LLM Red-teaming on Target Victim Model
2Unknown/Unsafe AttacksSFT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SFT2026.05 | 2 | 0.23 | |
| PPO2026.05 | 3 | 91.7 | |
| PPO + Curiosity2026.05 | 4 | 36.75 | |
| DPO2026.05 | 5.33 | 0.52 | |
| GFN2026.05 | 17.67 | 93.75 | |
| ICL2026.05 | 21 | 2.54 | |
| Rainbow Teaming2026.05 | 33 | 66.11 | |
| Jailbreak R1CoT=enabled2026.05 | 75.33 | 7.36 | |
| S-GFN2026.05 | 134 | 92.55 |