Stealthy and Adjustable Text-Guided Backdoor Attacks on Multimodal Pretrained Models
This work addresses security vulnerabilities in multimodal pretrained models, which is critical for real-world applications, though it is incremental as it builds on existing backdoor attack methods.
The paper tackles the vulnerability of multimodal pretrained models to backdoor attacks by proposing a Text-Guided Backdoor (TGB) attack that uses common words as triggers, improving stealthiness and practicality, and achieves adjustable attack success rates in tasks like Composed Image Retrieval and Visual Question Answering.
Multimodal pretrained models are vulnerable to backdoor attacks, yet most existing methods rely on visual or multimodal triggers, which are impractical since visually embedded triggers rarely occur in real-world data. To overcome this limitation, we propose a novel Text-Guided Backdoor (TGB) attack on multimodal pretrained models, where commonly occurring words in textual descriptions serve as backdoor triggers, significantly improving stealthiness and practicality. Furthermore, we introduce visual adversarial perturbations on poisoned samples to modulate the model's learning of textual triggers, enabling a controllable and adjustable TGB attack. Extensive experiments on downstream tasks built upon multimodal pretrained models, including Composed Image Retrieval (CIR) and Visual Question Answering (VQA), demonstrate that TGB achieves practicality and stealthiness with adjustable attack success rates across diverse realistic settings, revealing critical security vulnerabilities in multimodal pretrained models.