Jailbreak Attack on Harmbench Malicious (full)
1.13Harmful ScoreIrrelevant Fine-tuning
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Irrelevant Fine-tuningModel=Llama2-7b2026.01 | 1.13 | 29.52 | |
| Irrelevant Fine-tuningModel=GPT-3.5-Turbo2026.01 | 1.74 | 33.18 | |
| Irrelevant Fine-tuningModel=Qwen2.5-7b2026.01 | 2.27 | 36.1 | |
| Irrelevant Fine-tuningModel=Llama3-70b2026.01 | 2.7 | 42.15 | |
| Irrelevant Fine-tuningModel=GPT-4o-mini2026.01 | 2.85 | 37.01 | |
| Irrelevant Fine-tuningModel=Qwen2.5-3b2026.01 | 2.89 | 41.05 | |
| Irrelevant Fine-tuningModel=Llama3-8b2026.01 | 3.12 | 44.89 | |
| Malicious Fine-tuningModel=Llama3-8b2026.01 | 4.38 | 0.915 | |
| Malicious Fine-tuningModel=Qwen2.5-7b2026.01 | 4.79 | 0.9843 | |
| Malicious Fine-tuningModel=GPT-3.5-Turbo2026.01 | 4.8 | 0.9799 | |
| Malicious Fine-tuningModel=Llama2-7b2026.01 | 4.82 | 0.9614 | |
| Malicious Fine-tuningModel=GPT-4o-mini2026.01 | 4.85 | 0.9622 | |
| Malicious Fine-tuningModel=Qwen2.5-3b2026.01 | 4.87 | 0.9744 | |
| Malicious Fine-tuningModel=Llama3-70b2026.01 | 4.91 | 0.9588 |