Safety Evaluation on SafetyLLM
95.4Task PerformanceCausal Detoxify
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Causal DetoxifyAttack Method=Causal Detoxify2025.12 | 95.4 | 0.8243 | 6.76 | 0.0277 | |
| AdaptiveAttack Method=Adaptive Poison2025.12 | 94.25 | 0.9257 | 17.56 | 0.1666 | |
| CleanAttack Method=Clean2025.12 | 93.1 | — | 4.73 | — | |
| Two-StepAttack Method=Two-Step2025.12 | 93.1 | 0.8243 | 14.87 | 0.1213 | |
| Extreme PoisonAttack Method=Extreme Poison2025.12 | 92.34 | 0.9597 | 22.97 | 0.1945 | |
| OverpoisonAttack Method=Overpoison2025.12 | 88.89 | 0.9527 | 24.32 | 0.1944 | |
| FusionAttack Method=Fusion2025.12 | 88.51 | 0.5473 | 19.59 | 0.2218 |