Jailbreak Attack on Gemma 4B 3
66NRSWhisper
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SWhisperSurrogate Model=Gemma-3-4B2026.03 | 66 | 42.3 | |
| SWhisperSurrogate Model=Qwen2.5-7B2026.03 | 64 | 37.5 | |
| SWhisperSurrogate Model=LLaMA-3.1-8B2026.03 | 62 | 43.5 | |
| PAIRSurrogate Model=Gemma-3-4B2026.03 | 58 | 35 | |
| Auto-DANSurrogate Model=Mistral-7B2026.03 | 54 | 36 | |
| Auto-DANSurrogate Model=Qwen2.5-7B2026.03 | 52 | 35.8 | |
| Auto-DANSurrogate Model=Gemma-3-4B2026.03 | 50 | 28 | |
| SWhisperSurrogate Model=Mistral-7B2026.03 | 50 | 35 | |
| PAIRSurrogate Model=Mistral-7B2026.03 | 45 | 32.5 | |
| PAIRSurrogate Model=Qwen2.5-7B2026.03 | 44 | 30.3 | |
| Auto-DANSurrogate Model=LLaMA-3.1-8B2026.03 | 40 | 19.3 | |
| IGCGSurrogate Model=Gemma-3-4B2026.03 | 32 | 11.8 | |
| PAIRSurrogate Model=LLaMA-3.1-8B2026.03 | 32 | 21.3 | |
| GCGSurrogate Model=Qwen2.5-7B2026.03 | 24 | 10 | |
| IGCGSurrogate Model=LLaMA-3.1-8B2026.03 | 24 | 9.3 | |
| GCGSurrogate Model=Mistral-7B2026.03 | 22 | 8 | |
| IGCGSurrogate Model=Qwen2.5-7B2026.03 | 18 | 7.5 | |
| IGCGSurrogate Model=Mistral-7B2026.03 | 18 | 2 | |
| GCGSurrogate Model=Gemma-3-4B2026.03 | 16 | 9.3 | |
| GCGSurrogate Model=LLaMA-3.1-8B2026.03 | 10 | 4.3 |