Jailbreak Defense on LLaMA3 8B Instruct
88GCG Success RateOriginal
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Original2026.01 | 88 | 87 | 98 | 41 | 2 | 63.2 | |
| DeRTa2026.01 | 97 | 89 | 100 | 98 | 32 | 83.2 | |
| HumorReject2026.01 | 98 | 99 | 99 | 100 | 98 | 98.8 | |
| DeepAug2026.01 | 99 | 40 | 100 | 59 | 3 | 60.2 | |
| CB2026.01 | 99 | 98 | 97 | 95 | 98 | 97.4 | |
| TraceRouter2026.01 | 99 | 100 | 100 | 100 | 99 | 99.6 |