Jailbreak Attack on 50 prompts Abliterated victim
12Positive ExitsPAIR
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| PAIRDefense=LLaMA-Guard-3-8B, Total=502026.06 | 12 | 38 | 0 | 7 | 5 | 10 | 41.4 | |
| LLaMA-Guard-3-8BVictim=Abliterated, Attacker=GPTFuzz, Total prompts=502026.06 | 22 | 28 | 0 | 12 | 10 | 20 | 34 | |
| PAIRDefense=LLaMA-Guard-3-8B + CMPE, Total=502026.06 | 47 | 3 | 47 | 0 | 0 | 0 | 15.8 | |
| LLaMA-Guard-3-8B + CMPEVictim=Abliterated, Attacker=GPTFuzz, Total prompts=502026.06 | 48 | 2 | 42 | 5 | 1 | 2 | 6.4 |