List your company

Jailbreak Attacks on MM-Safety

99.9Safety RateTRR
-1.91624.51750.9577.383Dec 8, 2025Jan 11, 2026Feb 15, 2026Mar 21, 2026Apr 25, 2026May 29, 2026Jul 3, 2026

Evaluation results

MethodLinks
TRRBackbone=Qwen2.5VL-7B, Alignment Method=TRR2025.1299.9
TRRBackbone=Qwen2.5VL-32B, Alignment Method=TRR2025.1299.4
MSR-AlignBackbone=Qwen2.5VL-32B, Alignment Method=MSR-Align2025.1299.1
TiSBackbone=Qwen2.5VL-32B, Alignment Method=TiS2025.1298.9
MSR-AlignBackbone=Qwen2.5VL-7B, Alignment Method=MSR-Align2025.1298.2
Gemini-2.5-proModel=Gemini-2.5-pro2025.1294.8
Claude-3.5-SonnetModel=Claude-3.5-Sonnet2025.1291.9
Safework-R1-7BModel=Safework-R1-7B2025.1288.3
TiSBackbone=Qwen2.5VL-7B, Alignment Method=TiS2025.1285.8
INFER+Test Model=Gemini 2.5 Flash2026.0783
INFER+Test Model=Qwen3-VL2026.0777
GPT-4oModel=GPT-4o2025.1274.7
INFERTest Model=Gemini 2.5 Flash2026.0766
Qwen2.5VL-32BBackbone=Qwen2.5VL-32B, Alignment Method=None2025.1254.7
CS-DJTest Model=Gemini 2.5 Flash2026.0754
Qwen2.5VL-7BBackbone=Qwen2.5VL-7B, Alignment Method=None2025.1250.1
ImageTypoTest Model=Gemini 2.5 Flash2026.0742
INFERTest Model=Qwen3-VL2026.0741
CS-DJTest Model=Qwen3-VL2026.0732
ImageTypoTest Model=Qwen3-VL2026.0715
TVPITest Model=Gemini 2.5 Flash2026.0711
TVPITest Model=Qwen3-VL2026.072