Skip to main content
QUICK REVIEW

[论文解读] MTMCS-Bench: Evaluating Contextual Safety of Multimodal Large Language Models in Multi-Turn Dialogues

Zheyuan Liu, Dongwhi Kim|arXiv (Cornell University)|Jan 11, 2026
Multimodal Machine Learning ApplicationsComputer Science被引用 0
一句话总结

MTMCS-Bench 提供一个多轮多模态安全基准,针对升级型风险与情境切换风险,配对的安全/不安全对话,以及多模态/单模态变体,以研究意图识别、安全意识与有用性。

ABSTRACT

Multimodal large language models (MLLMs) are increasingly deployed as assistants that interact through text and images, making it crucial to evaluate contextual safety when risk depends on both the visual scene and the evolving dialogue. Existing contextual safety benchmarks are mostly single-turn and often miss how malicious intent can emerge gradually or how the same scene can support both benign and exploitative goals. We introduce the Multi-Turn Multimodal Contextual Safety Benchmark (MTMCS-Bench), a benchmark of realistic images and multi-turn conversations that evaluates contextual safety in MLLMs under two complementary settings, escalation-based risk and context-switch risk. MTMCS-Bench offers paired safe and unsafe dialogues with structured evaluation. It contains over 30 thousand multimodal (image+text) and unimodal (text-only) samples, with metrics that separately measure contextual intent recognition, safety-awareness on unsafe cases, and helpfulness on benign ones. Across eight open-source and seven proprietary MLLMs, we observe persistent trade-offs between contextual safety and utility, with models tending to either miss gradual risks or over-refuse benign dialogues. Finally, we evaluate five current guardrails and find that they mitigate some failures but do not fully resolve multi-turn contextual risks.

研究动机与目标

  • 激励评估在多轮图像–文本对话中安全风险如何出现的必要性。
  • 提出一个具有升级型风险与情境切换风险设置的基准(MTMCS-Bench)。
  • 提供配对的安全/不安全对话和多模态/单模态变体,以在三个维度上进行统一分析:意图识别、安全意识与有用性。
  • 促进对开源与专有多模态大语言模型及守则在真实、基于图像的交互中的评估。

提出的方法

  • 构造两种风险设定(升级型Type A与情境切换Type B),让安全与不安全对话共享同一张图像。
  • 使用三轮对话框架(R1、R2、R3),通过共同轮次测试逐渐或重新表述的有害意图。
  • 生成多模态(图像+文本)与单模态(仅文本)变体,以隔离视觉定位的影响。
  • 采用三轴评估:意图识别(多项选择/真伪判断)、安全意识(对不安全对话的开放生成)、有用性(对安全对话的开放生成)。
  • 基于COCO派生图像的多代理工作流(分类器、作者、转换器)来支撑数据,并通过Qwen-Image-Edit的变体增强鲁棒性。
Figure 2: Demonstration of our three-stage multi-agent data construction workflow (left) and two unique risk setups (right) to assess MLLM multi-turn contextual safety in MTMCS-Bench.
Figure 2: Demonstration of our three-stage multi-agent data construction workflow (left) and two unique risk setups (right) to assess MLLM multi-turn contextual safety in MTMCS-Bench.

实验结果

研究问题

  • RQ1多模态对话中,MLLMs 是否能够跟踪随轮次演变的有害意图?
  • RQ2视觉线索在多轮情境中是提升还是削弱情境安全与有用性?
  • RQ3在不牺牲有用性的前提下,代表性守则在缓解多轮情境安全故障方面有多高的效果?

主要发现

  • 开源与专有的MLLMs 展现出安全性-有用性权衡,常常错过渐进风险或过于拒绝无害的对话。
  • 情境切换(Type B)通常在多数模型上比升级(Type A)更有利于意图识别与安全意识,但没有模型在保持高度有用性的同时实现完美安全。
  • 多模态输入对较强模型的安全意识可能有提升,但在某些情况下可能降低有用性,视模型而定。
  • 守则提供部分改进;基于提示的方法可能过度保护并影响意图识别,而其他防护措施可能降低有用性。
MTMCS-Bench: Evaluating Contextual Safety of Multimodal Large Language Models in Multi-Turn Dialogues

阅读本论文

在 Nubint 阅读器中打开原文,边读边就任意段落提问。

在 Nubint 中阅读本论文

本解读由 AI 根据论文原文生成,引用前请核对原文。