Toward Universal and Transferable Jailbreak Attacks on Vision-Language Models

TL;DR

This paper introduces UltraBreak, a novel framework for creating universal, transferable jailbreak attacks on vision-language models by leveraging vision-space transformations and semantic textual objectives, significantly improving attack transferability.

Contribution

UltraBreak is the first method to combine vision-level regularisation with semantic textual objectives to generate universal, transferable jailbreaks for vision-language models, overcoming prior transferability limitations.

Findings

UltraBreak outperforms previous jailbreak methods in transferability.

Semantic objectives help smooth the loss landscape, enhancing transferability.

The approach generalizes across diverse models and attack targets.

Abstract

Vision-language models (VLMs) extend large language models (LLMs) with vision encoders, enabling text generation conditioned on both images and text. However, this multimodal integration expands the attack surface by exposing the model to image-based jailbreaks crafted to induce harmful responses. Existing gradient-based jailbreak methods transfer poorly, as adversarial patterns overfit to a single white-box surrogate and fail to generalise to black-box models. In this work, we propose Universal and transferable jailbreak (UltraBreak), a framework that constrains adversarial patterns through transformations and regularisation in the vision space, while relaxing textual targets through semantic-based objectives. By defining its loss in the textual embedding space of the target LLM, UltraBreak discovers universal adversarial patterns that generalise across diverse jailbreak objectives. This combination of vision-level regularisation and semantically guided textual supervision mitigates surrogate overfitting and enables strong transferability across both models and attack targets. Extensive experiments show that UltraBreak consistently outperforms prior jailbreak methods. Further analysis reveals why earlier approaches fail to transfer, highlighting that smoothing the loss landscape via semantic objectives is crucial for enabling universal and transferable jailbreaks. The code is publicly available in our \href{https://github.com/kaiyuanCui/UltraBreak}{GitHub repository}.