Jailbreak Attack on Claude 3.5
0ASRFig-step
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Fig-stepSource=AAAI’252026.03 | 0 | — | |
| VanillaSource=/2026.03 | 0.08 | — | |
| HadesSource=ECCV’242026.03 | 0.26 | — | |
| MM-safetySource=ECCV’242026.03 | 0.53 | — | |
| Bonquery count limit=502026.05 | 1.67 | 49 | |
| PAPlow_perplexity=true, single_turn_attack=true, require_human_strategies=true2025.12 | 2 | — | |
| TAP2026.05 | 2.33 | 48 | |
| PiF2026.05 | 2.67 | 49 | |
| GCGlow_perplexity=false, single_turn_attack=true, require_human_strategies=false2025.12 | 3 | — | |
| PAIRlow_perplexity=true, single_turn_attack=true, require_human_strategies=true2025.12 | 3 | — | |
| CipherChatlow_perplexity=false, single_turn_attack=true, require_human_strategies=true2025.12 | 6.5 | — | |
| JOODSource=CVPR’252026.03 | 28.41 | — | |
| CS-DJSource=CVPR’252026.03 | 31.25 | — | |
| FlipAttack2026.05 | 38.33 | 1 | |
| Code Attacklow_perplexity=true, single_turn_attack=true, require_human_strategies=true2025.12 | 39.5 | — | |
| SI-AttackSource=ICCV’252026.03 | 47.2 | — | |
| MMLSource=ACL’252026.03 | 60.4 | — | |
| J2 Attacklow_perplexity=true, single_turn_attack=false, require_human_strategies=true2025.12 | 60.5 | — | |
| TreeTeamingSource=/2026.03 | 61.6 | — | |
| ActorAttacklow_perplexity=true, single_turn_attack=false, require_human_strategies=false2025.12 | 66.5 | — | |
| Babel2026.05 | 75.67 | 40 | |
| BoN Attacklow_perplexity=false, single_turn_attack=true, require_human_strategies=false2025.12 | 78 | — | |
| Bijectionlow_perplexity=false, single_turn_attack=true, require_human_strategies=false2025.12 | 91.2 | — | |
| Jailbreak-Zerolow_perplexity=true, single_turn_attack=true, require_human_strategies=false2025.12 | 96 | — |