Jailbreaking on Attack Prompts Out-of-domain
67.7ASR (TA)RogueMerge
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| RogueMerge2026.06 | 67.7 | 72.3 | 61.3 | 71.7 | 68.3 | |
| LoBAM*2026.06 | 55.7 | 54.3 | 49.3 | 54.7 | 53.5 | |
| SFT2026.06 | 41.7 | 43 | 32.7 | 42.3 | 39.9 | |
| MergeHijacking*2026.06 | 40.3 | 42.3 | 35 | 45.7 | 40.8 | |
| Clean2026.06 | 18.3 | 36.3 | 20.3 | 33 | 27 |