Directional Jailbreak Detection on UltraChat (Safe) × StrongREJECT (Harmful)
0.797AUROC (mono)Qwen3-8B
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Qwen3-8BProbe Layer=L25, Relative Depth=~69%, Total Layers=36, Direction=downward2026.06 | 0.797 | 0.541 | 0.543 | |
| Llama-3.1-8BProbe Layer=L22, Relative Depth=~69%, Total Layers=32, Direction=downward2026.06 | 0.79 | 0.713 | 0.715 | |
| Gemma-7bProbe Layer=L19, Relative Depth=~68%, Total Layers=28, Direction=upward2026.06 | 0.704 | 0.643 | 0.643 |