arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2510.02677cs.AIcs.LG
收藏

ARMs:面向多模态模型的即插即用攻击自适应红队智能体

ARMs: Adaptive Red-Teaming Agent against Multimodal Models with Plug-and-Play Attacks

University of Chicago(芝加哥大学);University of Illinois(伊利诺伊大学);Virtue AI;Meta

Zhaorun Chen, Xun Liu, Mintong Kang, Jiawei Zhang, Minzhou Pan, Shuang Yang, Bo Li

AI总结:

针对视觉语言模型多模态接口安全漏洞评估的局限,研究提出自适应红队智能体ARMs,集成多种攻击策略与算法,实现了最优攻击成功率,还构建了大规模安全数据集以提升模型鲁棒性。

摘要:

随着视觉语言模型(VLM)日益普及,其多模态接口也带来了新的安全漏洞,使得安全评估既具挑战性又至关重要。现有红队工作要么局限于狭窄的对抗模式集合,要么严重依赖人工设计,缺乏对新兴真实世界VLM漏洞的可扩展探索。为弥补这一差距,我们提出ARMs——一种自适应红队智能体,可系统地对VLM开展全面风险评估。给定目标有害行为或风险定义,ARMs通过推理增强的多步编排自动优化多样化红队策略,以有效诱导目标VLM生成有害输出。我们提出11种新型多模态攻击策略,覆盖VLM的多种对抗模式(例如推理劫持、上下文隐藏),并通过模型上下文协议(MCP)将17种红队算法集成到ARMs中。为平衡攻击的多样性与有效性,我们设计了带有ε-贪心攻击探索算法的分层记忆机制。在基于实例和基于策略的基准上开展的大量实验表明,ARMs取得了SOTA(当前最优)攻击成功率,平均超出基线52.1%,在Claude-4-Sonnet上的攻击成功率超过90%。我们发现,ARMs生成的红队实例多样性显著更高,能够揭示VLM中涌现的漏洞。借助ARMs,我们构建了ARMs-Bench——一个大规模多模态安全数据集,包含超3万个红队实例,覆盖51个不同风险类别,兼具真实世界多模态威胁与监管风险背景。使用ARMs-Bench进行安全微调,可在保留VLM通用效用的同时大幅提升其鲁棒性,为提升多模态安全对齐以应对新兴威胁提供了可操作的指导。

英文摘要:

As vision-language models (VLMs) gain prominence, their multimodal interfaces also introduce new safety vulnerabilities, making the safety evaluation challenging and critical. Existing red-teaming efforts are either restricted to a narrow set of adversarial patterns or depend heavily on manual engineering, lacking scalable exploration of emerging real-world VLM vulnerabilities. To bridge this gap, we propose ARMs, an adaptive red-teaming agent that systematically conducts comprehensive risk assessments for VLMs. Given a target harmful behavior or risk definition, ARMs automatically optimizes diverse red-teaming strategies with reasoning-enhanced multi-step orchestration, to effectively elicit harmful outputs from target VLMs. We propose 11 novel multimodal attack strategies, covering diverse adversarial patterns of VLMs (e.g., reasoning hijacking, contextual cloaking), and integrate 17 red-teaming algorithms into ARMs via model context protocol (MCP). To balance the diversity and effectiveness of the attack, we design a layered memory with an epsilon-greedy attack exploration algorithm. Extensive experiments on instance- and policy-based benchmarks show that ARMs achieves SOTA attack success rates, exceeding baselines by an average of 52.1% and surpassing 90% on Claude-4-Sonnet. We show that the diversity of red-teaming instances generated by ARMs is significantly higher, revealing emerging vulnerabilities in VLMs. Leveraging ARMs, we construct ARMs-Bench, a large-scale multimodal safety dataset comprising over 30K red-teaming instances spanning 51 diverse risk categories, grounded in both real-world multimodal threats and regulatory risks. Safety fine-tuning with ARMs-Bench substantially improves the robustness of VLMs while preserving their general utility, providing actionable guidance to improve multimodal safety alignment against emerging threats.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