List your company

Safety Evaluation on JailBreakV

0.72ASRCrossGuard
-2.879621.417745.71570.0123May 30, 2025Jul 19, 2025Sep 7, 2025Oct 28, 2025Dec 17, 2025Feb 5, 2026Mar 28, 2026

Evaluation results

MethodLinks
CrossGuardCategory=MLLM Guardrails2025.100.72————
Qwen2.5-VL-7BCategory=Offline MLLMs2025.102.14————
HiddenDetectCategory=MLLM Guardrails2025.104.64————
Claude-3.5-SonnetCategory=Online MLLMs2025.105————
GPT-4oCategory=Online MLLMs2025.106.08————
CAREBackbone=Qwen2.5-VL2026.036.55————
ShiftDCBackbone=Qwen2.5-VL2026.0311.17————
AM3SafetyBackbone=Qwen2.5-VL-7B2026.0111.87——63.7168.14
ProFSBackbone=Qwen2.5-VL2026.0316.3————
SPA-VLBackbone=Qwen2.5-VL-7B2026.0121.46——28.153.81
WILDBREAKBase Model=Mistral-7B, Alignment=WILDBREAK2025.0524142.43——
TRIDENT-EDGEBase Model=Mistral-7B, Alignment=TRIDENT-EDGE2025.0526272.32——
JailDAMCategory=MLLM Guardrails2025.1032.5————
MM-DPOBackbone=Qwen2.5-VL-7B2026.0133.79——53.9255.45
Llama-Guard3-VisionCategory=MLLM Guardrails2025.1034.29————
AM3SafetyBackbone=LLaVA-NeXT-7B2026.0135.16——68.4770.56
Mistral-7B (Unaligned)Base Model=Mistral-7B, Alignment=None (Unaligned)2025.0543333.13——
Safe RLHF-VBackbone=Qwen2.5-VL-7B2026.0143.84——62.2156.96
Qwen2.5-VL-7BAlignment=Original Base Model2026.0144.29——5050
LLaVA-1.5-7B (base)Category=Offline MLLMs2025.1051.43————
SPA-VLBackbone=LLaVA-NeXT-7B2026.0152.05——17.0163.4
RLHF-VBackbone=Qwen2.5-VL-7B2026.0154.33——31.4750.16
LLaVA-NeXT-7BAlignment=Original Base Model2026.0174.89——5050
MM-DPOBackbone=LLaVA-NeXT-7B2026.0175.8——50.7249.03
Safe RLHF-VBackbone=LLaVA-NeXT-7B2026.0176.26——4650.43
RLHF-VBackbone=LLaVA-NeXT-7B2026.0178.08——25.7648.35
LlavaGuardCategory=MLLM Guardrails2025.1090.71————