Jailbreak Attack on Model-based ASR evaluation prompt set
96ASR (GPT-4o)HILL
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| HILLJudge Model=GPT-5, Attack Type=Single-turn, API Settings=Default2025.09 | 96 | 96 | 96 | 86 | 80 | |
| PAPsJudge Model=GPT-5, Attack Type=Single-turn2025.09 | 90 | 82 | 54 | 86 | 82 | |
| PAIRJudge Model=GPT-5, Attack Type=Single-turn2025.09 | 22 | 38 | 8 | 38 | 40 | |
| DrAttackJudge Model=GPT-5, Attack Type=Single-turn2025.09 | 22 | 20 | 2 | 54 | 72 | |
| MasterKeyJudge Model=GPT-5, Attack Type=Single-turn2025.09 | 0 | 10 | 2 | 0 | 2 |