Agentic Security and Utility Evaluation on AgentDojo
0ASRCaMeL
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| CaMeLModel=GPT-4o-mini2026.05 | 0 | 48 | 32 | |
| CaMeLModel=GPT-4o2026.05 | 0 | 53 | 41 | |
| AuthGraphModel=GPT-4o-mini2026.05 | 1 | 69 | 44 | |
| AuthGraphModel=GPT-4o2026.05 | 1 | 76 | 58 | |
| SecAlignModel=GPT-4o-mini2026.05 | 2 | 76 | 74 | |
| ProgentModel=GPT-4o-mini2026.05 | 2 | 64 | 3 | |
| SecAlignModel=GPT-4o2026.05 | 2 | 76 | 74 | |
| ProgentModel=GPT-4o2026.05 | 2 | 76 | 61 | |
| DRIFTModel=GPT-4o2026.05 | 2 | 70 | 62 | |
| Tool FilterModel=GPT-4o-mini2026.05 | 3 | 68 | 37 | |
| DRIFTModel=GPT-4o-mini2026.05 | 3 | 52 | 45 | |
| Tool FilterModel=GPT-4o2026.05 | 6 | 66 | 61 | |
| PromptArmorModel=GPT-4o-mini2026.05 | 7 | 65 | 15 | |
| Trans. DetectModel=GPT-4o-mini2026.05 | 8 | 43 | 3 | |
| Trans. DetectModel=GPT-4o2026.05 | 8 | 37 | 19 | |
| PromptArmorModel=GPT-4o2026.05 | 8 | 71 | 21 | |
| Repeat PromptModel=GPT-4o-mini2026.05 | 11 | 68 | 57 | |
| SpotlightingModel=GPT-4o-mini2026.05 | 14 | 65 | 48 | |
| NoneModel=GPT-4o-mini2026.05 | 17 | 73 | 43 | |
| SpotlightingModel=GPT-4o2026.05 | 23 | 73 | 62 | |
| Repeat PromptModel=GPT-4o2026.05 | 25 | 84 | 69 | |
| NoneModel=GPT-4o2026.05 | 40 | 79 | 54 |