Multi-turn Jailbreak Evaluation on MHJ LLaMA-3.1-70B (test)
95.2ASR (%)HMNS
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| HMNSGrading=GPT-4o-graded2026.04 | 95.2 | 2 | |
| ArrAttackGrading=GPT-4o-graded2026.04 | 85.3 | 7.2 | |
| TempestGrading=GPT-4o-graded2026.04 | 81.1 | 9.1 | |
| AutoDANGrading=GPT-4o-graded2026.04 | 66 | 11.8 | |
| Defense-onlyGrading=GPT-4o-graded2026.04 | 3.2 | — |