arXiv cs.AI· Zhiwei Xu, Xuyang Liu, Xiaolin Peng, Weixian Deng, Jiabao Gao, Zirui Li, Tian Qiu, Boya Zhou, Wei Jiao, Hai Wan, Xibin Zhao·· 2 小时前AI 评分37
TrustJev：在 LLM 决策中将策略权威与不可信证据分离
TrustJev: Separating Policy Authority from Untrusted Evidence in LLM-Based Decisions
AI 导读
研究者提出 TrustJev 框架，通过受保护的策略打分与有界的证据修正，将策略权威与可能包含指令注入的外部证据分离。0.6B 实现对比 14 个基线与三个数据集，平均 Balance 超最强基线 0.57-1.05 个百分点；授权基准中对 95.77% 的来源正确应对全部 32 个攻击，对比方法为 80.98%。推理内存与中位延迟相当，额外开销主要在训练阶段。
正文 · 原文
Authors:Zhiwei Xu, Xuyang Liu, Xiaolin Peng, Weixian Deng, Jiabao Gao, Zirui Li, Tian Qiu, Boya Zhou, Wei Jiao, Hai Wan, Xibin Zhao
Abstract:Decision models enable software to select from predefined actions using natural language inputs. However, the external evidence that supplies facts for these decisions may also contain instructions that attempt to override the application's policy. We propose TrustJev, a framework combining protected policy scoring with bounded evidence corrections. A single Transformer backbone computes policy scores and a correction budget from trusted policy and candidates, then bounds evidence corrections by that budget. Paired supervision encourages stable predictions under instruction insertions that preserve task facts and changed preferences when edits change the correct answer. Authority supervision accommodates evidence when the policy alone cannot determine the decision. We evaluate a 0.6B implementation against 14 baselines on three datasets covering authentication, instruction hierarchy, and executable authorization rules. TrustJev achieves the highest mean Balance, the harmonic mean of accuracy on non-attack and attack inputs, exceeding the strongest baseline on each by 0.57-1.05 percentage points. In the authorization benchmark, it correctly answers all 32 attacks per source in a fixed pool for 95.77% of sources, compared with 80.98% for the strongest comparator. TrustJev maintains comparable inference memory and median latency, with additional overhead primarily during training.
| Comments: | 12 pages |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.34826 [cs.AI] |
| (or arXiv:2610.34826v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.34826 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Zhiwei Xu [view email]
[v1]
Mon, 26 Oct 2026 04:17:53 UTC (930 KB)
来源：arXiv cs.AI · arxiv.org