List your company

Multimodal Jailbreaking on HADES-Dataset

99.07ASR (%)MML
-3.827622.886249.676.3138Nov 30, 2024

Evaluation results

MethodLinks
MMLTarget Model=GPT-4o, Encryption Method=Base64 Encoding (B64), System Prompt=None2024.1199.07
MMLTarget Model=GPT-4o, Encryption Method=Image Mirroring (M), System Prompt=None2024.1198.8
MMLTarget Model=GPT-4o, Encryption Method=Word Replacement (WR), System Prompt=None2024.1198.4
MMLTarget Model=GPT-4o, Encryption Method=Image Rotation (R), System Prompt=None2024.1198.4
MMLTarget Model=GPT-4o-Mini, Encryption Method=Image Mirroring (M), System Prompt=None2024.1198.27
MMLTarget Model=GPT-4o-Mini, Encryption Method=Image Rotation (R), System Prompt=None2024.1198.13
MMLTarget Model=GPT-4o-Mini, Encryption Method=Word Replacement (WR), System Prompt=None2024.1197.6
MMLTarget Model=Qwen-VL-Max, Encryption Method=Image Rotation (R), System Prompt=None2024.1197.2
MMLTarget Model=Qwen-VL-Max, Encryption Method=Word Replacement (WR), System Prompt=None2024.1196.93
MMLTarget Model=Qwen-VL-Max, Encryption Method=Image Mirroring (M), System Prompt=None2024.1196.67
MMLTarget Model=GPT-4o-Mini, Encryption Method=Base64 Encoding (B64), System Prompt=None2024.1194.13
MMLTarget Model=Qwen-VL-Max, Encryption Method=Base64 Encoding (B64), System Prompt=None2024.1192.13
MMLTarget Model=Claude-3.5-Sonnet, Encryption Method=Image Mirroring (M), System Prompt=None2024.1145.73
HADESTarget Model=Qwen-VL-Max, System Prompt=None2024.1140.93
MMLTarget Model=Claude-3.5-Sonnet, Encryption Method=Word Replacement (WR), System Prompt=None2024.1139.33
MMLTarget Model=Claude-3.5-Sonnet, Encryption Method=Image Rotation (R), System Prompt=None2024.1133.47
MMLTarget Model=Claude-3.5-Sonnet, Encryption Method=Base64 Encoding (B64), System Prompt=None2024.1110.93
HADESTarget Model=GPT-4o-Mini, System Prompt=None2024.114.93
HADESTarget Model=GPT-4o, System Prompt=None2024.114
HADESTarget Model=Claude-3.5-Sonnet, System Prompt=None2024.110.13