Adversarial Detection on AutoDAN
100DSRSALO
Evaluation Results
| Method | Links | |
|---|---|---|
| SALOTarget Model=Llama-3.1-8B-Instruct, Params=<20M, Grad=false2026.05 | 100 | |
| Linear ProbeTarget Model=Qwen2.5-7B-Instruct, Params=<1M, Grad=false2026.05 | 99.2 | |
| Linear ProbeTarget Model=Llama-3.1-8B-Instruct, Params=<1M, Grad=false2026.05 | 98.8 | |
| SALOTarget Model=Qwen2.5-7B-Instruct, Params=<20M, Grad=false2026.05 | 98.8 | |
| SALOTarget Model=Mistral-7B-Instruct-v0.3, Params=<20M, Grad=false2026.05 | 98.8 | |
| Smooth LLMTarget Model=Qwen2.5-7B-Instruct, Grad=false2026.05 | 82 | |
| GradSafeTarget Model=Llama-3.1-8B-Instruct, Grad=true2026.05 | 76 | |
| Smooth LLMTarget Model=Llama-3.1-8B-Instruct, Grad=false2026.05 | 72 | |
| PPL FilterTarget Model=Llama-3.1-8B-Instruct, Grad=false2026.05 | 67.2 | |
| No Defense (1-ASR)Target Model=Qwen2.5-7B-Instruct2026.05 | 55.6 | |
| No Defense (1-ASR)Target Model=Llama-3.1-8B-Instruct2026.05 | 54.4 | |
| PPL FilterTarget Model=Qwen2.5-7B-Instruct, Grad=false2026.05 | 32.8 | |
| Linear ProbeTarget Model=Mistral-7B-Instruct-v0.3, Params=<1M, Grad=false2026.05 | 32 | |
| GradSafeTarget Model=Qwen2.5-7B-Instruct, Grad=true2026.05 | 10 | |
| No Defense (1-ASR)Target Model=Mistral-7B-Instruct-v0.32026.05 | 3.2 | |
| GradSafeTarget Model=Mistral-7B-Instruct-v0.3, Grad=true2026.05 | 2 | |
| Smooth LLMTarget Model=Mistral-7B-Instruct-v0.3, Grad=false2026.05 | 1.6 | |
| PPL FilterTarget Model=Mistral-7B-Instruct-v0.3, Grad=false2026.05 | 0 |