Robustness of Vision Language Models Against Split-Image Harmful Input Attacks

TL;DR

This paper uncovers a new vulnerability in vision-language models where split-image harmful inputs bypass safety measures, introduces novel attacks exploiting this flaw, and proposes methods to improve model robustness.

Contribution

The work identifies a gap in safety alignment for split-image inputs, develops novel split-image jailbreak attacks including Adv-KD, and suggests solutions to enhance VLM robustness.

Findings

Split-image inputs often bypass safety filters.

Proposed attacks achieve up to 60% higher transfer success.

Adv-KD improves cross-model attack transferability.

Abstract

Vision-Language Models (VLMs) are now a core part of modern AI. Recent work proposed several visual jailbreak attacks using single/ holistic images. However, contemporary VLMs demonstrate strong robustness against such attacks due to extensive safety alignment through preference optimization (e.g., RLHF). In this work, we identify a new vulnerability: while VLM pretraining and instruction tuning generalize well to split-image inputs, safety alignment is typically performed only on holistic images and does not account for harmful semantics distributed across multiple image fragments. Consequently, VLMs often fail to detect and refuse harmful split-image inputs, where unsafe cues emerge only after combining images. We introduce novel split-image visual jailbreak attacks (SIVA) that exploit this misalignment. Unlike prior optimization-based attacks, which exhibit poor black-box transferability due to architectural and prior mismatches across models, our attacks evolve in progressive phases from naive splitting to an adaptive white-box attack, culminating in a black-box transfer attack. Our strongest strategy leverages a novel adversarial knowledge distillation (Adv-KD) algorithm to substantially improve cross-model transferability. Evaluations on three state-of-the-art modern VLMs and three jailbreak datasets demonstrate that our strongest attack achieves up to 60% higher transfer success than existing baselines. Lastly, we propose efficient ways to address this critical vulnerability in the current VLM safety alignment.