CalvinBackup/OmniSafeBench-MM
1
0
Fork 0
mirror of https://github.com/jiaxiaojunQAQ/OmniSafeBench-MM.git synced 2026-02-12 17:52:46 +00:00
Code Issues Packages Projects Releases Wiki Activity

Update table format and add venue information

Browse Source
This commit is contained in:
gq-max
2025-12-03 20:52:41 +08:00
committed by GitHub
parent 915db82299
commit 94ffd2e59c
1 changed files with 17 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
README.md
View File

@@ -34,23 +34,23 @@ OmniSafeBench-MM is a unified benchmark and open-source toolbox for evaluating m
The benchmark unifies multi-modal jailbreak attack–defense evaluation, 13 attack and 15 defense methods, and a three-dimensional scoring protocol measuring harmfulness, alignment, and detail.
## 🗡️ Integrated Attack Methods
| Name | Title | Paper | Code | Classification |
|:----------------------------:|:-----------------------------------------------------------------------------------------------------------------------------:|:---------------------------------------------------------------:|:-------------------------------------------------------------------------------------------------------:|:--------------------------------------------------------:|
| FigStep / FigStep-Pro | FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Prompts | [link](https://arxiv.org/abs/2311.05608) | [link](https://github.com/ThuCCSLab/FigStep) | Black-box — Structured visual carriers |
| QR-Attack (MM-SafetyBench) | MM-SafetyBench: A Benchmark for Safety Evaluation of Multimodal Large Language Models | [link](https://arxiv.org/abs/2311.17600) | [link](https://github.com/isXinLiu/MM-SafetyBench) | Black-box — Structured visual carriers |
| MML | Jailbreak Large Vision-Language Models Through Multi-Modal Linkage | [link](https://aclanthology.org/2025.acl-long.74/) | [link](https://github.com/wangyu-ovo/MML) | Black-box — Structured visual carriers |
| CS-DJ | Distraction is All You Need for Multimodal Large Language Model Jailbreaking | [link](https://arxiv.org/abs/2502.10794) | [link](https://github.com/TeamPigeonLab/CS-DJ) | Black-box — OOD (attention / distribution manipulation) |
| SI-Attack | Jailbreaking Multimodal Large Language Models via Shuffle Inconsistency | [link](https://arxiv.org/abs/2501.04931) | [link](https://github.com/zhaoshiji123/SI-Attack) | Black-box — OOD (shuffle / attention inconsistency) |
| JOOD | Playing the Fool: Jailbreaking LLMs and Multimodal LLMs with Out-of-Distribution Strategy | [link](https://arxiv.org/abs/2503.20823) | [link](https://github.com/naver-ai/JOOD) | Black-box — OOD (OOD strategy) |
| HIMRD | Heuristic-Induced Multimodal Risk Distribution (HIMRD) Jailbreak Attack | [link](https://arxiv.org/abs/2412.05934) | [link](https://github.com/MaTengSYSU/HIMRD-jailbreak) | Black-box — OOD / risk distribution |
| HADES | Images are Achilles’ Heel of Alignment: Exploiting Visual Vulnerabilities for Jailbreaking Multimodal Large Language Models | [link](https://arxiv.org/abs/2403.09792) | [link](https://github.com/AoiDragon/HADES) | Black-box — Structured visual carriers |
| BAP | Jailbreak Vision Language Models via Bi-Modal Adversarial Prompt (BAP) | [link](https://arxiv.org/abs/2406.04031) | [link](https://github.com/NY1024/BAP-Jailbreak-Vision-Language-Models-via-Bi-Modal-Adversarial-Prompt) | White-box — Cross-modal |
| visual_adv | Visual Adversarial Examples Jailbreak Aligned Large Language Models | [link](https://ojs.aaai.org/index.php/AAAI/article/view/30150) | [link](https://github.com/Unispac/Visual-Adversarial-Examples-Jailbreak-Large-Language-Models) | White-box — Single-modal |
| VisCRA | VisCRA: A Visual Chain Reasoning Attack for Jailbreaking Multimodal Large Language Models | [link](https://arxiv.org/abs/2505.19684) | [link](https://github.com/DyMessi/VisCRA) | Black-box — Structured visual carriers |
| UMK | White-box Multimodal Jailbreaks Against Large Vision-Language Models (Universal Master Key, UMK) | [link](https://arxiv.org/abs/2405.17894) | [link](https://github.com/roywang021/UMK) | White-box — Cross-modal |
