Jailbreaking on MaliciousInstruct
4.8BERT ScoreLINT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| LINTAPI Compatibility=false2025.05 | 4.8 | 2.36 | 0.4 | |
| FLIPAPI Compatibility=true2025.05 | 4.42 | 2.03 | 0.46 | |
| NaiveAPI Compatibility=true2025.05 | 4.07 | 1.54 | 0.17 | |
| JULITarget Model=Llama2-7B-CHAT, Defense=DEEPALIGN, Evaluation Setting=open-source setting2025.05 | 4.01 | 3.84 | 3.19 | |
| GCGAPI Compatibility=false2025.05 | 3.63 | 1.74 | 0.37 | |
| EDAPI Compatibility=false2025.05 | 3.55 | 3.66 | 2.74 | |
| JULIAPI Compatibility=false2025.05 | 3.39 | 2.52 | 1.72 | |
| JULI (API)API Compatibility=true2025.05 | 2.86 | 1.95 | 0.71 | |
| WTSAPI Compatibility=false2025.05 | 2.04 | 1.33 | 0.21 | |
| BaseAPI Compatibility=true2025.05 | 1.98 | 1.16 | 0.04 | |
| NaiveTarget Model=Llama2-7B-CHAT, Defense=DEEPALIGN, Evaluation Setting=open-source setting2025.05 | 1.21 | 1 | 0 |