Prompt Injection Defense on IPI
94.4AccuracySCOUT
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SCOUTJudge=GPT-5.12026.05 | 94.4 | 8.6 | 97.4 | 1,147 | |
| Always-DLLMJudge=GPT-5.12026.05 | 94 | 12 | 100 | 1,651 | |
| Always-DLLMJudge=GPT-4o2026.05 | 90.3 | 14.4 | 95 | 1,504 | |
| SCOUTJudge=GPT-4o2026.05 | 90.1 | 12.8 | 93 | 915 |