Backdoor on attack prompts In-domain
100ASR (TA)RogueMerge
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| RogueMerge2026.06 | 100 | 100 | 100 | 100 | 100 | |
| LoBAM2026.06 | 13 | 26 | 45 | 6 | 22.5 | |
| Clean2026.06 | 0 | 0 | 0 | 0 | 0 | |
| MergeHijacking2026.06 | 0 | 0 | 0 | 0 | 0 | |
| SFT2026.06 | 0 | 0 | 0 | 0 | 0 |
| Method | Links | |||||
|---|---|---|---|---|---|---|
| RogueMerge2026.06 | 100 | 100 | 100 | 100 | 100 | |
| LoBAM2026.06 | 13 | 26 | 45 | 6 | 22.5 | |
| Clean2026.06 | 0 | 0 | 0 | 0 | 0 | |
| MergeHijacking2026.06 | 0 | 0 | 0 | 0 | 0 | |
| SFT2026.06 | 0 | 0 | 0 | 0 | 0 |