arXiv cs.CL· Aditya Nawal, Manit Baser, Mohan Gurusamy·· 12 小时前AI 评分38
语言如何影响 LLM 智能体的工具准入与检索后安全
Disentangling Language Effects on Tool Admission and Post-Retrieval Safety
AI 导读
研究在受控 URL 检索设置下跨十个模型、七种语言独立改变请求语言、检索页面语言和应答语言，识别出跨语言处理导致的安全损失（SLIP）：非英语有害请求使平均工具准入率从 46.9% 升至 67.4%，且每个模型均呈正向变化。
正文 · 原文
Abstract:LLM agents increasingly use retrieval tools across languages, yet their safety evaluation remains English-centric. It remains unclear whether language affects tool admission, post-retrieval generation, or both. We independently vary request, retrieved-page, and required response languages in a controlled URL-retrieval setting across ten models and seven languages. We identify Safety Loss through Interlingual Processing (SLIP). Non-English harmful requests increase mean tool admission from 46.9% to 67.4%, with a positive mean change in every model. This pattern also persists in an autonomous retrieval setup across all four models tested in this setting. Requiring answers in the page language increases pooled harmfulness on matched tool calls from 50.3% to 57.4% across four closed-weight models, mainly in Chinese and Hindi. Translating pages alone produces no consistent harm increase in eight models with comparable inference settings. Automated audits show that a non-harmful response may be a genuine safety refusal, a deflection citing language difficulties, or an unusable answer. Harmfulness labels alone cannot distinguish these outcomes. Multilingual safety evaluations should therefore measure tool admission and response harmfulness separately and distinguish refusal from capability failure.
| Comments: | 33 pages, including appendix |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.23093 [cs.CL] |
| (or arXiv:2610.23093v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.23093 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Aditya Nawal [view email]
[v1]
Fri, 16 Oct 2026 08:05:14 UTC (907 KB)
来源：arXiv cs.CL · arxiv.org