List your company

Jailbreaking on AdvBench (Detoxify & Llama Guard 3 ASR)

0ASR (Detoxify)VAE-JB
-0.4482.5765.68.624Apr 11, 2026

Evaluation results

MethodLinks
VAE-JBModel=InternVL, epsilon (ε)=322026.0401
VAE-JBModel=MiniGPT-4, epsilon (ε)=32, transfer=true2026.04075.4
BAPModel=Qwen-VL, epsilon (ε)=322026.0404.2
BAPModel=LLaVA-1.5, epsilon (ε)=322026.04054.8
BAPModel=InternVL, epsilon (ε)=322026.0400.8
JB-PcsModel=Qwen-VL, epsilon (ε)=–2026.0401.5
JB-PcsModel=InternVL, epsilon (ε)=–2026.0400.6
JB-PcsModel=LLaVA-1.5, epsilon (ε)=–2026.040.258.7
Attention-Guided Visual JailbreakingModel=Qwen-VL, epsilon (ε)=162026.040.244.8
BAPModel=MiniGPT-4, epsilon (ε)=32, transfer=true2026.040.475.8
Attention-Guided Visual JailbreakingModel=MiniGPT-4, epsilon (ε)=16, transfer=true2026.040.474.3
VAE-JBModel=LLaVA-1.5, epsilon (ε)=322026.040.657.9
JB-PcsModel=MiniGPT-4, epsilon (ε)=–, transfer=true2026.040.862.7
Attention-Guided Visual JailbreakingModel=MiniGPT-4, epsilon (ε)=32, transfer=true2026.041.376.3
Attention-Guided Visual JailbreakingModel=LLaVA-1.5, epsilon (ε)=162026.041.562.3
VAE-JBModel=Qwen-VL, epsilon (ε)=322026.041.968.8
Attention-Guided Visual JailbreakingModel=InternVL, epsilon (ε)=322026.043.81.5
Attention-Guided Visual JailbreakingModel=InternVL, epsilon (ε)=162026.047.718.3
Attention-Guided Visual JailbreakingModel=LLaVA-1.5, epsilon (ε)=322026.048.377.5
Attention-Guided Visual JailbreakingModel=Qwen-VL, epsilon (ε)=322026.0411.294.4