通过方向性嵌入噪声提升视觉语言模型抗越狱攻击能力
Directional Embedding Smoothing for Robust Vision Language Models
- 在嵌入层注入与原向量同方向的噪声，增强模型鲁棒性
- 在JailBreakV-28K数据集上将攻击成功率显著降低
- 轻量级推理时防御方案，适合集成到AI安全框架中
视觉语言模型（VLM）的安全性和可靠性是构建可信代理型AI系统的关键。然而，当前VLM仍易受越狱攻击影响，导致模型输出有害内容。本文将随机嵌入平滑与令牌聚合（RESTA）防御方法扩展至VLM，并在包含28,000条样本的多模态越狱攻击基准测试JailBreakV-28K上评估其性能。结果表明，采用方向性嵌入噪声（即噪声方向与原始嵌入向量一致）时，RESTA能有效降低多种攻击的成功概率。实验验证了该方法作为轻量级推理时防御层，在整体安全框架中可显著提升VLM的安全性。
原文摘要 · Abstract (English)
The safety and reliability of vision-language models (VLMs) are a crucial part of deploying trustworthy agentic AI systems. However, VLMs remain vulnerable to jailbreaking attacks that undermine their safety alignment to yield harmful outputs. In this work, we extend the Randomized Embedding Smoothing and Token Aggregation (RESTA) defense to VLMs and evaluate its performance against the JailBreakV-28K benchmark of multi-modal jailbreaking attacks. We find that RESTA is effective in reducing attack success rate over this diverse corpus of attacks, in particular, when employing directional embedding noise, where the injected noise is aligned with the original token embedding vectors. Our results demonstrate that RESTA can contribute to securing VLMs within agentic systems, as a lightweight, inference-time defense layer of an overall security framework.
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。