Prompt Injection on In-domain attack prompts
53ASR (TA)RogueMerge
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| RogueMerge2026.06 | 53 | 49 | 45 | 46 | 48.3 | |
| MergeHijacking*2026.06 | 46 | 37 | 46 | 44 | 43.3 | |
| SFT2026.06 | 46 | 40 | 42 | 43 | 42.8 | |
| LoBAM*2026.06 | 45 | 40 | 41 | 43 | 42.3 | |
| Clean2026.06 | 33 | 25 | 32 | 28 | 29.5 |