Directional Jailbreak Detection on WildJailbreak (Safe) × StrongREJECT (Harmful)
98.1AUROC (Mono)Qwen3-8B
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Qwen3-8BProbe Layer=L25, Relative Depth=~69%, Total Layers=36, Direction=downward2026.06 | 98.1 | 65.6 | 65.9 | |
| Llama-3.1-8BProbe Layer=L22, Relative Depth=~69%, Total Layers=32, Direction=downward2026.06 | 93.6 | 81.6 | 81.4 | |
| Gemma-7bProbe Layer=L19, Relative Depth=~68%, Total Layers=28, Direction=upward2026.06 | 80.1 | 75.2 | 75.5 |