Agent Task Performance on AgentDojo Slack
67.62UtilityNo Defense
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| No DefenseModel=GPT-4o, Attack Type=Injecagent2024.12 | 67.62 | 1,333 | |
| Task ShieldModel=GPT-4o, Attack Type=Injecagent2024.12 | 66.67 | 95 | |
| Task ShieldModel=GPT-4o, Attack Type=Important Instructions2024.12 | 64.76 | 95 | |
| No DefenseModel=GPT-4o, Attack Type=Important Instructions2024.12 | 63.81 | 9,238 | |
| Task ShieldModel=GPT-4o, Attack Type=Ignore Previous2024.12 | 63.81 | 95 | |
| No DefenseModel=GPT-4o, Attack Type=Ignore Previous2024.12 | 61.9 | 2,095 |