Malicious Advice Defense on PoisonRAG
5ASRPropGuard
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| PropGuardArchitecture=Star2026.05 | 5 | — | — | — | — | — | — | 95 | |
| PropGuardArchitecture=Tree2026.05 | 5.7 | — | — | — | — | — | — | 96.7 | |
| PropGuardArchitecture=Chain2026.05 | 6.7 | — | — | — | — | — | — | 93.3 | |
| INFA-GuardArchitecture=Tree2026.05 | 7.3 | — | — | — | — | — | — | 93.3 | |
| PropGuardArchitecture=Random2026.05 | 7.7 | — | — | — | — | — | — | 93.3 | |
| INFA-GuardArchitecture=Star2026.05 | 8 | — | — | — | — | — | — | 91.7 | |
| INFA-GuardArchitecture=Chain2026.05 | 10.3 | — | — | — | — | — | — | 88.3 | |
| G-SafeguardArchitecture=Chain2026.05 | 11.3 | — | — | — | — | — | — | 88.3 | |
| INFA-GuardArchitecture=Random2026.05 | 12.3 | — | — | — | — | — | — | 86.7 | |
| GUARDIANArchitecture=Chain2026.05 | 13.3 | — | — | — | — | — | — | 86.7 | |
| GUARDIANArchitecture=Tree2026.05 | 13.3 | — | — | — | — | — | — | 88.3 | |
| G-SafeguardArchitecture=Tree2026.05 | 14 | — | — | — | — | — | — | 86.7 | |
| GUARDIANArchitecture=Star2026.05 | 15.7 | — | — | — | — | — | — | 85 | |
| GUARDIANArchitecture=Random2026.05 | 17.7 | — | — | — | — | — | — | 81.7 | |
| G-SafeguardArchitecture=Star2026.05 | 18 | — | — | — | — | — | — | 81.7 | |
| G-SafeguardArchitecture=Random2026.05 | 18.3 | — | — | — | — | — | — | 80 | |
| InspectorArchitecture=Random2026.05 | 25 | — | — | — | — | — | — | 75 | |
| AgentXposedArchitecture=Tree2026.05 | 27 | — | — | — | — | — | — | 75 | |
| InspectorArchitecture=Tree2026.05 | 27 | — | — | — | — | — | — | 75 | |
| AgentXposedArchitecture=Chain2026.05 | 28 | — | — | — | — | — | — | 71.7 | |
| No DefenseArchitecture=Chain2026.05 | 29.3 | — | — | — | — | — | — | 75 | |
| ChallengerArchitecture=Tree2026.05 | 29.5 | — | — | — | — | — | — | 75 | |
| AgentSafeArchitecture=Tree2026.05 | 29.7 | — | — | — | — | — | — | 73.3 | |
| InspectorArchitecture=Star2026.05 | 30 | — | — | — | — | — | — | 70 | |
| InspectorArchitecture=Chain2026.05 | 31.3 | — | — | — | — | — | — | 68.3 | |
| AgentSafeArchitecture=Chain2026.05 | 31.7 | — | — | — | — | — | — | 68.3 | |
| ChallengerArchitecture=Chain2026.05 | 32.7 | — | — | — | — | — | — | 71.7 | |
| AgentXposedArchitecture=Star2026.05 | 34 | — | — | — | — | — | — | 66.7 | |
| No DefenseArchitecture=Star2026.05 | 34.3 | — | — | — | — | — | — | 65 | |
| No DefenseArchitecture=Tree2026.05 | 35 | — | — | — | — | — | — | 70 | |
| AgentXposedArchitecture=Random2026.05 | 35 | — | — | — | — | — | — | 68.3 | |
| AgentSafeArchitecture=Star2026.05 | 36 | — | — | — | — | — | — | 65 | |
| ChallengerArchitecture=Star2026.05 | 36.3 | — | — | — | — | — | — | 65 | |
| ChallengerArchitecture=Random2026.05 | 43 | — | — | — | — | — | — | 60 | |
| AgentSafeArchitecture=Random2026.05 | 43.7 | — | — | — | — | — | — | 55 | |
| No DefenseArchitecture=Random2026.05 | 45.7 | — | — | — | — | — | — | 53.3 | |
| AgentSafeBackbone=GPT-4o-mini, Topology=random2026.01 | — | 24.3 | 78.3 | 13.3 | 20.3 | 91.7 | 85 | — | |
| AgentSafeBackbone=Qwen3-235B-A22B, Topology=random2026.01 | — | 47.7 | 55 | 24.7 | 38 | 76.7 | 66.7 | — | |
| AgentXposed-GuideBackbone=GPT-4o-mini, Topology=random2026.01 | — | 27 | 75 | 15.3 | 23.7 | 88.3 | 80 | — | |
| AgentXposed-GuideBackbone=Qwen3-235B-A22B, Topology=random2026.01 | — | 46 | 58.3 | 23.3 | 36 | 83.3 | 66.6 | — | |
| AgentXposed-KickBackbone=GPT-4o-mini, Topology=random2026.01 | — | 25.3 | 75 | 13.3 | 21.7 | 88.3 | 80 | — | |
| AgentXposed-KickBackbone=Qwen3-235B-A22B, Topology=random2026.01 | — | 44 | 51.7 | 23.3 | 33.7 | 80 | 68.3 | — | |
| ChallengerBackbone=GPT-4o-mini, Topology=random2026.01 | — | 28 | 71.7 | 17 | 24.7 | 86.7 | 76.7 | — | |
| ChallengerBackbone=Qwen3-235B-A22B, Topology=random2026.01 | — | 54.3 | 45 | 23.3 | 39.7 | 78.3 | 65 | — | |
| G-safeguardBackbone=GPT-4o-mini, Topology=random2026.01 | — | 18 | 83.3 | 17.3 | 18 | 85 | 83.3 | — | |
| G-safeguardBackbone=Qwen3-235B-A22B, Topology=random2026.01 | — | 13.3 | 88.3 | 12.3 | 13 | 88.3 | 90 | — | |
| INFA-GUARDBackbone=GPT-4o-mini2026.01 | — | 6.1 | 96.7 | — | — | — | — | — | |
| INFA-GUARDBackbone=Qwen3-235B-A22B2026.01 | — | 8.7 | 91.7 | — | — | — | — | — | |
| INFA-GUARDBackbone=GPT-4o-mini, Topology=random2026.01 | — | 6.1 | 96.7 | 6.1 | 6.1 | 96.7 | 96.7 | — | |
| INFA-GUARDBackbone=Qwen3-235B-A22B, Topology=random2026.01 | — | 8.7 | 91.7 | 9 | 9.7 | 91.7 | 91.7 | — | |
| InspectorBackbone=GPT-4o-mini, Topology=random2026.01 | — | 25.5 | 78.7 | 24.3 | 24.7 | 78.7 | 76.7 | — | |
| InspectorBackbone=Qwen3-235B-A22B, Topology=random2026.01 | — | 34 | 70 | 24.7 | 30.3 | 80 | 71.7 | — | |
| No DefenseBackbone=GPT-4o-mini, Topology=random2026.01 | — | 38.3 | 63.3 | 27.3 | 33.3 | 75 | 68.3 | — | |
| No DefenseBackbone=Qwen3-235B-A22B, Topology=random2026.01 | — | 44 | 53.3 | 22.7 | 34 | 78.3 | 71.7 | — |