用能量驱动方法检测大模型响应中的指令冲突，提升安全防护能力。
Combating Instruction Conflict via Energy-Driven Latent Conflict Detection
- 通过拼接末词嵌入与平均池化表示，构建响应级潜在表征
- 在1.5B到14B参数的五款模型上，对齐度提升约30个百分点
- 适合自托管或开源大模型的安全审核场景
大型语言模型（LLMs）越来越多地采用分层指令，但仍易受用户指令覆盖系统约束的冲突影响。现有防御机制多依赖静态输入检查，难以发现响应漂移现象——即尽管输入看似合规，但最终输出仍违反系统约束。为此，本文提出ELCD，一种面向生成后、交付前的响应级潜在冲突检测器。给定完整生成结果，ELCD通过拼接末词嵌入与均值池化响应嵌入，构建复合隐状态表征，并优化成对边缘排序目标，使合规与漂移响应在潜在空间中分离。在涵盖1.5B至14B参数的五款主流大模型上进行的广泛实验表明，ELCD显著优于竞争基线。尤其在Llama-2-7B上，PR-AUC提升约30个百分点；在Mistral-7B上，95%真阳性率下的假阳性率（FPR95）降至2.67%。结果表明，ELCD为开放权重或自托管大模型部署中的潜在指令冲突检测提供了可行方案。
原文摘要 · Abstract (English)
Large Language Models (LLMs) are increasingly deployed with hierarchical instructions, yet they remain vulnerable to conflicts in which user directives override system-level constraints. Existing defense mechanisms predominantly focus on static input inspection and therefore fail to detect Response Drift, a phenomenon in which the model's final response violates system-level constraints despite seemingly compliant inputs. To bridge this gap, we introduce ELCD, a response-level latent conflict detector for post-generation, pre-delivery verification. Given the full generated output, ELCD constructs a composite hidden-state representation by concatenating the final-token embedding with the mean-pooled response embedding. It then optimizes a pairwise margin ranking objective to separate compliant and drifting responses in latent space. Extensive experiments across five mainstream LLMs ranging from 1.5B to 14B parameters demonstrate that ELCD significantly outperforms competitive baselines. Notably, it improves the PR-AUC on Llama-2-7B by approximately 30 percentage points and reduces the False Positive Rate at 95% TPR (FPR95) on Mistral-7B to 2.67%. These results suggest that ELCD provides a promising approach for latent instruction-conflict detection in open-weight or self-hosted LLM deployments.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。