RunawayEvil：越狱图像到视频生成模型
RunawayEvil: Jailbreaking the Image-to-Video Generative Models
针对图像到视频生成模型，给出一种可复用的多模态越狱攻击范式：用强化学习与 LLM 自动搜索文本指令与图像篡改的协同组合，可作为评估 I2V 系统安全对齐的自动化红队工具，也提示防御方需同时约束视觉与文本两条条件通路。
原文依据 we propose Runaway Evil, the first multimodal jailbreaking framework for I2V models with dynamic evolutionary capability
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论图像到视频（I2V）生成是内容创作的前沿方向，模型通过联合推理图像与文本提示来合成动态视觉序列。这种多模态视觉定位使得对视频属性的多样化可控成为可能。然而，正是这一能力引入了关键的安全盲区：攻击者可以利用视觉与文本线索之间的相互作用，发起多模态越狱攻击，从而严重破坏输出的安全性。尽管现实世界的 I2V 系统越来越多地部署安全机制，此类跨模态威胁仍未被探索。现有攻击方法仍局限于单模态设定，仅依赖孤立的文本或图像扰动，这严重限制了其有效性。为弥补这一空白，我们提出 Runaway Evil，这是首个面向 I2V 模型、具备动态演化能力的多模态越狱框架。该框架基于“策略—战术—行动”范式，通过三个核心组件实现自增强攻击：（1）策略感知命令单元，通过强化学习驱动的策略定制与基于大语言模型（LLM）的策略探索，使攻击能够自我演化其策略；（2）多模态战术规划单元，根据所选策略生成协同的文本越狱指令与图像篡改指南；（3）战术行动单元，执行并评估协同攻击。这种自演化架构使框架无需人工干预即可持续适应并强化其攻击策略。大量实验表明，Runaway Evil 在 Open-Sora 2.0 与 CogVideoX 等商用 I2V 模型上取得了最优的攻击成功率。本工作为探测与缓解多模态脆弱性提供了关键工具，为构建更鲁棒的视频生成系统奠定了基础。代码已发布于：https://github.com/DeepSota/RunawayEvil。
英文原文摘要 2038 字符
Image-to-Video (I2V) generation represents a frontier in content creation, where models synthesize dynamic visual sequences by jointly reasoning from both image and text prompts. This multimodal grounding enables diverse controllability over video attributes. However, it is precisely this capability that introduces a critical security blind spot: by exploiting the interplay between visual and textual cues, attackers can launch multimodal jailbreak attacks that severely compromise output security. Despite the increasing implementation of security mechanisms in real-world I2V systems, such cross-modal threats remain unexplored. Existing attack methods remain confined to single-modal settings, relying solely on isolated text or image perturbations, which severely limits their effectiveness. To bridge this gap, we propose Runaway Evil, the first multimodal jailbreaking framework for I2V models with dynamic evolutionary capability. Built on a Strategy-Tactic-Action paradigm, our framework exhibits self-amplifying attack through three core components: (1) a strategy-aware command unit that enables the attack to self-evolve its strategies through reinforcement learning-driven strategy customization and large language model (LLM)-based strategy exploration; (2) a multimodal tactical planning unit that generates synergistic text jailbreak instructions and image tampering guidelines based on the selected strategies; and (3) an tactical action Unit executes and evaluates the coordinated attacks. This self-evolving architecture allows the framework to continuously adapt and intensify its attack strategies without human intervention. Extensive experiments demonstrate that Runaway Evil achieves state-of-the-art attack success rates on commercial I2V models, such as Open-Sora 2.0 and CogVideoX. This work provides a critical tool for probing and mitigating multimodal vulnerabilities, laying a foundation for building more robust video generation systems. The code is available at: https://github.com/DeepSota/RunawayEvil.