Jailbreak Attack on HarmfulQA
56JADESiMIST
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| iMISTTarget Model=DeepSeek-V32026.01 | 56 | 60 | 2 | |
| iMISTTarget Model=Qwen3-32B2026.01 | 48 | 58 | 4 | |
| FlipAttackTarget Model=DeepSeek-V32026.01 | 41 | 57 | 2 | |
| iMISTTarget Model=GPT-OSS-120B2026.01 | 41 | 38 | 32 | |
| TAPTarget Model=DeepSeek-V32026.01 | 37 | 52 | 8 | |
| TAPTarget Model=Qwen3-32B2026.01 | 36 | 41 | 14 | |
| PAIRTarget Model=Qwen3-32B2026.01 | 35 | 22 | 18 | |
| DRATarget Model=DeepSeek-V32026.01 | 31 | 53 | 18 | |
| FlipAttackTarget Model=Qwen3-32B2026.01 | 30 | 48 | 20 | |
| PAIRTarget Model=DeepSeek-V32026.01 | 29 | 7 | 4 | |
| DRATarget Model=Qwen3-32B2026.01 | 28 | 25 | 12 | |
| ArtPromptTarget Model=DeepSeek-V32026.01 | 24 | 31 | 10 | |
| PASSTarget Model=DeepSeek-V32026.01 | 24 | 27 | 14 | |
| PASSTarget Model=Qwen3-32B2026.01 | 24 | 18 | 8 | |
| WordGameTarget Model=Qwen3-32B2026.01 | 22 | 40 | 20 | |
| WordGameTarget Model=GPT-OSS-120B2026.01 | 22 | 24 | 46 | |
| ArtPromptTarget Model=Qwen3-32B2026.01 | 20 | 19 | 16 | |
| WordGameTarget Model=DeepSeek-V32026.01 | 20 | 24 | 2 | |
| SATATarget Model=Qwen3-32B2026.01 | 20 | 38 | 22 | |
| SATATarget Model=DeepSeek-V32026.01 | 18 | 34 | 30 | |
| RL-JackTarget Model=DeepSeek-V32026.01 | 17 | 28 | 24 | |
| RL-JackTarget Model=Qwen3-32B2026.01 | 14 | 7 | 40 | |
| BaselineTarget Model=Qwen3-32B2026.01 | 13 | 7 | 42 | |
| PASSTarget Model=GPT-OSS-120B2026.01 | 10 | 16 | 64 | |
| DRATarget Model=GPT-OSS-120B2026.01 | 8 | 18 | 78 | |
| SATATarget Model=GPT-OSS-120B2026.01 | 8 | 12 | 76 | |
| BaselineTarget Model=DeepSeek-V32026.01 | 6 | 4 | 70 | |
| ArtPromptTarget Model=GPT-OSS-120B2026.01 | 6 | 17 | 72 | |
| PAIRTarget Model=GPT-OSS-120B2026.01 | 5 | 4 | 58 | |
| FlipAttackTarget Model=GPT-OSS-120B2026.01 | 4 | 8 | 88 | |
| TAPTarget Model=GPT-OSS-120B2026.01 | 2 | 10 | 84 | |
| RL-JackTarget Model=GPT-OSS-120B2026.01 | 2 | 4 | 92 | |
| BaselineTarget Model=GPT-OSS-120B2026.01 | 0 | 0 | 100 |