Safety Evaluation on Harmful Prompts (ASR Metrics)
2ASR (Raw)No Intervention
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| No InterventionBackbone=DeepSeek2026.05 | 2 | — | — | |
| No InterventionBackbone=Qwen32026.05 | 2 | — | — | |
| No InterventionBackbone=Phi2026.05 | 2.5 | — | — | |
| No InterventionBackbone=OLMoE2026.05 | 4 | — | — | |
| No InterventionBackbone=GPT-oss2026.05 | 5 | — | — | |
| GCGBackbone=GPT-oss2026.05 | 6.5 | 5 | 1.5 | |
| GCGBackbone=Qwen32026.05 | 9 | 1 | 0 | |
| GCGBackbone=OLMoE2026.05 | 14 | 10.5 | 2.5 | |
| GCGBackbone=DeepSeek2026.05 | 15 | 2 | 0 | |
| GCGBackbone=Phi2026.05 | 18 | 0.5 | 0 | |
| RASETBackbone=Phi2026.05 | 61 | 55.5 | 29.5 | |
| RASETBackbone=GPT-oss2026.05 | 67.5 | 61 | 39.5 | |
| RASETBackbone=OLMoE2026.05 | 78 | 75.5 | 50.5 | |
| RASETBackbone=DeepSeek2026.05 | 92.5 | 90 | 61.5 | |
| RASETBackbone=Qwen32026.05 | 94 | 88 | 71.5 |