LLM Red-teaming on S-GFN-defended Target Model
7.33Unsuccessful Attack Rate (UA)S-GFN
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| S-GFN2026.05 | 7.33 | 75 | |
| Jailbreak R1CoT=enabled2026.05 | 5.67 | 55 | |
| Rainbow Teaming2026.05 | 2.33 | 23 | |
| ICL2026.05 | 1 | 10 | |
| GFN2026.05 | 0.33 | 3 | |
| SFT2026.05 | 0 | 0 | |
| DPO2026.05 | 0 | 0 | |
| PPO2026.05 | 0 | 0 | |
| PPO + Curiosity2026.05 | 0 | 0 |