Jailbreak Defense on AdvBench PAIR attack
98DSRSmoothLLM
Evaluation results
| Method | Links | |
|---|---|---|
| SmoothLLMTarget Model=Llama-2-13B, Attack Setting=w/o defense2024.02 | 98 | |
| BacktranslationTarget Model=Llama-2-13B, Attack Setting=w/o defense2024.02 | 98 | |
| BacktranslationTarget Model=Vicuna-13B, Attack Setting=w/o defense2024.02 | 94 | |
| BacktranslationTarget Model=Llama-2-13B, Attack Setting=w/ defense2024.02 | 94 | |
| ParaphrasingTarget Model=Llama-2-13B, Attack Setting=w/o defense2024.02 | 90 | |
| BacktranslationTarget Model=GPT-3.5-turbo, Attack Setting=w/o defense2024.02 | 88 | |
| Response checkTarget Model=Llama-2-13B, Attack Setting=w/o defense2024.02 | 82 | |
| SmoothLLMTarget Model=Llama-2-13B, Attack Setting=w/ defense2024.02 | 82 | |
| ParaphrasingTarget Model=Vicuna-13B, Attack Setting=w/o defense2024.02 | 80 | |
| ParaphrasingTarget Model=GPT-3.5-turbo, Attack Setting=w/o defense2024.02 | 78 | |
| SmoothLLMTarget Model=Vicuna-13B, Attack Setting=w/o defense2024.02 | 76 | |
| BacktranslationTarget Model=GPT-3.5-turbo, Attack Setting=w/ defense2024.02 | 76 | |
| SmoothLLMTarget Model=GPT-3.5-turbo, Attack Setting=w/o defense2024.02 | 70 | |
| BacktranslationTarget Model=GPT-3.5-turbo2024.02 | 70 | |
| Response checkTarget Model=Llama-2-13B, Attack Setting=w/ defense2024.02 | 68 | |
| Response checkTarget Model=GPT-3.5-turbo, Attack Setting=w/o defense2024.02 | 66 | |
| No defenseTarget Model=Llama-2-13B, Attack Setting=w/o defense2024.02 | 64 | |
| ParaphrasingTarget Model=GPT-3.5-turbo, Attack Setting=w/ defense2024.02 | 64 | |
| No defenseTarget Model=Llama-2-13B, Attack Setting=w/ defense2024.02 | 64 | |
| BacktranslationTarget Model=Vicuna-13B, Attack Setting=w/ defense2024.02 | 56 | |
| ParaphrasingTarget Model=Llama-2-13B, Attack Setting=w/ defense2024.02 | 54 | |
| Response checkTarget Model=GPT-3.5-turbo, Attack Setting=w/ defense2024.02 | 46 | |
| ParaphrasingTarget Model=GPT-3.5-turbo2024.02 | 38 | |
| No defenseTarget Model=GPT-3.5-turbo, Attack Setting=w/o defense2024.02 | 36 | |
| No defenseTarget Model=GPT-3.5-turbo, Attack Setting=w/ defense2024.02 | 36 | |
| Response checkTarget Model=Vicuna-13B, Attack Setting=w/o defense2024.02 | 32 | |
| Response checkTarget Model=GPT-3.5-turbo2024.02 | 30 | |
| SmoothLLMTarget Model=GPT-3.5-turbo, Attack Setting=w/ defense2024.02 | 28 | |
| SmoothLLMTarget Model=GPT-3.5-turbo2024.02 | 20 | |
| No defenseTarget Model=Vicuna-13B, Attack Setting=w/o defense2024.02 | 8 | |
| No defenseTarget Model=Vicuna-13B, Attack Setting=w/ defense2024.02 | 8 | |
| No defenseTarget Model=GPT-3.5-turbo2024.02 | 8 | |
| Response checkTarget Model=Vicuna-13B, Attack Setting=w/ defense2024.02 | 6 | |
| ParaphrasingTarget Model=Vicuna-13B, Attack Setting=w/ defense2024.02 | 4 | |
| SmoothLLMTarget Model=Vicuna-13B, Attack Setting=w/ defense2024.02 | 2 |