GuardTrace-VL：通过迭代安全监督检测不安全的多模态推理
GuardTrace-VL: Detecting Unsafe Multimodel Reasoning via Iterative Safety Supervision
把多模态大模型的中间推理链（Thinking）纳入安全审核范围，用图文联合分析在推理阶段就拦截不安全内容；对做 VLM 安全对齐或红队评测的人，可直接复用其 QTA 全流程审计设定与三阶段渐进训练方案。
原文依据 monitors the full Question-Thinking-Answer (QTA) pipeline via joint image-text analysis
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论多模态大推理模型（MLRM）越来越多地被部署于需要产出显式中间推理的视觉语言任务。然而，即使最终答案无害，推理轨迹也可能包含不安全内容，从而带来部署风险。现有多模态安全防护主要只评估输入问题与最终答案，忽视了中间推理过程。这一疏漏使得偏见性推断或违反策略地使用视觉上下文等危害在推理过程中出现而未被发现。我们提出 GuardTrace-VL，一个具备视觉感知能力的安全审计器，它通过图文联合分析监控完整的「问题—思考—答案」（QTA）流程，从而能够在推理阶段及时发现不安全内容。为支持训练与评测，我们构建了 GuardTrace 数据集，该数据集通过多样化的提示策略生成，并经基于 MLRM 与人工的投票与验证流程精炼。此外，我们提出一种三阶段渐进式训练方案，并与数据精炼过程相结合，使模型能够依据不同风险等级学习细致且依赖上下文的安全偏好。在我们提出的涵盖域内与域外场景的测试集上，GuardTrace-VL 模型在不安全推理检测任务上取得 93.1% 的 F1 分数，相比此前最强的多模态安全防御方法，F1 分数提升 13.5%。代码见 https://github.com/xiangyx2020/GuardTrace-VL。
英文原文摘要 1548 字符
Multimodal large reasoning models (MLRMs) are increasingly deployed for vision-language tasks that produce explicit intermediate rationales. However, reasoning traces can contain unsafe content even when the final answer is non-harmful, creating deployment risks. Existing multimodal safety guards primarily evaluate only the input question and the final answer, neglecting the intermediate reasoning process. This oversight allows undetected harm, such as biased inferences or policy-violating use of visual context, to emerge during reasoning. We introduce GuardTrace-VL, a vision-aware safety auditor that monitors the full Question-Thinking-Answer (QTA) pipeline via joint image-text analysis, enabling detection of unsafe content as it emerges in the reasoning stage. To support training and evaluation, we construct the GuardTrace dataset, which is generated through diverse prompting strategies and refined via a MLRM- and human-based voting and verification pipeline. Furthermore, we propose a three-stage progressive training scheme combined with the data refinement process, enabling the model to learn nuanced and context-dependent safety preferences according to different risk levels. On our proposed test set covering both in-domain and out-of-domain scenarios, GuardTrace-VL model achieves an F1 score of 93.1% on unsafe reasoning detection tasks, representing a 13.5% improvement in F1 score compared to the previous strongest multimodal safety defense methods.The codes is available at https://github.com/xiangyx2020/GuardTrace-VL.