Prompt Injection on Attack prompts Out-of-domain
47ASR (TA)RogueMerge
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| RogueMerge2026.06 | 47 | 46.5 | 48.5 | 47 | 47.3 | |
| LoBAM*2026.06 | 46 | 43.5 | 45.5 | 42.5 | 44.4 | |
| MergeHijacking*2026.06 | 43.5 | 40.5 | 42.5 | 41.5 | 42 | |
| SFT2026.06 | 42 | 36.5 | 41.5 | 36 | 39 | |
| Clean2026.06 | 34 | 35 | 35 | 34.5 | 34.6 |