AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

Toward Universal and Transferable Jailbreak Attacks on Vision-Language Models

Kaiyuan Cui, Yige Li, Yutao Wu, Xingjun Ma, Sarah Monazam Erfani, Christopher Leckie, Hanxun Huang

ICLR 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: Vision-language model robustness · 3 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add the optimised jailbreak image (vs text-only input) (ablation)UltraBreak · SafeBench black-box averageASR (%): 40.79 -> 71.83Improved
add the optimised jailbreak image (vs text-only input) (ablation)UltraBreak · AdvBench black-box averageASR (%): 25.48 -> 57.64Improved
add random transformations and TV constraints on the image (ablation)UltraBreak · SafeBench black-box averageASR (%): 51.99 -> 71.83Improved
add random transformations and TV constraints on the image (ablation)UltraBreak · AdvBench black-box averageASR (%): 29.86 -> 57.64Improved
add random transformations and TV constraints on the image (ablation)UltraBreak · SafeBench, white-box Qwen2-VL surrogateASR (%): 89.84 -> 81.59Hurt
replace cross-entropy with the attention semantic loss (ablation)UltraBreak · SafeBench black-box averageASR (%): 55.80 -> 71.83Improved
replace cross-entropy with the attention semantic loss (ablation)UltraBreak · AdvBench black-box averageASR (%): 40.15 -> 57.64Improved
add attention weighting to the semantic loss (ablation)UltraBreak · SafeBench black-box averageASR (%): 57.54 -> 71.83Improved
add attention weighting to the semantic loss (ablation)UltraBreak · AdvBench black-box averageASR (%): 41.83 -> 57.64Improved
increase TV loss weight from 0 to 0.2 (ablation)UltraBreak · SafeBench, Qwen-VL and GLM-4.1V averageASR (%): 65.88 -> 72.38Improved
increase TV loss weight from 0.5 to 1 (ablation)UltraBreak · SafeBench, Qwen-VL and GLM-4.1V averageASR (%): 69.37 -> 63.18Hurt
change the affirming phrase in TPG (ablation)UltraBreak · SafeBench, GLM-4.1V-9B-ThinkingASR (%): 66.03 / 72.06 / 66.98No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy