| Task Name | Dataset Name | SOTA Result | Trend | |
|---|---|---|---|---|
| Human-crafted jailbreak defense | mhj | Attack Success Rate (ASR)14 | 7 | |
| Multi-turn Jailbreak Evaluation | MHJ LLaMA-3.1-70B (test) | ASR (%)95.2 | 5 | |
| Multi-turn Jailbreak Evaluation | MHJ Phi-3-Medium-14B (test) | ASR91.6 | 5 | |
| Adversarial Attack | MHJ | Attack Success Rate76.72 | 4 |