Jailbreak Defense on Qwen2.5-7B (Adaptive AutoDAN-T Attack)
30ASREmbedding Disruption Method
Evaluation Results
| Method | Links | |
|---|---|---|
| Embedding Disruption MethodTarget Model=Qwen2.5-7B, Attack=Adaptive AutoDAN-T2026.05 | 30 | |
| Erase-and-CheckTarget Model=Qwen2.5-7B, Attack=Adaptive AutoDAN-T2026.05 | 41 | |
| RESTATarget Model=Qwen2.5-7B, Attack=Adaptive AutoDAN-T2026.05 | 46 | |
| SmoothLLMTarget Model=Qwen2.5-7B, Attack=Adaptive AutoDAN-T2026.05 | 49 | |
| VanillaTarget Model=Qwen2.5-7B, Attack=Adaptive AutoDAN-T2026.05 | 60 | |
| Perplexity FilterTarget Model=Qwen2.5-7B, Attack=Adaptive AutoDAN-T2026.05 | 60 |