Prompt Injection Discovery on AgentDyn
78Source DiversityPI-Hunter
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| PI-HunterModel=Gemini-3.1-pro2026.06 | 78 | 81 | — | — | |
| PI-HunterModel=Claude-4.6-sonnet2026.06 | 70 | 74 | — | — | |
| PI-HunterModel=GPT-5.4-mini2026.06 | 67 | 72 | — | — | |
| BaselineModel=Gemini-3.1-pro2026.06 | 13 | 37 | — | — | |
| BaselineModel=GPT-5.4-mini2026.06 | 10 | 25 | — | — | |
| BaselineModel=Claude-4.6-sonnet2026.06 | 10 | 27 | — | — | |
| PI-HunterDefense=None2026.06 | 0.526 | 0.551 | 0.417 | 0.514 | |
| PI-HunterDefense=Spotlight2026.06 | 0.473 | 0.441 | 0.475 | 0.509 | |
| PI-HunterDefense=MELON2026.06 | 0.44 | 0.386 | 0.349 | 0.467 | |
| PI-HunterDefense=PIGuard2026.06 | 0.318 | 0.151 | 0.252 | 0.141 | |
| BaselineDefense=None2026.06 | 0.075 | 0.131 | 0.216 | 0.091 | |
| BaselineDefense=Spotlight2026.06 | 0 | 0 | 0.273 | 0.181 | |
| BaselineDefense=MELON2026.06 | 0 | 0 | 0.181 | 0.166 | |
| BaselineDefense=PIGuard2026.06 | 0 | 0 | 0.109 | 0.083 |