LLM Agent Defense Evaluation on AgentLure (320 samples)
1.6ASRMELON
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| MELONDefense Category=Execution-level defenses2026.05 | 1.6 | 7.5 | 65 | 10.9 | 35 | 1.27 | 64 | |
| ARGUSDefense Category=Execution-level defenses2026.05 | 3.8 | 7.5 | 87.5 | 34.1 | 7.5 | 1.24 | 84.2 | |
| DRIFTDefense Category=Execution-level defenses2026.05 | 8.1 | 25 | 80 | 35 | 7.5 | 2.88 | 73.5 | |
| Tool FilterDefense Category=Execution-level defenses2026.05 | 8.8 | 17.5 | 65 | 35.3 | 25 | 0.81 | 59.3 | |
| ACEDefense Category=Execution-level defenses2026.05 | 11.6 | 25 | 80 | 35.3 | 12.5 | 1.26 | 70.7 | |
| InjecGuardDefense Category=Execution-level defenses2026.05 | 12.8 | 35 | 90 | 28.1 | 20 | 1.56 | 78.5 | |
| SandwichDefense Category=Text-level defenses2026.05 | 18.8 | 47.5 | 75 | 22.8 | — | 14.11 | 60.9 | |
| No DefenseDefense Category=Baseline2026.05 | 28.8 | 55 | 92.5 | 38.4 | — | 1 | 65.9 | |
| InstructionalDefense Category=Text-level defenses2026.05 | 33.1 | 77.5 | 90 | 40 | — | 0.95 | 60.2 | |
| DelimitersDefense Category=Text-level defenses2026.05 | 34.7 | 67.5 | 92.5 | 40.6 | — | 0.96 | 60.4 |