List your company

Jailbreak Attack on MaliciousInstruct

100ASRPiF
44.8859.1973.587.81Sep 15, 2025Oct 13, 2025Nov 11, 2025Dec 10, 2025Jan 7, 2026Feb 5, 2026Mar 6, 2026

Evaluation results

MethodLinks
PiFTarget Model=Vicuna-13B-V1.52025.12100————421——
PiFTarget Model=GPT-4-06132025.12100————299——
PiFTarget Model=Llama-3.1-8B-Instruct2025.1299————240——
PiFTarget Model=Mistral-7B-Instruct2025.1299————318——
PiFTarget Model=GPT-O1-Preview2025.1298————282——
REATarget Model=Mistral-7B-Instruct2026.0398———————
GCGTarget Model=Vicuna-13B-V1.52025.1296————248——
GCGTarget Model=Mistral-7B-Instruct2025.1295————368——
Audio Narrative AttacksTarget Model=Gemini 2.0 Flash2026.0194———————
REATarget Model=Qwen2.5-7B-Instruct2026.0394———————
DeepInception (Text Baseline)Target Model=Qwen2.5-Omni, Modality=Text2026.0193———————
FORCETarget Model=InstructBlip-Vicuna-7B2025.0992———————
SCAVTarget Model=Mistral-7B-Instruct2026.0392———————
Audio Narrative AttacksTarget Model=GPT-4o Realtime2026.0191———————
PAIRTarget Model=GPT-4-06132025.1291————178——
REATarget Model=Llama3.1-8B-Instruct2026.0390———————
SCAVTarget Model=Llama3.1-8B-Instruct2026.0389———————
AdvWaveTarget Model=Gemini 2.0 Flash2026.0187———————
GCGTarget Model=GPT-4-06132025.1287————227——
SAHATarget Model=Llama3.1-8B-Instruct2026.0387——————84
Phi-4Pruning Percentage=100%2025.0987———————
gemma-7b-itPruning Percentage=75%2025.0987———————
Audio Narrative AttacksTarget Model=Gemini 2.0 Flash, Modality=Audio2026.0186———————
Audio Narrative AttacksTarget Model=Qwen2.5-Omni, Modality=Audio2026.0186———————
SAHATarget Model=Qwen1.5-7B-Chat2026.0386——————81
Qwen2.5-14B-InstructPruning Percentage=75%2025.0986———————
gemma-7b-itPruning Percentage=100%2025.0986———————
ConVATarget Model=Qwen1.5-7B-Chat2026.0385——————40
Llama-3.2-1B-InstructPruning Percentage=100%2025.0985———————
Phi-4Pruning Percentage=75%2025.0985———————
IT2026.0284.3———————
AdvWaveTarget Model=GPT-4o Realtime2026.0184———————
PAIRTarget Model=Mistral-7B-Instruct2025.1284————181——
PGDTarget Model=InstructBlip-Vicuna-7B2025.0984———————
CAATarget Model=Qwen2.5-7B-Instruct2026.0384———————
Qwen2.5-14B-InstructPruning Percentage=100%2025.0984———————
Phi-4Pruning Percentage=50%2025.0984———————
AutoDANTarget Model=Qwen1.5-7B-Chat2026.0383——————64
Llama-3.2-1B-InstructPruning Percentage=75%2025.0983———————
PiFTarget Model=Llama-2-13B-Chat2025.1282————202——
Llama-3.2-3B-InstructPruning Percentage=75%2025.0982———————
QwQ-32BPruning Percentage=100%2025.0982———————
RL2026.0281.8———————
Audio Narrative AttacksTarget Model=Qwen2.5-Omni2026.0181———————
SAHATarget Model=Deepseek-llm-7B-Chat2026.0381——————77
Llama-3.2-3B-InstructPruning Percentage=100%2025.0981———————
DeepSeek-R1-Distill Llama-8BPruning Percentage=75%2025.0981———————
QwQ-32BPruning Percentage=75%2025.0981———————
SCAVTarget Model=Qwen1.5-7B-Chat2026.0380——————81
DeepSeek-R1-Distill Qwen-1.5BPruning Percentage=75%2025.0980———————
