Jailbreaking MLLMs on MM-SafetyBench
100WASROurs (Wide-Net-Casting)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Ours (Wide-Net-Casting)Safety Alignment Strategy=Original Safety Alignment2026.05 | 100 | 0.934 | |
| wide-net-casting jailbreak methodSafety Alignment=Original Safety Alignment2026.05 | 100 | 0.939 | |
| Naive Strategy 2Safety Alignment=Original Safety Alignment2026.05 | 95.1 | 0.907 | |
| Naive Strategy 1Safety Alignment=Original Safety Alignment2026.05 | 94.9 | 0.899 | |
| Naive Strategy 2Safety Alignment Strategy=Original Safety Alignment2026.05 | 93.7 | 0.903 | |
| MLAI+PixArt-αSafety Alignment=Original Safety Alignment2026.05 | 93.7 | 0.891 | |
| Naive Strategy 1Safety Alignment Strategy=Original Safety Alignment2026.05 | 93 | 0.892 | |
| Baseline (MLAI + PixArt-α)Safety Alignment Strategy=Original Safety Alignment2026.05 | 89.9 | 0.865 | |
| wide-net-casting jailbreak methodSafety Alignment=Original Safety Alignment + VLGuard2026.05 | 53.5 | 0.517 | |
| wide-net-casting jailbreak methodSafety Alignment=Original Safety Alignment + IMMUNE2026.05 | 50.1 | 0.469 | |
| Ours (Wide-Net-Casting)Safety Alignment Strategy=Original Safety Alignment + VLGuard2026.05 | 47.6 | 0.435 | |
| Ours (Wide-Net-Casting)Safety Alignment Strategy=Original Safety Alignment + IMMUNE2026.05 | 46.7 | 0.423 | |
| Naive Strategy 2Safety Alignment=Original Safety Alignment + VLGuard2026.05 | 44.1 | 0.418 | |
| wide-net-casting jailbreak methodSafety Alignment=Original Safety Alignment + ASTRA2026.05 | 43.6 | 0.382 | |
| Naive Strategy 1Safety Alignment=Original Safety Alignment + VLGuard2026.05 | 43.4 | 0.409 | |
| Ours (Wide-Net-Casting)Safety Alignment Strategy=Original Safety Alignment + ASTRA2026.05 | 42.3 | 0.381 | |
| Naive Strategy 2Safety Alignment=Original Safety Alignment + IMMUNE2026.05 | 40.8 | 0.363 | |
| MLAI+PixArt-αSafety Alignment=Original Safety Alignment + VLGuard2026.05 | 40.2 | 0.387 | |
| Naive Strategy 1Safety Alignment=Original Safety Alignment + IMMUNE2026.05 | 40.1 | 0.359 | |
| Naive Strategy 2Safety Alignment Strategy=Original Safety Alignment + VLGuard2026.05 | 37.8 | 0.343 | |
| MLAI+PixArt-αSafety Alignment=Original Safety Alignment + IMMUNE2026.05 | 37.2 | 0.321 | |
| Naive Strategy 1Safety Alignment Strategy=Original Safety Alignment + VLGuard2026.05 | 37.1 | 0.335 | |
| Naive Strategy 2Safety Alignment Strategy=Original Safety Alignment + IMMUNE2026.05 | 36.6 | 0.321 | |
| Naive Strategy 2Safety Alignment=Original Safety Alignment + ASTRA2026.05 | 35.6 | 0.311 | |
| Naive Strategy 1Safety Alignment Strategy=Original Safety Alignment + IMMUNE2026.05 | 35.3 | 0.318 | |
| Naive Strategy 1Safety Alignment=Original Safety Alignment + ASTRA2026.05 | 35.2 | 0.309 | |
| Baseline (MLAI + PixArt-α)Safety Alignment Strategy=Original Safety Alignment + VLGuard2026.05 | 34.3 | 0.303 | |
| Baseline (MLAI + PixArt-α)Safety Alignment Strategy=Original Safety Alignment + IMMUNE2026.05 | 33.1 | 0.294 | |
| MLAI+PixArt-αSafety Alignment=Original Safety Alignment + ASTRA2026.05 | 32.9 | 0.271 | |
| Naive Strategy 2Safety Alignment Strategy=Original Safety Alignment + ASTRA2026.05 | 32.2 | 0.295 | |
| Naive Strategy 1Safety Alignment Strategy=Original Safety Alignment + ASTRA2026.05 | 31.5 | 0.283 | |
| Baseline (MLAI + PixArt-α)Safety Alignment Strategy=Original Safety Alignment + ASTRA2026.05 | 29.4 | 0.257 |