Dynamic Agent Security and Utility Evaluation on AgentDyn
52ASRNone
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| NoneModel=GPT-4o-mini2026.05 | 52 | 50 | 38 | |
| SpotlightingModel=GPT-4o-mini2026.05 | 49 | 44 | 37 | |
| NoneModel=GPT-4o2026.05 | 39 | 53 | 55 | |
| Repeat PromptModel=GPT-4o-mini2026.05 | 32 | 39 | 42 | |
| Repeat PromptModel=GPT-4o2026.05 | 31 | 63 | 56 | |
| SpotlightingModel=GPT-4o2026.05 | 28 | 55 | 52 | |
| PromptArmorModel=GPT-4o-mini2026.05 | 15 | 16 | 13 | |
| ProgentModel=GPT-4o-mini2026.05 | 10 | 6 | 4 | |
| SecAlignModel=GPT-4o-mini2026.05 | 9 | 55 | 53 | |
| SecAlignModel=GPT-4o2026.05 | 9 | 55 | 53 | |
| Tool FilterModel=GPT-4o2026.05 | 8 | 5 | 5 | |
| Tool FilterModel=GPT-4o-mini2026.05 | 6 | 3 | 6 | |
| DRIFTModel=GPT-4o-mini2026.05 | 3 | 18 | 21 | |
| Trans. DetectModel=GPT-4o-mini2026.05 | 2 | 1 | 1 | |
| AuthGraphModel=GPT-4o-mini2026.05 | 2 | 37 | 31 | |
| Trans. DetectModel=GPT-4o2026.05 | 2 | 3 | 1 | |
| PromptArmorModel=GPT-4o2026.05 | 2 | 4 | 2 | |
| ProgentModel=GPT-4o2026.05 | 2 | 7 | 6 | |
| AuthGraphModel=GPT-4o2026.05 | 2 | 51 | 47 | |
| DRIFTModel=GPT-4o2026.05 | 1 | 30 | 27 | |
| CaMeLModel=GPT-4o-mini2026.05 | 0 | 0 | 0 | |
| CaMeLModel=GPT-4o2026.05 | 0 | 0 | 0 |