On the Adversarial Robustness of Multi-Modal Foundation Models
This addresses a security vulnerability for users of deployed multi-modal AI systems, highlighting a need for countermeasures against adversarial content.
The paper tackles the problem of adversarial attacks on multi-modal foundation models, showing that imperceptible image perturbations can manipulate caption outputs to harm users, such as by redirecting to malicious websites or spreading fake information.
Multi-modal foundation models combining vision and language models such as Flamingo or GPT-4 have recently gained enormous interest. Alignment of foundation models is used to prevent models from providing toxic or harmful output. While malicious users have successfully tried to jailbreak foundation models, an equally important question is if honest users could be harmed by malicious third-party content. In this paper we show that imperceivable attacks on images in order to change the caption output of a multi-modal foundation model can be used by malicious content providers to harm honest users e.g. by guiding them to malicious websites or broadcast fake information. This indicates that countermeasures to adversarial attacks should be used by any deployed multi-modal foundation model.