Prompt Injection Defense on AgentDojo No Attack
92.78Benign UtilityNo Defense
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| No DefenseBackbone Model=Gemini-3-Pro2026.02 | 92.78 | 1 | |
| Repeat PromptBackbone Model=Gemini-3-Pro2026.02 | 92.78 | 1.16 | |
| No DefenseBackbone Model=Claude-4-Sonnet2026.02 | 89.44 | 1 | |
| CausalArmorBackbone Model=Gemini-3-Pro2026.02 | 86.6 | 1.22 | |
| CausalArmorBackbone Model=Claude-4-Sonnet2026.02 | 82.47 | 1.27 | |
| DRIFTBackbone Model=Gemini-3-Pro2026.02 | 81.05 | 5.82 | |
| PromptArmorBackbone Model=Gemini-3-Pro2026.02 | 77.32 | 1.78 | |
| MelonBackbone Model=Claude-4-Sonnet2026.02 | 74.29 | 2.31 | |
| PromptArmorBackbone Model=Claude-4-Sonnet2026.02 | 74.23 | 2.2 | |
| DRIFTBackbone Model=Claude-4-Sonnet2026.02 | 72.27 | 5.43 | |
| Repeat PromptBackbone Model=Claude-4-Sonnet2026.02 | 67.01 | 1.2 | |
| Repeat PromptBackbone Model=Gemini-2.5-Flash2026.02 | 64.95 | 1.15 | |
| PiGuardBackbone Model=Gemini-3-Pro2026.02 | 64.95 | 1.06 | |
| PiGuardBackbone Model=Claude-4-Sonnet2026.02 | 62.89 | 1.15 | |
| DeBERTa DetectorBackbone Model=Gemini-3-Pro2026.02 | 58.76 | 1.04 | |
| CausalArmorBackbone Model=Gemini-2.5-Flash2026.02 | 56.26 | 1.3 | |
| DeBERTa DetectorBackbone Model=Claude-4-Sonnet2026.02 | 55.67 | 1.1 | |
| No DefenseBackbone Model=Gemini-2.5-Flash2026.02 | 53.61 | 1 | |
| DRIFTBackbone Model=Gemini-2.5-Flash2026.02 | 50.58 | 4.67 | |
| MelonBackbone Model=Gemini-2.5-Flash2026.02 | 49.48 | 2.21 | |
| PromptArmorBackbone Model=Gemini-2.5-Flash2026.02 | 41.24 | 2.44 | |
| PiGuardBackbone Model=Gemini-2.5-Flash2026.02 | 32.99 | 1.22 | |
| DeBERTa DetectorBackbone Model=Gemini-2.5-Flash2026.02 | 30.93 | 1.16 |