List your company

Jailbreaking MLLMs on MM-SafetyBench

100WASROurs (Wide-Net-Casting)
26.57645.63864.783.762May 16, 2026

Evaluation results

MethodLinks
Ours (Wide-Net-Casting)Safety Alignment Strategy=Original Safety Alignment2026.051000.934
wide-net-casting jailbreak methodSafety Alignment=Original Safety Alignment2026.051000.939
Naive Strategy 2Safety Alignment=Original Safety Alignment2026.0595.10.907
Naive Strategy 1Safety Alignment=Original Safety Alignment2026.0594.90.899
Naive Strategy 2Safety Alignment Strategy=Original Safety Alignment2026.0593.70.903
MLAI+PixArt-αSafety Alignment=Original Safety Alignment2026.0593.70.891
Naive Strategy 1Safety Alignment Strategy=Original Safety Alignment2026.05930.892
Baseline (MLAI + PixArt-α)Safety Alignment Strategy=Original Safety Alignment2026.0589.90.865
wide-net-casting jailbreak methodSafety Alignment=Original Safety Alignment + VLGuard2026.0553.50.517
wide-net-casting jailbreak methodSafety Alignment=Original Safety Alignment + IMMUNE2026.0550.10.469
Ours (Wide-Net-Casting)Safety Alignment Strategy=Original Safety Alignment + VLGuard2026.0547.60.435
Ours (Wide-Net-Casting)Safety Alignment Strategy=Original Safety Alignment + IMMUNE2026.0546.70.423
Naive Strategy 2Safety Alignment=Original Safety Alignment + VLGuard2026.0544.10.418
wide-net-casting jailbreak methodSafety Alignment=Original Safety Alignment + ASTRA2026.0543.60.382
Naive Strategy 1Safety Alignment=Original Safety Alignment + VLGuard2026.0543.40.409
Ours (Wide-Net-Casting)Safety Alignment Strategy=Original Safety Alignment + ASTRA2026.0542.30.381
Naive Strategy 2Safety Alignment=Original Safety Alignment + IMMUNE2026.0540.80.363
MLAI+PixArt-αSafety Alignment=Original Safety Alignment + VLGuard2026.0540.20.387
Naive Strategy 1Safety Alignment=Original Safety Alignment + IMMUNE2026.0540.10.359
Naive Strategy 2Safety Alignment Strategy=Original Safety Alignment + VLGuard2026.0537.80.343
MLAI+PixArt-αSafety Alignment=Original Safety Alignment + IMMUNE2026.0537.20.321
Naive Strategy 1Safety Alignment Strategy=Original Safety Alignment + VLGuard2026.0537.10.335
Naive Strategy 2Safety Alignment Strategy=Original Safety Alignment + IMMUNE2026.0536.60.321
Naive Strategy 2Safety Alignment=Original Safety Alignment + ASTRA2026.0535.60.311
Naive Strategy 1Safety Alignment Strategy=Original Safety Alignment + IMMUNE2026.0535.30.318
Naive Strategy 1Safety Alignment=Original Safety Alignment + ASTRA2026.0535.20.309
Baseline (MLAI + PixArt-α)Safety Alignment Strategy=Original Safety Alignment + VLGuard2026.0534.30.303
Baseline (MLAI + PixArt-α)Safety Alignment Strategy=Original Safety Alignment + IMMUNE2026.0533.10.294
MLAI+PixArt-αSafety Alignment=Original Safety Alignment + ASTRA2026.0532.90.271
Naive Strategy 2Safety Alignment Strategy=Original Safety Alignment + ASTRA2026.0532.20.295
Naive Strategy 1Safety Alignment Strategy=Original Safety Alignment + ASTRA2026.0531.50.283
Baseline (MLAI + PixArt-α)Safety Alignment Strategy=Original Safety Alignment + ASTRA2026.0529.40.257