List your company

Jailbreak Defense on JailBreakV

6.55Attack Success Rate (ASR)CARE
4.983615.556826.1336.7032Mar 28, 2026

Evaluation results

MethodLinks
CAREBackbone=Qwen-2.5VL2026.036.55
CAREBackbone=LLaVA-onevision2026.037.03
JailGuardBackbone=Qwen-2.5VL2026.039.47
SPO-VLMBackbone=LLaVA-onevision2026.0310.37
SPO-VLMBackbone=Qwen-2.5VL2026.0311.12
ASTRABackbone=LLaVA-onevision2026.0311.98
ASTRABackbone=Qwen-2.5VL2026.0312.3
CASTBackbone=Qwen-2.5VL2026.0315.68
JailGuardBackbone=LLaVA-onevision2026.0318.46
Refusal PairsBackbone=Qwen-2.5VL2026.0318.85
Original modelBackbone=Qwen-2.5VL2026.0320
CASTBackbone=LLaVA-onevision2026.0325.03
Refusal PairsBackbone=LLaVA-onevision2026.0332.81
Original modelBackbone=LLaVA-onevision2026.0345.71