Agent Task Performance on AgentDojo Overall
69.79UtilityTask Shield
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Task ShieldModel=GPT-4o, Attack Type=Important Instructions2024.12 | 69.79 | 0.0207 | |
| Task ShieldModel=GPT-4o, Attack Type=Injecagent2024.12 | 69.48 | 0.0111 | |
| No DefenseModel=GPT-4o, Attack Type=Injecagent2024.12 | 68.52 | 0.0572 | |
| Task ShieldModel=GPT-4o, Attack Type=Ignore Previous2024.12 | 66.93 | 0.0048 | |
| No DefenseModel=GPT-4o, Attack Type=Ignore Previous2024.12 | 66.77 | 0.0541 | |
| No DefenseModel=GPT-4o, Attack Type=Important Instructions2024.12 | 50.08 | 0.4769 |