DeepSeek-R1-Distill Qwen-1.5BPruning Percentage=100%2025.0980———————
DeepSeek-R1-Distill Llama-8BPruning Percentage=100%2025.0980———————
Llama-3.2-3B-InstructPruning Percentage=50%2025.0979———————
DeepSeek-R1-Distill Qwen-1.5BPruning Percentage=50%2025.0979———————
DeepSeek-R1-Distill Llama-8BPruning Percentage=50%2025.0978———————
Qwen2.5-7B-InstructPruning Percentage=75%2025.0977———————
Qwen2.5-7B-InstructPruning Percentage=100%2025.0977———————
Phi-4-mini-instructPruning Percentage=75%2025.0977———————
GCGTarget Model=Llava-v1.6-mistral-7b2025.0974———————
SCAVTarget Model=Deepseek-llm-7B-Chat2026.0374——————80
DeepInception (Text Baseline)Target Model=Gemini 2.0 Flash, Modality=Text2026.0173———————
Phi-4-mini-instructPruning Percentage=50%2025.0973———————
Phi-4-mini-instructPruning Percentage=100%2025.0973———————
DeepSeek-R1-Distill Qwen-1.5BPruning Percentage=0%2025.0973———————
DeepSeek-R1-Distill Qwen-1.5BPruning Percentage=25%2025.0973———————
GCGTarget Model=GPT-O1-Preview2025.1272————195——
CAATarget Model=Mistral-7B-Instruct2026.0372———————
PAIRTarget Model=Llama-3.1-8B-Instruct2025.1270————214——
PAIRTarget Model=Vicuna-13B-V1.52025.1270————171——
CAATarget Model=Qwen1.5-7B-Chat2026.0370——————37
CAATarget Model=Llama3.1-8B-Instruct2026.0370———————
FORCETarget Model=Llava-v1.6-mistral-7b2025.0969———————
SCAVTarget Model=Llama3.1-8B-Instruct2026.0368——————82
GCGTarget Model=Mistral-7B-Instruct2026.0368———————
DeepSeek-R1-Distill Llama-8BPruning Percentage=25%2025.0968———————
AutoDANTarget Model=Llama3.1-8B-Instruct2026.0367——————66
FORCETarget Model=Idefics3-8B-Llama3, Source Model=InstructBLIP-Vicuna-7B2025.0966—————38.48—
BasedTarget Model=Mistral-7B-Instruct2026.0366———————
gemma-2b-itPruning Percentage=100%2025.0966———————
FORCETarget Model=Idefics3-8B-Llama32025.0964———————
SCAVTarget Model=Qwen2.5-7B-Instruct2026.0364———————
AdvWaveTarget Model=Qwen2.5-Omni2026.0163———————
AutoDAN-TurboTarget Model=Qwen1.5-7B-Chat2026.0363——————72
AutoDANTarget Model=Deepseek-llm-7B-Chat2026.0363——————70
gemma-2b-itPruning Percentage=75%2025.0963———————
PGDTarget Model=Llava-v1.6-mistral-7b2025.0961———————
PAIRTarget Model=Llama3.1-8B-Instruct2026.0360——————6
GCGTarget Model=Qwen1.5-7B-Chat2026.0360——————66
PAIRTarget Model=Mistral-7B-Instruct2026.0356———————
Audio Narrative AttacksTarget Model=GPT-4o Realtime, Modality=Audio2026.0155———————
Qwen2.5-14B-InstructPruning Percentage=50%2025.0955———————
PAIRTarget Model=GPT-O1-Preview2025.1254————140——
GCGTarget Model=InstructBlip-Vicuna-7B2025.0953———————
PGDTarget Model=Idefics3-8B-Llama32025.0953———————
ConVATarget Model=Mistral-7B-Instruct2026.0352———————
PAIRTarget Model=Llama-2-13B-Chat2025.1251————133——
FORCETarget Model=Llava-v1.6-mistral-7b, Source Model=InstructBLIP-Vicuna-7B2025.0950—————52.93—
PGDTarget Model=Idefics3-8B-Llama3, Source Model=InstructBLIP-Vicuna-7B2025.0950—————53.09—
Qwen2.5-7B-InstructPruning Percentage=50%2025.0949———————
gemma-7b-itPruning Percentage=50%2025.0947———————