Prompt Injection Defense on Multi-Agent System (MAS) Topologies Average of Chain, Tree, Star, Random
94.26AccuracySAIGuard
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SAIGuardBackbone=GPT-4o-mini2026.06 | 94.26 | 3.36 | |
| SAIGuardBackbone=GPT-4o-mini2026.06 | 87.95 | 8.49 | |
| G-SafeguardBackbone=GPT-4o-mini2026.06 | 87.3 | 9.38 | |
| XG-GuardBackbone=GPT-4o-mini2026.06 | 86.95 | 9.06 | |
| BlindGuardBackbone=GPT-4o-mini2026.06 | 85.99 | 12.5 | |
| TAMBackbone=GPT-4o-mini2026.06 | 84.94 | 16.66 | |
| PREMBackbone=GPT-4o-mini2026.06 | 84.71 | 17 | |
| No DefenseBackbone=GPT-4o-mini2026.06 | 84.58 | 19.26 | |
| DominantBackbone=GPT-4o-mini2026.06 | 84.57 | 17.87 | |
| XG-GuardBackbone=GPT-4o-mini2026.06 | 78.44 | 38.99 | |
| G-SafeguardBackbone=GPT-4o-mini2026.06 | 77.77 | 39.65 | |
| BlindGuardBackbone=GPT-4o-mini2026.06 | 75.15 | 39.77 | |
| TAMBackbone=GPT-4o-mini2026.06 | 66.65 | 43.04 | |
| PREMBackbone=GPT-4o-mini2026.06 | 62.08 | 46.81 | |
| DominantBackbone=GPT-4o-mini2026.06 | 57.73 | 48.77 | |
| No DefenseBackbone=GPT-4o-mini2026.06 | 52.73 | 51.16 |