List your company
ResearchFormerly WizWand

Jailbreak Attack on AdvBench 150 Harmful Behaviors

0ASRRA-LLM
-4235077Sep 18, 2023Feb 15, 2024Jul 15, 2024Dec 13, 2024May 12, 2025Oct 10, 2025Mar 10, 2026
Updated 7mo ago

Evaluation results

MethodLinks
RA-LLMAttack=Adv Strings, Model Backbone=Vicuna-7B-v1.32023.09084
TAPTarget Model=Llama-2-7B-chat2026.034—
PAIRTarget Model=Llama-2-7B-chat2026.0310—
RA-LLMAttack Type=TAP, Base Model=Guanaco-7B-HF2023.0912.785.3
RA-LLMAttack Type=AutoDAN-GPT, Base Model=Guanaco-7B-HF2023.0915.384.7
RA-LLMAttack Type=TAP, Base Model=Vicuna-7B-chat-HF2023.091682
RA-LLMAttack Type=AutoDAN, Base Model=Guanaco-7B-HF2023.0916.782
AdvPrompterTarget Model=Llama-2-7B-chat2026.0324—
AutoDANTarget Model=Llama-2-7B-chat2026.0326—
AmpleGCGTarget Model=Llama-2-7B-chat2026.0328—
Baseline2026.0334.8—
RA-LLMAttack Type=AutoDAN-GPT, Base Model=Vicuna-7B-chat-HF2023.0941.347.4
RA-LLMAttack Type=AutoDAN, Base Model=Vicuna-7B-chat-HF2023.094248
GCGTarget Model=Llama-2-7B-chat2026.0354—
MACTarget Model=Llama-2-7B-chat2026.0356—
Probe-SamplingTarget Model=Llama-2-7B-chat2026.0356—
AdvPrompterTarget Model=Vicuna-7B-1.52026.0364—
AmpleGCGTarget Model=Vicuna-7B-1.52026.0366—
AdvPrompterTarget Model=Mistral-7B-Instruct-0.22026.0374—
Original LLMAttack=Adv Strings, Model Backbone=Vicuna-7B-v1.32023.0984—
Amnesia2026.0386.351.5
Original LLMAttack Type=AutoDAN-GPT, Base Model=Vicuna-7B-chat-HF2023.0988.7—
PAIRTarget Model=Mistral-7B-Instruct-0.22026.0390—
Original LLMAttack Type=AutoDAN, Base Model=Vicuna-7B-chat-HF2023.0990.7—
GCGTarget Model=Mistral-7B-Instruct-0.22026.0392—
TAPTarget Model=Mistral-7B-Instruct-0.22026.0392—
MACTarget Model=Mistral-7B-Instruct-0.22026.0394—
Probe-SamplingTarget Model=Mistral-7B-Instruct-0.22026.0394—
PAIRTarget Model=Vicuna-7B-1.52026.0394—
TAPTarget Model=Vicuna-7B-1.52026.0394—
AutoDANTarget Model=Mistral-7B-Instruct-0.22026.0396—
Original LLMAttack Type=TAP, Base Model=Vicuna-7B-chat-HF2023.0998—
Original LLMAttack Type=TAP, Base Model=Guanaco-7B-HF2023.0998—
GCGTarget Model=Vicuna-7B-1.52026.0398—
Original LLMAttack Type=AutoDAN, Base Model=Guanaco-7B-HF2023.0998.7—
Original LLMAttack Type=AutoDAN-GPT, Base Model=Guanaco-7B-HF2023.09100—
MACTarget Model=Vicuna-7B-1.52026.03100—
AutoDANTarget Model=Vicuna-7B-1.52026.03100—
Probe-SamplingTarget Model=Vicuna-7B-1.52026.03100—
I-GCGTarget Model=Vicuna-7B-1.52026.03100—
I-GCGTarget Model=Llama-2-7B-chat2026.03100—
I-GCGTarget Model=Mistral-7B-Instruct-0.22026.03100—
TAO-AttackTarget Model=Vicuna-7B-1.52026.03100—
TAO-AttackTarget Model=Llama-2-7B-chat2026.03100—
TAO-AttackTarget Model=Mistral-7B-Instruct-0.22026.03100—