Safety Evaluation on MultiJail Italian
100Safe Response RateDirect Instruction
Evaluation Results
| Method | Links | |
|---|---|---|
| Direct InstructionModel=GPT-OSS-120b2025.09 | 100 | |
| AIMModel=GPT-OSS-120b2025.09 | 100 | |
| SteerMoEModel=GPT-OSS-120b2025.09 | 90.2 | |
| SteerMoE + AIMModel=GPT-OSS-120b2025.09 | 11.4 |