Response Safety on JailBreakV-28K (avg)
0.975JBV-R ScoreDACO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DACOTarget Model=InternVL3.5-8B-Instruct [108], Steering Method=DACO (Ours)2026.04 | 0.975 | 0.97 | |
| MOP [56]Target Model=InternVL3.5-8B-Instruct [108], Steering Method=MOP [56]2026.04 | 0.917 | 0.926 | |
| DACOTarget Model=Qwen2.5-VL-7B-Instruct [5], Steering Method=DACO (Ours)2026.04 | 0.903 | 0.841 | |
| DACOTarget Model=LLaVA1.6-Mistral-7B [48], Steering Method=DACO (Ours)2026.04 | 0.885 | 0.803 | |
| ActAdd [50, 101, 134]Target Model=InternVL3.5-8B-Instruct [108], Steering Method=ActAdd [50, 101, 134]2026.04 | 0.86 | 0.847 | |
| MOP [56]Target Model=Qwen2.5-VL-7B-Instruct [5], Steering Method=MOP [56]2026.04 | 0.835 | 0.752 | |
| MOP [56]Target Model=LLaVA1.6-Mistral-7B [48], Steering Method=MOP [56]2026.04 | 0.811 | 0.69 | |
| Prompting [52]Target Model=InternVL3.5-8B-Instruct [108], Steering Method=Prompting [52]2026.04 | 0.809 | 0.755 | |
| ActAdd [50, 101, 134]Target Model=LLaVA1.6-Mistral-7B [48], Steering Method=ActAdd [50, 101, 134]2026.04 | 0.783 | 0.559 | |
| No SteeringTarget Model=InternVL3.5-8B-Instruct [108], Steering Method=No Steering2026.04 | 0.73 | 0.649 | |
| ActAdd [50, 101, 134]Target Model=Qwen2.5-VL-7B-Instruct [5], Steering Method=ActAdd [50, 101, 134]2026.04 | 0.691 | 0.675 | |
| Prompting [52]Target Model=Qwen2.5-VL-7B-Instruct [5], Steering Method=Prompting [52]2026.04 | 0.659 | 0.622 | |
| No SteeringTarget Model=Qwen2.5-VL-7B-Instruct [5], Steering Method=No Steering2026.04 | 0.564 | 0.543 | |
| Prompting [52]Target Model=LLaVA1.6-Mistral-7B [48], Steering Method=Prompting [52]2026.04 | 0.471 | 0.31 | |
| No SteeringTarget Model=LLaVA1.6-Mistral-7B [48], Steering Method=No Steering2026.04 | 0.303 | 0.203 |