Chain of Attack: On the Robustness of Vision-Language Models Against Transfer-Based Adversarial Attacks

TL;DR

This paper introduces Chain of Attack (CoA), a novel method for improving transfer-based adversarial attacks on vision-language models by leveraging multi-modal semantic updates, revealing significant vulnerabilities in these models.

Contribution

We propose CoA, an iterative adversarial attack method that enhances transferability by considering semantic correlations between vision and language modalities.

Findings

CoA achieves higher attack success rates than existing methods.

The method effectively misleads models to generate targeted responses.

Robustness evaluation reveals significant vulnerabilities in current VLMs.

Abstract

Pre-trained vision-language models (VLMs) have showcased remarkable performance in image and natural language understanding, such as image captioning and response generation. As the practical applications of vision-language models become increasingly widespread, their potential safety and robustness issues raise concerns that adversaries may evade the system and cause these models to generate toxic content through malicious attacks. Therefore, evaluating the robustness of open-source VLMs against adversarial attacks has garnered growing attention, with transfer-based attacks as a representative black-box attacking strategy. However, most existing transfer-based attacks neglect the importance of the semantic correlations between vision and text modalities, leading to sub-optimal adversarial example generation and attack performance. To address this issue, we present Chain of Attack (CoA), which iteratively enhances the generation of adversarial examples based on the multi-modal semantic update using a series of intermediate attacking steps, achieving superior adversarial transferability and efficiency. A unified attack success rate computing method is further proposed for automatic evasion evaluation. Extensive experiments conducted under the most realistic and high-stakes scenario, demonstrate that our attacking strategy can effectively mislead models to generate targeted responses using only black-box attacks without any knowledge of the victim models. The comprehensive robustness evaluation in our paper provides insight into the vulnerabilities of VLMs and offers a reference for the safety considerations of future model developments.