List your company

Jailbreak attacks on MML-Mirror (MML-M)

97Safety RateTRR
-3.15222.84948.8574.851Dec 8, 2025

Evaluation results

MethodLinks
TRRBackbone=Qwen2.5VL-7B, Alignment Method=TRR2025.1297
TRRBackbone=Qwen2.5VL-32B, Alignment Method=TRR2025.1280.5
TiSBackbone=Qwen2.5VL-32B, Alignment Method=TiS2025.1265
MSR-AlignBackbone=Qwen2.5VL-32B, Alignment Method=MSR-Align2025.1255.7
MSR-AlignBackbone=Qwen2.5VL-7B, Alignment Method=MSR-Align2025.1253.7
Claude-3.5-SonnetModel=Claude-3.5-Sonnet2025.1240
TiSBackbone=Qwen2.5VL-7B, Alignment Method=TiS2025.1228.8
Gemini-2.5-proModel=Gemini-2.5-pro2025.1215.7
Safework-R1-7BModel=Safework-R1-7B2025.1212.5
Qwen2.5VL-7BBackbone=Qwen2.5VL-7B, Alignment Method=None2025.126.5
GPT-4oModel=GPT-4o2025.121.9
Qwen2.5VL-32BBackbone=Qwen2.5VL-32B, Alignment Method=None2025.120.7