@inproceedings{d53ca57a085a4c15921069fb8233c891,
title = "Cipher-Prompt: Towards a Safe Diffusion Model via Learning Cryptographic Prompts",
abstract = "Security and privacy concerns associated with large generative models have recently attracted significant attention. In particular, there is a pressing need to address potential negative issues resulting from the generation of inappropriate images, including explicit, violent, or politically sensitive content. In this work, we propose a lightweight approach to learn cryptographic prompts, named Cipher-prompt, to prevent diffusion models from generating undesirable images that are semantically related to protected prompts. Cipher-prompt utilizes an untargeted attack objective to optimize a black-box model and generate perturbations that maximize the semantic distance between the protected class and the generated images. Therefore, Cipher-prompt does not require retraining or fine-tuning of the generative model or images as the training dataset. To evaluate the effectiveness of our proposed Cipher-prompt, we conduct thorough qualitative and quantitative experiments, measuring the protection failure rate and collateral impact rate. Experimental results show the efficacy of the proposed Cipher-prompt in balancing risk mitigation with the utility of diffusion-based image generation models.",
keywords = "Black-box attack, Diffusion models, Image protection",
author = "Sidong Jiang and Siyuan Wang and Rui Zhang and Xi Yang and Kaizhu Huang",
note = "Publisher Copyright: {\textcopyright} The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024.; 13th International Conference on Brain Inspired Cognitive Systems, BICS 2023 ; Conference date: 05-08-2023 Through 06-08-2023",
year = "2024",
month = may,
day = "22",
doi = "10.1007/978-981-97-1417-9_30",
language = "English",
isbn = "9789819714162",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Science and Business Media Deutschland GmbH",
pages = "322--332",
editor = "Jinchang Ren and Amir Hussain and Liao, {Iman Yi} and Rongjun Chen and Kaizhu Huang and Huimin Zhao and Xiaoyong Liu and Ping Ma and Thomas Maul",
booktitle = "Advances in Brain Inspired Cognitive Systems - 13th International Conference, BICS 2023, Proceedings",
}