List your company

Attack Success Rate (ASR) on AdvBench (Jailbreak Attack)

100ASREnumAttack
79.284.69095.4Mar 31, 2025Jun 10, 2025Aug 21, 2025Nov 1, 2025Jan 11, 2026Mar 24, 2026Jun 4, 2026

Evaluation results

MethodLinks
EnumAttackModel=gpt-4o, Grammar Guardrail=None2025.03100
I-GCGTarget Model=Llama-2-7b, Suffix Length=202025.09100
I-GCGTarget Model=Llama-2-7b, Suffix Length=302025.09100
I-GCG + Mask-GCGTarget Model=Llama-2-7b, Suffix Length=202025.09100
I-GCG + Mask-GCGTarget Model=Llama-2-7b, Suffix Length=302025.09100
GCGTarget Model=Vicuna-7b, Suffix Length=302025.09100
I-GCGTarget Model=Vicuna-7b, Suffix Length=202025.09100
I-GCGTarget Model=Vicuna-7b, Suffix Length=302025.09100
I-GCG + Mask-GCGTarget Model=Vicuna-7b, Suffix Length=202025.09100
I-GCG + Mask-GCGTarget Model=Vicuna-7b, Suffix Length=302025.09100
AmpleGCGTarget Model=Vicuna-7b, Suffix Length=302025.09100
AmpleGCG + Mask-GCGTarget Model=Vicuna-7b, Suffix Length=302025.09100
I-GCGTarget Model=Llama-2-13b, Suffix Length=202025.09100
I-GCGTarget Model=Llama-2-13b, Suffix Length=302025.09100
I-GCG + Mask-GCGTarget Model=Llama-2-13b, Suffix Length=302025.09100
AmpleGCGTarget Model=Llama-2-13b, Suffix Length=202025.09100
AmpleGCGTarget Model=Llama-2-13b, Suffix Length=302025.09100
AmpleGCG + Mask-GCGTarget Model=Llama-2-13b, Suffix Length=202025.09100
I-GCGTarget Model=Average, Suffix Length=202025.09100
I-GCGTarget Model=Average, Suffix Length=302025.09100
I-GCG + Mask-GCGTarget Model=Average, Suffix Length=302025.09100
DictAttackModel=gpt-4o, Grammar Guardrail=None, k=12025.0399.8
ICD-SEEDTarget Model=Vicuna-1.5-13B2026.0499.62
DictAttackModel=gemini-2.5-pro, Grammar Guardrail=disabled2025.0399.2
TemplateFuzzModel=Gemma3-4B, Attack Variant=Top-1 Template2026.0499.04
EnumAttackModel=gpt-4o-mini, Grammar Guardrail=None2025.0399
AmpleGCGTarget Model=Vicuna-7b, Suffix Length=202025.0999
I-GCG + Mask-GCGTarget Model=Average, Suffix Length=202025.0999
AmpleGCGTarget Model=Average, Suffix Length=302025.0999
AmpleGCG + Mask-GCGTarget Model=Average, Suffix Length=202025.0999
AmpleGCG + Mask-GCGTarget Model=Average, Suffix Length=302025.0999
DictAttackModel=deepseek-r1, Grammar Guardrail=disabled2025.0398.9
TemplateFuzzModel=DeepSeek-7B, Attack Variant=Top-1 Template2026.0498.84
DictAttackModel=gemini-2.0-flash, Grammar Guardrail=None, k=12025.0398.5
DictAttackModel=gpt-oss-120b, Grammar Guardrail=disabled2025.0398.3
DictAttackModel=gpt-4o, Grammar Guardrail=llama-guard-3-8b, k=12025.0398.1
AmpleGCGTarget Model=Llama-2-7b, Suffix Length=302025.0998
AmpleGCG + Mask-GCGTarget Model=Llama-2-7b, Suffix Length=202025.0998
AmpleGCG + Mask-GCGTarget Model=Llama-2-7b, Suffix Length=302025.0998
GCGTarget Model=Vicuna-7b, Suffix Length=202025.0998
AmpleGCG + Mask-GCGTarget Model=Vicuna-7b, Suffix Length=202025.0998
I-GCG + Mask-GCGTarget Model=Llama-2-13b, Suffix Length=202025.0998
AmpleGCG + Mask-GCGTarget Model=Llama-2-13b, Suffix Length=302025.0998
DictAttackModel=gemini-2.5-pro, Grammar Guardrail=enabled2025.0397.5
DictAttackModel=deepseek-r1, Grammar Guardrail=enabled2025.0397.2
DictAttackModel=gpt-4o-mini, Grammar Guardrail=None, k=12025.0397.1
AmpleGCGTarget Model=Average, Suffix Length=202025.0997
