Visual anchors in video AI increase risks of hidden harmful content
The Price of Consistency: Exploiting Visual Anchors for Multimodal Jailbreaking in Video Generation
Computer Vision and Pattern Recognition
Summary
Video generation AI is getting better at following instructions by using pictures as visual guides, which helps make videos look more consistent. However, this new work shows that these visual guides can also trap the AI into sticking with harmful ideas, making it harder for the AI to avoid creating bad content. The researchers found that this effect can be exploited to trick the AI into producing harmful videos by separating the harmful idea into a still image and moving text instructions. They tested this method and showed it works better than previous text-only tricks and created a safety test to help others study this problem.
video generationmultimodal inputvisual anchorsAI safetyjailbreakingspatiotemporal consistencymultimodal jailbreakattack success ratesafety benchmark
Authors
Peng Li, Qianqian Xu, Yangbangyan Jiang, Zhipeng Yu, Qingming Huang
Abstract
The rapid evolution of video generation has shifted the paradigm from pure text-driven to multi-conditional controllable generation, with reference images now widely adopted as conditional inputs to achieve superior spatiotemporal consistency. While these reference images serve as powerful visual anchors that significantly enhance controllability, their impact on safety remains largely unexplored. In this work, we reveal the visual anchoring effect: by enforcing consistency, the mechanism prevents the generated content from drifting away from the original harmful intent, thereby eliminating the model's natural safety escape route from harmful to benign content. Consequently, visual anchors inherently increase the safety risk---this is the price of consistency. Building on this insight, we propose Decoupling Intent via Visual Anchors (DIVA), a training-free multimodal jailbreak framework for video generation that exploits this vulnerability. DIVA decouples harmful intent into a static visual anchor image and a dynamic motion text prompt, and employs dual-criteria selection to balance attack stealthiness with semantic preservation. Extensive experiments across various leading commercial platforms and mainstream open-source video generation models demonstrate that DIVA achieves a substantially higher Attack Success Rate than existing text-only methods. To facilitate future research, we additionally contribute TI2VSafetyBench, the first safety benchmark for multi-conditional video generation.