Video Jailbreaking on VIDEOSAFETYBENCH
88.6Refusal RateSASA
Evaluation Results
| Method | Links | |
|---|---|---|
| SASAModel=QWEN3.52026.06 | 88.6 | |
| MARSModel=QWEN3-VL2026.06 | 88.2 | |
| SASAModel=MOLMO22026.06 | 88.1 | |
| MARSModel=QWEN3.52026.06 | 78.2 | |
| SASAModel=INTERNVL 3.52026.06 | 76.9 | |
| MARSModel=MOLMO22026.06 | 69.2 | |
| AdaSteerModel=QWEN3.52026.06 | 67.8 | |
| AdaSteerModel=QWEN3-VL2026.06 | 49.7 | |
| ECSOModel=MOLMO22026.06 | 47.2 | |
| SASAModel=QWEN3-VL2026.06 | 40.3 | |
| AdaSteerModel=MOLMO22026.06 | 39.6 | |
| ECSOModel=QWEN3.52026.06 | 34.9 | |
| ECSOModel=QWEN3-VL2026.06 | 32 | |
| Zero-shotModel=MOLMO22026.06 | 28.8 | |
| Zero-shotModel=QWEN3-VL2026.06 | 28.8 | |
| Zero-shotModel=QWEN3.52026.06 | 26.2 | |
| MARSModel=INTERNVL 3.52026.06 | 11.9 | |
| ECSOModel=INTERNVL 3.52026.06 | 9 | |
| AdaSteerModel=INTERNVL 3.52026.06 | 6.6 | |
| Zero-shotModel=INTERNVL 3.52026.06 | 4.7 |