DictAttackModel=gemini-2.0-flash, Grammar Guardrail=llama-guard-3-8b, k=12025.0396.8
DictAttackModel=gpt-oss-120b, Grammar Guardrail=enabled2025.0396.6
DictAttackModel=gpt-5, Grammar Guardrail=disabled2025.0396
GCG + Mask-GCGTarget Model=Vicuna-7b, Suffix Length=202025.0996
GCG + Mask-GCGTarget Model=Vicuna-7b, Suffix Length=302025.0996
I-GCG + SlotGCGDefense=Smooth LLM (swap)2026.0696
GCG-Hij + SlotGCGDefense=Smooth LLM (swap)2026.0696
I-GCGModel=Vicuna-7B, Attack Strategy=I-GCG2026.0696
SlotGCG + I-GCGModel=Vicuna-7B, Attack Strategy=I-GCG2026.0696
EnumAttackModel=gemini-2.0-flash, Grammar Guardrail=None2025.0395.8
TemplateFuzzModel=Llama2-7B, Attack Variant=Top-1 Template2026.0495.58
DictAttackModel=gpt-4o-mini, Grammar Guardrail=llama-guard-3-8b, k=12025.0395.5
DictAttackModel=gpt-5, Grammar Guardrail=enabled2025.0394.4
SlotGCG + I-GCGModel=Llama-2-13B, Attack Strategy=I-GCG2026.0694
AttnGCGModel=Mistral-7B, Attack Strategy=AttnGCG2026.0694
ICD-PREFILLTarget Model=Vicuna-1.5-13B2026.0493.08
AmpleGCGTarget Model=Llama-2-7b, Suffix Length=202025.0993
TemplateFuzzModel=Llama3-8B, Attack Variant=Top-1 Template2026.0492.16
AttnGCG + SlotGCGDefense=Smooth LLM (swap)2026.0692
SlotGCG + AttnGCGModel=Mistral-7B, Attack Strategy=AttnGCG2026.0692
TemplateFuzzModel=Qwen3-8B, Attack Variant=Top-1 Template2026.0491.54
TemplateFuzzModel=Qwen2.5-7B, Attack Variant=Top-1 Template2026.0490.96
SlotGCG + AttnGCGModel=Llama-2-7B, Attack Strategy=AttnGCG2026.0690
SlotGCG + I-GCGModel=Llama-2-7B, Attack Strategy=I-GCG2026.0690
SlotGCG + GCG-HijModel=Llama-2-7B, Attack Strategy=GCG-Hij2026.0690
SlotGCG + GCG-HijModel=Llama-2-13B, Attack Strategy=GCG-Hij2026.0690
I-GCGModel=Mistral-7B, Attack Strategy=I-GCG2026.0690
SlotGCG + I-GCGModel=Mistral-7B, Attack Strategy=I-GCG2026.0690
AttnGCGModel=Vicuna-7B, Attack Strategy=AttnGCG2026.0688
SlotGCG + AttnGCGModel=Average, Attack Strategy=AttnGCG2026.0686.3
GCG + SlotGCGDefense=Smooth LLM (swap)2026.0686
GCGModel=Mistral-7B, Attack Strategy=GCG2026.0686
SlotGCG + GCGModel=Mistral-7B, Attack Strategy=GCG2026.0686
SlotGCG + GCGModel=Vicuna-7B, Attack Strategy=GCG2026.0686
GCG-HijModel=Vicuna-7B, Attack Strategy=GCG-Hij2026.0686
SlotGCG + GCG-HijModel=Qwen-2.5, Attack Strategy=GCG-Hij2026.0686
SlotGCG + I-GCGModel=Average, Attack Strategy=I-GCG2026.0685.7
SlotGCG + GCG-HijModel=Average, Attack Strategy=GCG-Hij2026.0684.3
GCG-HijModel=Mistral-7B, Attack Strategy=GCG-Hij2026.0684
ICD-PREFILLTarget Model=Llama-3.1-70B2026.0482.88
I-GCG + SlotGCGDefense=Erase-and-Check (infusion)2026.0682
I-GCG + SlotGCGDefense=SmoothLLM (insert)2026.0682
SlotGCG + AttnGCGModel=Llama-2-13B, Attack Strategy=AttnGCG2026.0682
SlotGCG + GCGModel=Llama-3.1-8B, Attack Strategy=GCG2026.0682
SlotGCG + AttnGCGModel=Llama-3.1-8B, Attack Strategy=AttnGCG2026.0682
SlotGCG + AttnGCGModel=Vicuna-7B, Attack Strategy=AttnGCG2026.0682
SlotGCG + GCG-HijModel=Vicuna-7B, Attack Strategy=GCG-Hij2026.0682
GCG-HijModel=Qwen-2.5, Attack Strategy=GCG-Hij2026.0682
GCGTarget Model=Average, Suffix Length=302025.0981
GCGTarget Model=Llama-2-13b, Suffix Length=302025.0980
I-GCG + SlotGCGDefense=SmoothLLM (patch)2026.0680
SlotGCG + GCGModel=Llama-2-7B, Attack Strategy=GCG2026.0680
SlotGCG + GCG-HijModel=Mistral-7B, Attack Strategy=GCG-Hij2026.0680