Secure LLM Agent Task Completion on AgentDojo
76.3Benign UtilityProgent
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| ProgentBase Model=GPT-4o2025.06 | 76.3 | 61.2 | 2.2 | |
| DRIFTBase Model=GPT-4o2025.06 | 71.61 | 62 | 1.66 | |
| PlainHarness=gpt-4o-mini2026.06 | 70.1 | 47.22 | 31.48 | |
| IPIGuardHarness=gpt-4o-mini2026.06 | 64.95 | 51.99 | 15.26 | |
| FarameshHarness=gpt-4o-mini2026.06 | 60.82 | 47.69 | 22.89 | |
| SecureClawHarness=gpt-4o-mini2026.06 | 60.82 | 56.6 | 0.64 | |
| DRIFTHarness=gpt-4o-mini2026.06 | 59.79 | 52.78 | 2.38 | |
| DRIFTBase Model=GPT-4o-mini2025.06 | 57.29 | 50.93 | 1.35 | |
| ProgentBase Model=GPT-4o-mini2025.06 | 54.66 | 45.58 | 9.39 |