List your company

Prompt Injection on Attack prompts Out-of-domain

47ASR (TA)RogueMerge
33.4836.9940.544.01Jun 2, 2026

Evaluation results

MethodLinks
RogueMerge2026.064746.548.54747.3
LoBAM*2026.064643.545.542.544.4
MergeHijacking*2026.0643.540.542.541.542
SFT2026.064236.541.53639
Clean2026.0634353534.534.6