Multimodal Jailbreak Attack on MM-SafetyBench (full)
95.42ASRMML
Evaluation Results
| Method | Links | |
|---|---|---|
| MMLTarget Model=Qwen2.5-VL-7B2026.05 | 95.42 | |
| Furina (Typo)Target Model=Qwen2.5-VL-7B2026.05 | 93.93 | |
| Furina (Typo)Target Model=GPT-4o-mini2026.05 | 93.81 | |
| MMLTarget Model=GPT-4o2026.05 | 93.75 | |
| Furina (Typo)Target Model=GPT-4o2026.05 | 93.51 | |
| Furina (Typo)Target Model=Gemini-2.5-Flash2026.05 | 92.79 | |
| Furina (SD)Target Model=GPT-4o2026.05 | 92.44 | |
| MMLTarget Model=GPT-4o-mini2026.05 | 92.14 | |
| MMLTarget Model=Gemini-2.5-Flash2026.05 | 91.96 | |
| Furina (SD)Target Model=GPT-4o-mini2026.05 | 91.96 | |
| Furina (SD)Target Model=Gemini-2.5-Flash2026.05 | 91.9 | |
| Furina (SD)Target Model=Qwen2.5-VL-7B2026.05 | 91.61 | |
| JailBoundTarget Model=Qwen2.5-VL-7B2026.05 | 91.4 | |
| VAJMTarget Model=Qwen2.5-VL-7B2026.05 | 88.25 | |
| UMKTarget Model=Qwen2.5-VL-7B2026.05 | 80.43 | |
| Furina (Typo)Target Model=Claude-Haiku-4.52026.05 | 77.2 | |
| Furina (SD)Target Model=Claude-Haiku-4.52026.05 | 75.48 | |
| JailBoundTarget Model=GPT-4o2026.05 | 75.24 | |
| JailBoundTarget Model=GPT-4o-mini2026.05 | 73.57 | |
| JailBoundTarget Model=Gemini-2.5-Flash2026.05 | 70.06 | |
| UMKTarget Model=Gemini-2.5-Flash2026.05 | 51.79 | |
| JailBoundTarget Model=Claude-Haiku-4.52026.05 | 51.37 | |
| BaselineTarget Model=Qwen2.5-VL-7B2026.05 | 51.19 | |
| UMKTarget Model=GPT-4o2026.05 | 50.77 | |
| UMKTarget Model=GPT-4o-mini2026.05 | 46.85 | |
| FigStepTarget Model=Qwen2.5-VL-7B2026.05 | 46.61 | |
| MMLTarget Model=Claude-Haiku-4.52026.05 | 40.76 | |
| FigStepTarget Model=GPT-4o-mini2026.05 | 38.81 | |
| VAJMTarget Model=Gemini-2.5-Flash2026.05 | 37.98 | |
| BaselineTarget Model=GPT-4o-mini2026.05 | 36.49 | |
| UMKTarget Model=Claude-Haiku-4.52026.05 | 35.89 | |
| FigStepTarget Model=GPT-4o2026.05 | 35.36 | |
| FigStepTarget Model=Claude-Haiku-4.52026.05 | 33.27 | |
| BaselineTarget Model=GPT-4o2026.05 | 32.74 | |
| BaselineTarget Model=Gemini-2.5-Flash2026.05 | 31.37 | |
| FigStepTarget Model=Gemini-2.5-Flash2026.05 | 31.19 | |
| VAJMTarget Model=GPT-4o2026.05 | 28.45 | |
| VAJMTarget Model=GPT-4o-mini2026.05 | 27.38 | |
| BaselineTarget Model=Claude-Haiku-4.52026.05 | 27.14 | |
| VAJMTarget Model=Claude-Haiku-4.52026.05 | 24.46 |