List your company

Jailbreak Attack on JailbreakBench (ASR)

100ASRDictAttack
61.998471.864281.7391.5958Mar 31, 2025May 31, 2025Jul 31, 2025Sep 30, 2025Nov 30, 2025Jan 30, 2026Apr 1, 2026

Evaluation results

MethodLinks
DictAttackModel=gemini-2.5-pro, Grammar Guardrail=disabled2025.03100
EnumAttackModel=gpt-4o, Grammar Guardrail=None2025.03100
DictAttackModel=gpt-4o, Grammar Guardrail=None, k=12025.03100
DictAttackModel=gemini-2.5-pro, Grammar Guardrail=enabled2025.0398.3
DictAttackModel=gpt-4o, Grammar Guardrail=llama-guard-3-8b, k=12025.0398.3
DictAttackModel=deepseek-r1, Grammar Guardrail=disabled2025.0397
EnumAttackModel=gpt-4o-mini, Grammar Guardrail=None2025.0397
Audio Narrative AttacksTarget Model=GPT-4o Realtime2026.0196.33
ICD-SEEDBackbone=Qwen-3-4B2026.0495.77
DictAttackModel=deepseek-r1, Grammar Guardrail=enabled2025.0395.4
DictAttackModel=gpt-5, Grammar Guardrail=disabled2025.0395
DictAttackModel=gpt-4o-mini, Grammar Guardrail=None, k=12025.0394
DictAttackModel=gpt-5, Grammar Guardrail=enabled2025.0393.4
Audio Narrative AttacksTarget Model=Gemini 2.0 Flash2026.0193.33
MAPATarget Model=Llava V1.6-Mistral-7B2026.0293.33
MAPATarget Model=Qwen2.5-VL-7B-Instruct2026.0293.33
Prompt+RSTarget Model=GPT-3.5, Model Type=Closed-Source2026.0193
DictAttackModel=gpt-4o-mini, Grammar Guardrail=llama-guard-3-8b, k=12025.0392.4
EnumAttackModel=gemini-2.0-flash, Grammar Guardrail=None2025.0392
DictAttackModel=gpt-oss-120b, Grammar Guardrail=disabled2025.0391
DictAttackModel=gemini-2.0-flash, Grammar Guardrail=None, k=12025.0391
ICD-SEEDBackbone=Qwen-2.5-3B2026.0490.96
ICD-PREFILLBackbone=DeepSeek-R1-Distill-Llama-70B2026.0490.58
Prompt+RSTarget Model=LLaMA2-7B, Model Type=Open-Source2026.0190
ICD-PREFILLBackbone=Qwen-2.5-3B2026.0489.81
ICD-SEEDBackbone=Qwen-2.5-7B2026.0489.81
DictAttackModel=gemini-2.0-flash, Grammar Guardrail=llama-guard-3-8b, k=12025.0389.5
DictAttackModel=gpt-oss-120b, Grammar Guardrail=enabled2025.0389.4
ICD-PREFILLBackbone=DeepSeek-R1-Distill-Llama-8B2026.0489.04
AdvWaveTarget Model=Gemini 2.0 Flash2026.0189
ICD-SEEDTarget Model=Vicuna-1.5-13B2026.0489
MAPATarget Model=Average2026.0288.89
PrefillBackbone=DeepSeek-R1-Distill-Qwen-32B2026.0488.46
PrefillBackbone=Llama-3.3-70B2026.0487.5
ICD-SEEDBackbone=DeepSeek-R1-Distill-Llama-8B2026.0487.12
AdvWaveTarget Model=GPT-4o Realtime2026.0187
ICD-PREFILLBackbone=DeepSeek-R1-Distill-Qwen-32B2026.0486.92
ICD-PREFILLBackbone=DeepSeek-R1-Distill-Qwen-14B2026.0486.54
ICD-PREFILLBackbone=Qwen-3-8B2026.0485.58
ICD-PREFILLBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0485.38
ICD-SEEDBackbone=Llama-3.1-8B2026.0485.19
