Prompt Injection Exposure on AgentDojo
48Source RecallPI-Hunter
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| PI-HunterDefense=None2026.06 | 48 | 42.5 | 48.6 | 67.9 | |
| PI-HunterDefense=Spotlight2026.06 | 45.6 | 38.2 | 39 | 47.5 | |
| PI-HunterDefense=MELON2026.06 | 41.7 | 36.5 | 40.3 | 55 | |
| BaselineDefense=None2026.06 | 34.8 | 7.8 | 27.5 | 2.5 | |
| BaselineDefense=Spotlight2026.06 | 30 | 0 | 22 | 0 | |
| PI-HunterDefense=PIGuard2026.06 | 29.6 | 31.8 | 19.4 | 27.2 | |
| BaselineDefense=MELON2026.06 | 28.9 | 0 | 42.7 | 0 | |
| BaselineDefense=PIGuard2026.06 | 15.4 | 0 | 0 | 0 |