List your company

Jailbreak Detection on MM-SafetyBench

99.18AUROCMahal-OOD
50.008862.774475.5488.3056Dec 12, 2025Jan 10, 2026Feb 8, 2026Mar 9, 2026Apr 7, 2026May 6, 2026Jun 5, 2026

Evaluation results

MethodLinks
Mahal-OODModel=LLaVA, Uses target model hidden states=true2025.1299.1899.32
Mahal-OOD*Model=FLAVA, Uses target model hidden states=false2025.1297.0197.62
HiddenDetectVLM=LLaVa2026.0696.8—
KNN-OODModel=LLaVA, Uses target model hidden states=true2025.1295.2395.91
JailDAM*Model=CLIP, Uses target model hidden states=false2025.1291.2698.04
MTKVLM=LLaVa2026.0691.1—
KNN-OOD*Model=FLAVA, Uses target model hidden states=false2025.1285.4790.65
GradSafeModel=LLaVA2025.1285.1487.52
ECSOVLM=LLaVa2026.0684.8—
HiddenDetectModel=LLaVA2025.1282.6993.53
MTKVLM=Qwen-VL2026.0681.8—
MirrorCheckVLM=Qwen-VL2026.0678.1—
HiddenDetectVLM=Qwen-VL2026.0677.4—
JailGuardVLM=Qwen-VL2026.0675—
LLaVaGuardModel=Qwen2025.1274.2787.29
MirrorCheckVLM=LLaVa2026.0668.8—
ECSOVLM=Qwen-VL2026.0668.6—
CIDERVLM=LLaVa2026.0667.3—
CIDERVLM=Qwen-VL2026.0667.3—
JailDAMVLM=LLaVa2026.0661.2—
VLGuardModel=LLaVA2025.1261.0680.2
JailGuardVLM=LLaVa2026.0653.3—
JailDAMVLM=Qwen-VL2026.0651.9—