On the Adversarial Robustness of Multi-Modal Foundation Models

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 ({ _ = 1/255}) 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. Note: This paper contains fake information to illustrate the outcome of our attacks. It does not reflect the opinion of the authors.