repository·main·Indexed 41 days ago
https://github.com/allen-piexl/jailbreakzooA comprehensive research repository and survey documenting the security landscape of Large Language Models (LLMs) and Vision-Language Models (VLMs). It provides structured timelines and resources covering attack (jailbreaking) and defense methodologies, including specific research on text-to-image models, multi-modal adversarial prompts, and voice attacks.
JailbreakZoo is a research repository and survey focused on the security of Large Language Models (LLMs) and Vision-Language Models (VLMs). It provides a structured landscape of vulnerabilities, jailbreaking exploit methods, and defense mechanisms.
Key resources include:
LLM_Jailbreak.md file provides a chronological timeline of research papers and code repositories related to jailbreaking Large Language Models (LLMs). The timeline is categorized by months using seasonal emojis (e.g., ❄️ for Jan, ☀️ for Jun, 🍂 for Sep). Each entry typically includes the publication date, the research title, a link to the full paper (often on arXiv), and, where available, a link to the official implementation code on GitHub or other hosting services.The repository is organized into four primary research areas. You can navigate to these specific documentation files to study techniques and strategies:
This document provides a chronological timeline of research papers and tools focused on defending Vision-Language Models (VLMs) and Multimodal Large Language Models (MLLMs) against jailbreak attacks. It includes links to research papers and, where available, the corresponding implementation code.
### Research Timeline
- **[2024-08-17]** BaThe: Defense against the Jailbreak Attack in Multimodal Large Language Models by Treating Harmful Instruction as Backdoor Trigger
- **[2024-07-31]** Defending Jailbreak Attack in VLMs via Cross-modality Information Detector
- **[2024-06-11]** MLLMGuard: A Multi-dimensional Safety Evaluation Suite for Multimodal Large Language Models
- **[2024-04-08]** Unbridled Icarus: A Survey of the Potential Perils of Image Inputs in Multimodal Large Language Model Security
- **[2024-03-14]** AdaShield: Safeguarding Multimodal Large Language Models from Structure-based Attack via Adaptive Shield Prompting
- **[2024-02-03]** Safety Fine-Tuning at (Almost) No Cost: A Baseline for Vision Large Language Models
- **[2023-12-17]** A Mutation-Based Method for Multi-Modal Jailbreaking Attack Detection
- **[2023-11-29]** MM-SafetyBench: A Benchmark for Safety Evaluation of Multimodal Large Language ModelsIf you use this work in your research or publications, please use the following BibTeX citation:
@article{jin2024jailbreakzoo,
title={JailbreakZoo: Survey, Landscapes, and Horizons in Jailbreaking Large Language and Vision-Language Models},
author={Jin, Haibo and Hu, Leyang and Li, Xinuo and Zhang, Peiyan and Chen, Chonghan and Zhuang, Jun and Wang, Haohan},
journal={arXiv preprint arXiv:2407.01599},
year={2024}
}