List your company

Multimodal Jailbreak Attack on MM-SafetyBench (full)

95.42ASRMML
21.621640.780859.9479.0992May 24, 2026

Evaluation results

MethodLinks
MMLTarget Model=Qwen2.5-VL-7B2026.0595.42
Furina (Typo)Target Model=Qwen2.5-VL-7B2026.0593.93
Furina (Typo)Target Model=GPT-4o-mini2026.0593.81
MMLTarget Model=GPT-4o2026.0593.75
Furina (Typo)Target Model=GPT-4o2026.0593.51
Furina (Typo)Target Model=Gemini-2.5-Flash2026.0592.79
Furina (SD)Target Model=GPT-4o2026.0592.44
MMLTarget Model=GPT-4o-mini2026.0592.14
MMLTarget Model=Gemini-2.5-Flash2026.0591.96
Furina (SD)Target Model=GPT-4o-mini2026.0591.96
Furina (SD)Target Model=Gemini-2.5-Flash2026.0591.9
Furina (SD)Target Model=Qwen2.5-VL-7B2026.0591.61
JailBoundTarget Model=Qwen2.5-VL-7B2026.0591.4
VAJMTarget Model=Qwen2.5-VL-7B2026.0588.25
UMKTarget Model=Qwen2.5-VL-7B2026.0580.43
Furina (Typo)Target Model=Claude-Haiku-4.52026.0577.2
Furina (SD)Target Model=Claude-Haiku-4.52026.0575.48
JailBoundTarget Model=GPT-4o2026.0575.24
JailBoundTarget Model=GPT-4o-mini2026.0573.57
JailBoundTarget Model=Gemini-2.5-Flash2026.0570.06
UMKTarget Model=Gemini-2.5-Flash2026.0551.79
JailBoundTarget Model=Claude-Haiku-4.52026.0551.37
BaselineTarget Model=Qwen2.5-VL-7B2026.0551.19
UMKTarget Model=GPT-4o2026.0550.77
UMKTarget Model=GPT-4o-mini2026.0546.85
FigStepTarget Model=Qwen2.5-VL-7B2026.0546.61
MMLTarget Model=Claude-Haiku-4.52026.0540.76
FigStepTarget Model=GPT-4o-mini2026.0538.81
VAJMTarget Model=Gemini-2.5-Flash2026.0537.98
BaselineTarget Model=GPT-4o-mini2026.0536.49
UMKTarget Model=Claude-Haiku-4.52026.0535.89
FigStepTarget Model=GPT-4o2026.0535.36
FigStepTarget Model=Claude-Haiku-4.52026.0533.27
BaselineTarget Model=GPT-4o2026.0532.74
BaselineTarget Model=Gemini-2.5-Flash2026.0531.37
FigStepTarget Model=Gemini-2.5-Flash2026.0531.19
VAJMTarget Model=GPT-4o2026.0528.45
VAJMTarget Model=GPT-4o-mini2026.0527.38
BaselineTarget Model=Claude-Haiku-4.52026.0527.14
VAJMTarget Model=Claude-Haiku-4.52026.0524.46