Content Safety Detection on JailbreakBench
96.2F1 Scoreprompt–response verification framework
Evaluation results
| Method | Links | ||
|---|---|---|---|
| prompt–response verification framework2026.06 | 96.2 | 3.8 | |
| AutoDefense / SelfDefendType=Strongest Applicable Baseline2026.06 | 93.1 | — |