Jailbreak Attack Safety Evaluation on GPT-as-a-Judge
89ASRPCSA
Evaluation Results
| Method | Links | |
|---|---|---|
| PCSATarget Model=Llama-3.1-8B2026.04 | 89 | |
| PCSATarget Model=Psycho-8B2026.04 | 88 | |
| PCSATarget Model=Qwen-2-7B2026.04 | 84 | |
| PCSATarget Model=Qwen-2.5-72B2026.04 | 82 | |
| PCSATarget Model=Casper-7B2026.04 | 81 | |
| PCSATarget Model=GPT-3.52026.04 | 79 | |
| PCSATarget Model=Llama-3.1-70B2026.04 | 78 | |
| PCSATarget Model=GPT-4o-mini2026.04 | 71 | |
| AMATarget Model=Llama-3.1-70B2026.04 | 66 | |
| AMATarget Model=Llama-3.1-8B2026.04 | 64 | |
| AMATarget Model=Qwen-2.5-72B2026.04 | 64 | |
| AMATarget Model=Psycho-8B2026.04 | 59 | |
| CrescendoTarget Model=GPT-3.52026.04 | 53 | |
| CrescendoTarget Model=Llama-3.1-8B2026.04 | 52 | |
| CrescendoTarget Model=Llama-3.1-70B2026.04 | 51 | |
| AMATarget Model=Qwen-2-7B2026.04 | 48 | |
| CrescendoTarget Model=Qwen-2.5-72B2026.04 | 48 | |
| CrescendoTarget Model=Casper-7B2026.04 | 47 | |
| CrescendoTarget Model=GPT-4o-mini2026.04 | 44 | |
| CrescendoTarget Model=Psycho-8B2026.04 | 44 | |
| Actor-AttackTarget Model=GPT-3.52026.04 | 42 | |
| Actor-AttackTarget Model=Casper-7B2026.04 | 38 | |
| Actor-AttackTarget Model=Llama-3.1-70B2026.04 | 34 | |
| Actor-AttackTarget Model=Qwen-2-7B2026.04 | 33 | |
| Actor-AttackTarget Model=GPT-4o-mini2026.04 | 32 | |
| CoA (new)Target Model=Casper-7B2026.04 | 30 | |
| Actor-AttackTarget Model=Psycho-8B2026.04 | 29 | |
| AMATarget Model=GPT-4o-mini2026.04 | 27 | |
| CoA (new)Target Model=GPT-3.52026.04 | 25 | |
| Actor-AttackTarget Model=Qwen-2.5-72B2026.04 | 25 | |
| CoA (new)Target Model=Qwen-2-7B2026.04 | 22 | |
| CrescendoTarget Model=Qwen-2-7B2026.04 | 22 | |
| CoA (new)Target Model=GPT-4o-mini2026.04 | 21 | |
| CoA (new)Target Model=Psycho-8B2026.04 | 21 | |
| CoA (new)Target Model=Qwen-2.5-72B2026.04 | 21 | |
| CoA (new)Target Model=Llama-3.1-8B2026.04 | 19 | |
| AMATarget Model=GPT-3.52026.04 | 19 | |
| CoA (new)Target Model=Llama-3.1-70B2026.04 | 18 | |
| Actor-AttackTarget Model=Llama-3.1-8B2026.04 | 15 | |
| AMATarget Model=Casper-7B2026.04 | 11 |