List your company

Jailbreak Attack Evaluation on SafeBench 100 sampled harmful queries

97ASRStructBreak
15.8836.945879.06May 25, 2026

Evaluation results

MethodLinks
StructBreakTarget MLLM=Gemini2026.0597
StructBreakTarget MLLM=Max2026.0597
StructBreakTarget MLLM=GPT-52026.0595
StructBreakTarget MLLM=Qwen2.5-VL2026.0595
StructBreakTarget MLLM=GPT-4o2026.0593
FigStepTarget MLLM=Qwen2.5-VL2026.0592
FigStepTarget MLLM=Max2026.0592
StructBreakTarget MLLM=Avg.2026.0592
StructBreakTarget MLLM=GPT-5-mini2026.0590
MM-SafetyBenchTarget MLLM=Gemini2026.0588
MM-SafetyBenchTarget MLLM=Max2026.0588
MM-SafetyBenchTarget MLLM=Qwen2.5-VL2026.0585
StructBreakTarget MLLM=Claude2026.0582
Typeset RewrittenTarget MLLM=Gemini2026.0577
Typeset RewrittenTarget MLLM=Max2026.0577
FigStepTarget MLLM=Gemini2026.0576
RewrittenTarget MLLM=Gemini2026.0570
RewrittenTarget MLLM=Max2026.0570
Typeset RewrittenTarget MLLM=Qwen2.5-VL2026.0570
MM-SafetyBenchTarget MLLM=Avg.2026.0561.2
MM-SafetyBenchTarget MLLM=GPT-4o2026.0561
RewrittenTarget MLLM=GPT-4o2026.0560
Typeset RewrittenTarget MLLM=GPT-4o2026.0557
RewrittenTarget MLLM=Claude2026.0555
FigStepTarget MLLM=Avg.2026.0553.8
Typeset RewrittenTarget MLLM=Avg.2026.0553
RewrittenTarget MLLM=Avg.2026.0551.7
RewrittenTarget MLLM=Qwen2.5-VL2026.0549
MM-SafetyBenchTarget MLLM=GPT-52026.0546
FigStepTarget MLLM=GPT-4o2026.0545
MM-SafetyBenchTarget MLLM=Claude2026.0545
Typeset RewrittenTarget MLLM=GPT-5-mini2026.0543
Typeset RewrittenTarget MLLM=GPT-52026.0542
MM-SafetyBenchTarget MLLM=GPT-5-mini2026.0542
FigStepTarget MLLM=GPT-5-mini2026.0541
RewrittenTarget MLLM=GPT-5-mini2026.0538
RewrittenTarget MLLM=GPT-52026.0538
FigStepTarget MLLM=GPT-52026.0538
OriginalTarget MLLM=GPT-52026.0533
OriginalTarget MLLM=Max2026.0533
FigStepTarget MLLM=Claude2026.0531
OriginalTarget MLLM=GPT-4o2026.0530
OriginalTarget MLLM=GPT-5-mini2026.0529
OriginalTarget MLLM=Claude2026.0529
Typeset RewrittenTarget MLLM=Claude2026.0529
OriginalTarget MLLM=Avg.2026.0527.7
OriginalTarget MLLM=Gemini2026.0526
OriginalTarget MLLM=Qwen2.5-VL2026.0519