Robustness against Adaptive Attacks on AgentDojo 61 instances
0Static ASRPromptArmor
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| PromptArmorModel=Gemini-2.52026.03 | 0 | 29.51 | 21.5 | |
| MELONModel=Gemini-2.52026.03 | 0 | 81.97 | 10.5 | |
| AttriGuardModel=Gemini-2.52026.03 | 0 | 6.56 | 6.5 | |
| PIGuardModel=Llama3.32026.03 | 0 | 55.74 | 9.5 | |
| PromptArmorModel=Llama3.32026.03 | 0 | 40.98 | 65 | |
| AttriGuardModel=Llama3.32026.03 | 0 | 9.84 | 4.5 | |
| MetaSecAlignModel=Llama3.32026.03 | 1.64 | 24.59 | 42 | |
| PIGuardModel=Gemini-2.52026.03 | 4.92 | 52.46 | 4.5 |