List your company

Malicious Prompt Detection on MMBench OOD

0.16FPRVLMShield
-0.22682.38414.9957.6059Apr 7, 2026

Evaluation results

MethodLinks
VLMShieldTarget Model=LLaVA-1.5-13B2026.040.16
VLMShieldTarget Model=Qwen2.5-VL-7B-Instruct2026.040.16
VLMGuardTarget Model=Qwen2.5-VL-7B-Instruct2026.042
ASTRATarget Model=LLaVA-1.5-13B2026.042.34
CIDERTarget Model=LLaVA-1.5-13B2026.042.54
CIDERTarget Model=Qwen2.5-VL-7B-Instruct2026.042.54
VLMGuardTarget Model=LLaVA-1.5-13B2026.043.08
ECSOTarget Model=Qwen2.5-VL-7B-Instruct2026.044.93
JailGuardTarget Model=Qwen2.5-VL-7B-Instruct2026.045
ASTRATarget Model=Qwen2.5-VL-7B-Instruct2026.045.36
ECSOTarget Model=LLaVA-1.5-13B2026.047.2
JailGuardTarget Model=LLaVA-1.5-13B2026.048.75
MirrorCheckTarget Model=LLaVA-1.5-13B2026.049.83
MirrorCheckTarget Model=Qwen2.5-VL-7B-Instruct2026.049.83