VL-Trojan: Multimodal Instruction Backdoor Attacks against Autoregressive Visual Language Models
This addresses a security threat for users of multimodal AI systems, though it is incremental as it builds on existing backdoor attack methods.
The paper tackles the vulnerability of autoregressive visual language models to backdoor attacks during instruction tuning, proposing VL-Trojan, which uses isolating and clustering for image triggers and iterative text triggers to achieve a 62.52% higher attack success rate than baselines.
Autoregressive Visual Language Models (VLMs) showcase impressive few-shot learning capabilities in a multimodal context. Recently, multimodal instruction tuning has been proposed to further enhance instruction-following abilities. However, we uncover the potential threat posed by backdoor attacks on autoregressive VLMs during instruction tuning. Adversaries can implant a backdoor by injecting poisoned samples with triggers embedded in instructions or images, enabling malicious manipulation of the victim model's predictions with predefined triggers. Nevertheless, the frozen visual encoder in autoregressive VLMs imposes constraints on the learning of conventional image triggers. Additionally, adversaries may encounter restrictions in accessing the parameters and architectures of the victim model. To address these challenges, we propose a multimodal instruction backdoor attack, namely VL-Trojan. Our approach facilitates image trigger learning through an isolating and clustering strategy and enhance black-box-attack efficacy via an iterative character-level text trigger generation method. Our attack successfully induces target outputs during inference, significantly surpassing baselines (+62.52\%) in ASR. Moreover, it demonstrates robustness across various model scales and few-shot in-context reasoning scenarios.