Helpfulness Evaluation on AdvBench-audio Harmful-Safe
86.83BRR ScoreSARSteer
Evaluation Results
| Method | Links | |
|---|---|---|
| SARSteerModel=Kimi-Audio2025.10 | 86.83 | |
| No DefenseModel=Qwen2-Audio2025.10 | 85.19 | |
| SARSteerModel=Qwen2-Audio2025.10 | 85 | |
| MDSteer-c2rModel=Qwen2-Audio2025.10 | 84.23 | |
| MDSteer-c2rModel=Kimi-Audio2025.10 | 83.62 | |
| MDSteer-h2sModel=Kimi-Audio2025.10 | 81.25 | |
| MDSteer-h2sModel=Qwen2-Audio2025.10 | 81.15 | |
| AdaShieldModel=Qwen2-Audio2025.10 | 79.81 | |
| FSDModel=Qwen2-Audio2025.10 | 63.95 | |
| No DefenseModel=Kimi-Audio2025.10 | 60.77 | |
| FSDModel=Kimi-Audio2025.10 | 54.81 | |
| AdaShieldModel=Kimi-Audio2025.10 | 45.29 |