| Task Name | Dataset Name | SOTA Result | Trend | |
|---|---|---|---|---|
| Red-Teaming (Attack Success Rate) | DANGEROUSQA | ASR0 | 30 | |
| Safety Evaluation | DANGEROUSQA Llama-2 base | Chinese Safety Score15.3 | 8 | |
| Jailbreak Attack | DangerousQA | Harmful Rate1.01 | 6 | |
| Safety Evaluation | DangerousQA (test) | Harmful Rate0.0122 | 3 |