Fooling Contrastive Language-Image Pre-trained Models with CLIPMasterPrints

TL;DR

This paper reveals that CLIP models are vulnerable to specially crafted images called CLIPMasterPrints, which can fool the model into high confidence for many prompts, posing security risks and requiring new mitigation strategies.

Contribution

The paper introduces the concept of CLIPMasterPrints, demonstrates how to mine them using various optimization methods, and proposes mitigation and detection techniques to enhance model robustness.

Findings

Fooling master images can maximize CLIP confidence scores across many prompts.

Blackbox optimization effectively mines CLIPMasterPrints without model access.

Mitigation strategies can improve CLIP robustness and detect malicious images.

Abstract

Models leveraging both visual and textual data such as Contrastive Language-Image Pre-training (CLIP), are the backbone of many recent advances in artificial intelligence. In this work, we show that despite their versatility, such models are vulnerable to what we refer to as fooling master images. Fooling master images are capable of maximizing the confidence score of a CLIP model for a significant number of widely varying prompts, while being either unrecognizable or unrelated to the attacked prompts for humans. The existence of such images is problematic as it could be used by bad actors to maliciously interfere with CLIP-trained image retrieval models in production with comparably small effort as a single image can attack many different prompts. We demonstrate how fooling master images for CLIP (CLIPMasterPrints) can be mined using stochastic gradient descent, projected gradient descent, or blackbox optimization. Contrary to many common adversarial attacks, the blackbox optimization approach allows us to mine CLIPMasterPrints even when the weights of the model are not accessible. We investigate the properties of the mined images, and find that images trained on a small number of image captions generalize to a much larger number of semantically related captions. We evaluate possible mitigation strategies, where we increase the robustness of the model and introduce an approach to automatically detect CLIPMasterPrints to sanitize the input of vulnerable models. Finally, we find that vulnerability to CLIPMasterPrints is related to a modality gap in contrastive pre-trained multi-modal networks. Code available at https://github.com/matfrei/CLIPMasterPrints.