Jailbreak Defense on BeaverTails
69Reference PerformanceCHASE Defender
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| CHASE DefenderAttack=Translation*2026.06 | 69 | 2.71 | 7.9 | 0.012 | 13.2 | |
| CHASE DefenderAttack=AutoDAN2026.06 | 57 | 3.77 | 30.4 | 0.314 | 50.8 | |
| Base ModelAttack=Translation*2026.06 | 54 | 2.2 | 9.1 | — | — | |
| CHASE DefenderAttack=TAP2026.06 | 51 | 2.99 | 31 | 0.265 | 46.1 | |
| CHASE DefenderAttack=PAP2026.06 | 47 | 2.75 | 24.1 | 0.231 | 48.9 | |
| Base ModelAttack=TAP2026.06 | 30 | 4.04 | 57.5 | — | — | |
| Base ModelAttack=PAP2026.06 | 26 | 3.16 | 47.2 | — | — | |
| Base ModelAttack=AutoDAN2026.06 | 21 | 3.7 | 61.8 | — | — | |
| CHASE DefenderAttack=PAIR2026.06 | 15 | 2.98 | 45.1 | 0.178 | 28.3 | |
| Base ModelAttack=PAIR2026.06 | 8 | 3.39 | 62.9 | — | — |