Tool Attack Defense on Multi-Agent System (MAS) Topologies Average of Chain, Tree, Star, Random
89.83AccuracySAIGuard
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SAIGuardBackbone=GPT-4o-mini2026.06 | 89.83 | 5.37 | |
| SAIGuardBackbone=GPT-4o-mini2026.06 | 88.16 | 9.62 | |
| XG-GuardBackbone=GPT-4o-mini2026.06 | 80.13 | 16.51 | |
| G-SafeguardBackbone=GPT-4o-mini2026.06 | 79.52 | 17.36 | |
| BlindGuardBackbone=GPT-4o-mini2026.06 | 76.07 | 20.29 | |
| TAMBackbone=GPT-4o-mini2026.06 | 70.35 | 27.27 | |
| PREMBackbone=GPT-4o-mini2026.06 | 70.12 | 26.51 | |
| XG-GuardBackbone=GPT-4o-mini2026.06 | 68.15 | 17.51 | |
| DominantBackbone=GPT-4o-mini2026.06 | 68.02 | 30.68 | |
| G-SafeguardBackbone=GPT-4o-mini2026.06 | 66.91 | 18.09 | |
| No DefenseBackbone=GPT-4o-mini2026.06 | 63.78 | 34.62 | |
| BlindGuardBackbone=GPT-4o-mini2026.06 | 55 | 28.1 | |
| PREMBackbone=GPT-4o-mini2026.06 | 50.26 | 38.8 | |
| TAMBackbone=GPT-4o-mini2026.06 | 46.65 | 45.07 | |
| DominantBackbone=GPT-4o-mini2026.06 | 46.64 | 52.3 | |
| No DefenseBackbone=GPT-4o-mini2026.06 | 40.15 | 54.06 |