List your company

Jailbreak Attack Safety Evaluation on GPT-as-a-Judge

89ASRPCSA
7.8828.945071.06Apr 6, 2026

Evaluation results

MethodLinks
PCSATarget Model=Llama-3.1-8B2026.0489
PCSATarget Model=Psycho-8B2026.0488
PCSATarget Model=Qwen-2-7B2026.0484
PCSATarget Model=Qwen-2.5-72B2026.0482
PCSATarget Model=Casper-7B2026.0481
PCSATarget Model=GPT-3.52026.0479
PCSATarget Model=Llama-3.1-70B2026.0478
PCSATarget Model=GPT-4o-mini2026.0471
AMATarget Model=Llama-3.1-70B2026.0466
AMATarget Model=Llama-3.1-8B2026.0464
AMATarget Model=Qwen-2.5-72B2026.0464
AMATarget Model=Psycho-8B2026.0459
CrescendoTarget Model=GPT-3.52026.0453
CrescendoTarget Model=Llama-3.1-8B2026.0452
CrescendoTarget Model=Llama-3.1-70B2026.0451
AMATarget Model=Qwen-2-7B2026.0448
CrescendoTarget Model=Qwen-2.5-72B2026.0448
CrescendoTarget Model=Casper-7B2026.0447
CrescendoTarget Model=GPT-4o-mini2026.0444
CrescendoTarget Model=Psycho-8B2026.0444
Actor-AttackTarget Model=GPT-3.52026.0442
Actor-AttackTarget Model=Casper-7B2026.0438
Actor-AttackTarget Model=Llama-3.1-70B2026.0434
Actor-AttackTarget Model=Qwen-2-7B2026.0433
Actor-AttackTarget Model=GPT-4o-mini2026.0432
CoA (new)Target Model=Casper-7B2026.0430
Actor-AttackTarget Model=Psycho-8B2026.0429
AMATarget Model=GPT-4o-mini2026.0427
CoA (new)Target Model=GPT-3.52026.0425
Actor-AttackTarget Model=Qwen-2.5-72B2026.0425
CoA (new)Target Model=Qwen-2-7B2026.0422
CrescendoTarget Model=Qwen-2-7B2026.0422
CoA (new)Target Model=GPT-4o-mini2026.0421
CoA (new)Target Model=Psycho-8B2026.0421
CoA (new)Target Model=Qwen-2.5-72B2026.0421
CoA (new)Target Model=Llama-3.1-8B2026.0419
AMATarget Model=GPT-3.52026.0419
CoA (new)Target Model=Llama-3.1-70B2026.0418
Actor-AttackTarget Model=Llama-3.1-8B2026.0415
AMATarget Model=Casper-7B2026.0411