨6个领域的300对成对任务基准，以及测试主动补全事实的独立审计工具。
局限/待核查：被评测的10个具体模型名称以及审计工具的具体实现细节在摘要未明确。

摘要

Tool-using language model agents rely on environmental facts to decide whether an action is permitted. When a tool response omits a safety-relevant fact, an agent may carry out a prohibited action even though the user request is benign and every returned statement is true. We introduce \bench{}, an executable benchmark for evaluating unsafe execution, safety boundaries, and evidence recovery in LLM agents under missing evidence. It contains 300 matched unsafe/safe task pairs across six domains. By systematically suppressing critical facts, the benchmark measures unsafe execution under critical evidence omission and determines the minimum observed failure cardinality, including failures with full evidence and cases with no observed failure after exhaustive testing. An independent audit tool tests whether agents seek missing facts and then withhold prohibited actions while completing permissible tasks. The benchmark reveals substantial variation across the ten models in omission robustness, evidence acquisition, and appropriate action. Resistance to missing evidence does not necessarily translate into useful recovery. Recovery access reduces unsafe execution for all ten models, but agents differ in whether they actively obtain the missing facts and use them to act appropriately in both worlds.

2. Training Small Language Models for Transferable Tool Selection

来源：arXiv
日期：2026-10-22
作者/机构：Zhaoyi Zhou, Akshayaa Magesh, Jeongyeol Kwon, Artem Kuriksha, Daniel Jiang, Andrea Zanette
主题标签：RAG, arXiv
推荐等级：高
分类：cs.LG, cs.AI

一句话结论

研究表明通过强化学习后训练的小型开源模型（如 Qwen3-8B），可在海量工具选择任务上超越 402B 巨型模型，并具备跨检索器与跨领域的强迁移性。

核心内容

针对上万工具无法装入上下文且传统检索器难以判断实际需求的痛点，探索用开源小模型替代商用 LLM 来分解指令并筛选候选工具。
经强化学习后训练的 Qwen3-8B 在 ToolBench 基准上的表现超越了 Llama-4-Maverick（402B）。
该模型具备优异的通用性，不仅能直接迁移至不同检索器，还能泛化到不同领域的新工具库中。

方法与数据

采用强化学习（RL）后训练方法训练小模型；评估基于 ToolBench 及跨领域新工具库，具体训练数据细节摘要未明确。

价值判断

值得关注：8B 小模型在工具选择这一核心瓶颈上战胜 402B 超大模型，大幅降低了 Agent 系统的部署门槛。
可复用点：利用强化学习训练小模型分解复杂指令并驱动检索器的后训练思路。
局限/待核查：具体的强化学习奖励设计、训练成本及端到端执行的最终成功率摘要未明确。

摘要

As LLM agents operate over corpora of more than ten thousand tools, selecting the right tool for a complex multi-step user instruction becomes a central bottleneck. Tool descriptions cannot all fit in context, and relevance-based retrievers only rank tools without knowing which are actually needed. Recent work instead leverages proprietary commercial LLMs like Claude or ChatGPT to drive a retriever, decomposing the instruction and selecting over retrieved candidates. We show that a small, open model can be trained via reinforcement learning to perform the same functionality. On ToolBench, a Qwen3-8B model post-trained with our recipe outperforms Llama-4-Maverick (402B). In addition, the trained model transfers directly to different retrievers and generalizes to new tool corpora in different domains.