ICD-PREFILLBackbone=Qwen-2.5-7B2026.0485.19
ICD-SEEDBackbone=Qwen-3-8B2026.0485.19
ICD-PREFILLBackbone=Llama-3.1-8B2026.0485
PrefillBackbone=DeepSeek-R1-Distill-Llama-8B2026.0484.62
PrefillBackbone=DeepSeek-R1-Distill-Llama-70B2026.0484.23
ICD-PREFILLTarget Model=Vicuna-1.5-13B2026.0483
ICD-PREFILLBackbone=Llama-3.3-70B2026.0482.88
ICD-PREFILLBackbone=Qwen-3-4B2026.0482.69
RICEModel=Bagel2026.0382
RICEModel=Janus2026.0382
PrefillBackbone=Llama-3.2-3B2026.0481.34
PrefillBackbone=DeepSeek-R1-Distill-Qwen-14B2026.0481.15
ICD-SEEDBackbone=Qwen-3-14B2026.0480.38
AdvWaveTarget Model=Qwen2.5-Omni2026.0180
MAPATarget Model=Llama-3.2-Vision-11B-Instruct2026.0280
PrefillBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0480
Safe2HarmTarget Model=Llama-3-8B-Instruct2025.1279.8
CoATarget Model=Llava V1.6-Mistral-7B2026.0278.33
PrefillBackbone=Qwen-3-14B2026.0478.27
Prompt+RSTarget Model=GPT-4, Model Type=Closed-Source2026.0178
PAIRTarget Model=Llama-3.1-70B2026.0478
ICD-SEEDBackbone=Qwen-3-32B2026.0477.5
ICD-SEEDBackbone=DeepSeek-R1-Distill-Qwen-32B2026.0477.31
PrefillBackbone=Qwen-2.5-7B2026.0477.12
ICD-PREFILLBackbone=Qwen-2.5-72B2026.0476.35
PrefillBackbone=Llama-3.1-8B2026.0475.96
ICD-PREFILLBackbone=Qwen-2.5-32B2026.0475.58
FigStep-ProModel=Janus2026.0375
ICD-PREFILLBackbone=Llama-3.2-3B2026.0474.23
PlainModel=Janus2026.0374
ICD-PREFILLBackbone=Qwen-2.5-14B2026.0473.27
Audio Narrative AttacksTarget Model=Qwen2.5-Omni2026.0173
PrefillBackbone=Qwen-2.5-3B2026.0472.69
FigStepModel=Janus2026.0372
ICD-PREFILLBackbone=Gemma-3-27B2026.0471.77
ICD-AUTOBackbone=Llama-3.1-8B2026.0471.73
ICD-PREFILLBackbone=Gemma-3-12B2026.0471.35
PAIRTarget Model=GPT-3.5, Model Type=Closed-Source2026.0171
FigStep-ProModel=Bagel2026.0371
CoATarget Model=Qwen2.5-VL-7B-Instruct2026.0270
ICD-PREFILLBackbone=Qwen-3-14B2026.0468.85
ICD-AUTOBackbone=Qwen-2.5-3B2026.0468.27
ICD-AUTOBackbone=Qwen-2.5-7B2026.0468.08
FigStepModel=Bagel2026.0368
AMATarget Model=Llama-3.1-70B2026.0468
ICD-SEEDBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0467.88
CoATarget Model=Average2026.0267.78
ICD-SEEDBackbone=Gemma-3-4B2026.0467.12
FootInTheDoorTarget Model=Llava V1.6-Mistral-7B2026.0266.66
PrefillBackbone=Gemma-3-12B2026.0466.54
ICD-SEEDBackbone=Gemma-3-12B2026.0466.16
ICD-SEEDTarget Model=Llama-3.1-70B2026.0466
GCGModel=Bagel2026.0365
PrefillBackbone=Qwen-2.5-32B2026.0464.81
ICD-SEEDBackbone=Qwen-2.5-14B2026.0464.42
ICD-SEEDBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0464.23
PAIRModel=Bagel2026.0364
PrefillBackbone=Gemma-3-4B2026.0463.85
ICD-PREFILLBackbone=Gemma-3-4B2026.0463.46