| PBI-Attack | Prior-Guided Bimodal Interactive Black-Box Jailbreak Attack for Toxicity Maximization (PBI-Attack) | [link](https://aclanthology.org/2025.emnlp-main.32.pdf) | [link](https://github.com/Rosy0912/PBI-Attack) | Black-box — Query optimization & transfer |
| ImgJP / DeltaJP | Jailbreaking Attack against Multimodal Large Language Model (imgJP / deltaJP) | [link](https://arxiv.org/abs/2402.02309) | [link](https://github.com/abc03570128/Jailbreaking-Attack-against-Multimodal-Large-Language-Model) | White-box — Single-modal |
| JPS | JPS: Jailbreak Multimodal Large Language Models with Collaborative Visual Perturbation and Textual Steering | [link](https://arxiv.org/abs/2508.05087) | [link](https://github.com/thu-coai/JPS) | White-box — Cross-modal |
| Name | Title | Venue | Paper | Code |
| :------------------------: | :----------------------------------------------------------: | :--------: | :----------------------------------------------------: | :----------------------------------------------------------: |
| FigStep / FigStep-Pro | FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Prompts | AAAI 2025 | https://arxiv.org/abs/2311.05608 | https://github.com/ThuCCSLab/FigStep |
| QR-Attack (MM-SafetyBench) | MM-SafetyBench: A Benchmark for Safety Evaluation of Multimodal Large Language Models | ECCV 2024 | https://arxiv.org/abs/2311.17600 | https://github.com/isXinLiu/MM-SafetyBench |
| MML | Jailbreak Large Vision-Language Models Through Multi-Modal Linkage | ACL 2025 | https://aclanthology.org/2025.acl-long.74/ | https://github.com/wangyu-ovo/MML |
| CS-DJ | Distraction is All You Need for Multimodal Large Language Model Jailbreaking | CVPR 2025 | https://arxiv.org/abs/2502.10794 | https://github.com/TeamPigeonLab/CS-DJ |
| SI-Attack | Jailbreaking Multimodal Large Language Models via Shuffle Inconsistency | ICCV 2025 | https://arxiv.org/abs/2501.04931 | https://github.com/zhaoshiji123/SI-Attack |
| JOOD | Playing the Fool: Jailbreaking LLMs and Multimodal LLMs with Out-of-Distribution Strategy | CVPR 2025 | https://arxiv.org/abs/2503.20823 | https://github.com/naver-ai/JOOD |
| HIMRD | Heuristic-Induced Multimodal Risk Distribution (HIMRD) Jailbreak Attack | ICCV 2025 | https://arxiv.org/abs/2412.05934 | https://github.com/MaTengSYSU/HIMRD-jailbreak |
| HADES | Images are Achilles’ Heel of Alignment: Exploiting Visual Vulnerabilities for Jailbreaking MLLMs | ECCV 2024 | https://arxiv.org/abs/2403.09792 | https://github.com/AoiDragon/HADES |
| BAP | Jailbreak Vision Language Models via Bi-Modal Adversarial Prompt (BAP) | TIFS 2025 | https://arxiv.org/abs/2406.04031 | https://github.com/NY1024/BAP-Jailbreak-Vision-Language-Models-via-Bi-Modal-Adversarial-Prompt |
| visual_adv | Visual Adversarial Examples Jailbreak Aligned Large Language Models | AAAI 2024 | https://ojs.aaai.org/index.php/AAAI/article/view/30150 | https://github.com/Unispac/Visual-Adversarial-Examples-Jailbreak-Large-Language-Models |
| VisCRA | VisCRA: A Visual Chain Reasoning Attack for Jailbreaking Multimodal Large Language Models | EMNLP 2025 | https://arxiv.org/abs/2505.19684 | https://github.com/DyMessi/VisCRA |
| UMK | White-box Multimodal Jailbreaks Against Large Vision-Language Models (Universal Master Key) | ACMMM 2024 | https://arxiv.org/abs/2405.17894 | https://github.com/roywang021/UMK |
| PBI-Attack | Prior-Guided Bimodal Interactive Black-Box Jailbreak Attack for Toxicity Maximization | EMNLP 2025 | https://aclanthology.org/2025.emnlp-main.32.pdf | https://github.com/Rosy0912/PBI-Attack |
| ImgJP / DeltaJP | Jailbreaking Attack against Multimodal Large Language Models | arXiv 2024 | https://arxiv.org/abs/2402.02309 | https://github.com/abc03570128/Jailbreaking-Attack-against-Multimodal-Large-Language-Model |
| JPS | JPS: Jailbreak Multimodal Large Language Models with Collaborative Visual Perturbation and Textual Steering | ACMMM 2025 | https://arxiv.org/abs/2508.05087 | https://github.com/thu-coai/JPS |
## 🛡️Integrated Defense Methods