End-to-End Defense in RAG on FiQA
0ASRTrustRAG
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TrustRAGAttack Scenario=GL2026.05 | 0 | 0 | |
| CleanBaseAttack Scenario=GL2026.05 | 0 | 0 | |
| CleanBase + MetaSecAlignAttack Scenario=GL2026.05 | 0 | 0 | |
| TrustRAGAttack Scenario=PB2026.05 | 5 | 0.4 | |
| PromptGuardAttack Scenario=CI2026.05 | 5 | 0 | |
| PromptGuardAttack Scenario=CA2026.05 | 6 | 0 | |
| PromptGuardAttack Scenario=MA2026.05 | 6 | 0 | |
| CleanBaseAttack Scenario=CI2026.05 | 6 | 0 | |
| TrustRAGAttack Scenario=NA2026.05 | 7 | 97 | |
| CleanBase + MetaSecAlignAttack Scenario=NA2026.05 | 7 | 0 | |
| TrustRAGAttack Scenario=CI2026.05 | 8 | 0.4 | |
| TrustRAGAttack Scenario=FC2026.05 | 8 | 0 | |
| PromptGuardAttack Scenario=NA2026.05 | 8 | 0 | |
| PromptGuardAttack Scenario=FC2026.05 | 8 | 0 | |
| CleanBaseAttack Scenario=FC2026.05 | 8 | 0 | |
| CleanBaseAttack Scenario=CA2026.05 | 8 | 0 | |
| CleanBaseAttack Scenario=MA2026.05 | 8 | 0.2 | |
| CleanBase + MetaSecAlignAttack Scenario=CI2026.05 | 8 | 0 | |
| CleanBase + MetaSecAlignAttack Scenario=CA2026.05 | 8 | 0 | |
| CleanBase + MetaSecAlignAttack Scenario=MA2026.05 | 8 | 0.2 | |
| TrustRAGAttack Scenario=PW2026.05 | 9 | 0.2 | |
| CleanBase + MetaSecAlignAttack Scenario=FC2026.05 | 9 | 0 | |
| TrustRAGAttack Scenario=CA2026.05 | 10 | 0.2 | |
| TrustRAGAttack Scenario=MA2026.05 | 10 | 0 | |
| PromptGuardAttack Scenario=GL2026.05 | 10 | 10 | |
| CleanBaseAttack Scenario=NA2026.05 | 12 | 0 | |
| CleanBase + MetaSecAlignAttack Scenario=PB2026.05 | 12 | 4.4 | |
| CleanBaseAttack Scenario=PB2026.05 | 13 | 4.4 | |
| TrustRAGAttack Scenario=PT2026.05 | 21 | 0.8 | |
| PromptGuardAttack Scenario=PB2026.05 | 22 | 21.2 | |
| CleanBaseAttack Scenario=PT2026.05 | 22 | 12.4 | |
| CleanBase + MetaSecAlignAttack Scenario=PT2026.05 | 22 | 12.4 | |
| DataSentinelAttack Scenario=CA2026.05 | 23 | 7.6 | |
| CleanBase + MetaSecAlignAttack Scenario=PW2026.05 | 24 | 37.8 | |
| DataSentinelAttack Scenario=FC2026.05 | 25 | 11.2 | |
| CleanBaseAttack Scenario=PW2026.05 | 25 | 37.8 | |
| PromptGuardAttack Scenario=PT2026.05 | 26 | 28.2 | |
| PromptGuardAttack Scenario=PW2026.05 | 30 | 36.8 | |
| DataSentinelAttack Scenario=MA2026.05 | 42 | 16.4 | |
| MetaSecAlignAttack Scenario=PB2026.05 | 43 | 98.8 | |
| DataSentinelAttack Scenario=PB2026.05 | 43 | 95.8 | |
| No DefenseAttack Scenario=PB2026.05 | 45 | 98.8 | |
| DataSentinelAttack Scenario=PW2026.05 | 45 | 94.8 | |
| MetaSecAlignAttack Scenario=PT2026.05 | 46 | 97 | |
| DataSentinelAttack Scenario=PT2026.05 | 46 | 94.8 | |
| MetaSecAlignAttack Scenario=PW2026.05 | 47 | 98.8 | |
| No DefenseAttack Scenario=PT2026.05 | 49 | 97 | |
| DataSentinelAttack Scenario=CI2026.05 | 50 | 24.6 | |
| No DefenseAttack Scenario=PW2026.05 | 52 | 98.8 | |
| MetaSecAlignAttack Scenario=NA2026.05 | 52 | 99.2 | |
| DataSentinelAttack Scenario=GL2026.05 | 55 | 54 | |
| DataSentinelAttack Scenario=NA2026.05 | 58 | 54.2 | |
| MetaSecAlignAttack Scenario=FC2026.05 | 60 | 97.8 | |
| MetaSecAlignAttack Scenario=CI2026.05 | 62 | 97.2 | |
| MetaSecAlignAttack Scenario=MA2026.05 | 67 | 97.2 | |
| MetaSecAlignAttack Scenario=CA2026.05 | 68 | 97.6 | |
| MetaSecAlignAttack Scenario=GL2026.05 | 70 | 100 | |
| No DefenseAttack Scenario=GL2026.05 | 90 | 100 | |
| No DefenseAttack Scenario=FC2026.05 | 95 | 97.8 | |
| No DefenseAttack Scenario=CA2026.05 | 95 | 97.6 | |
| No DefenseAttack Scenario=MA2026.05 | 97 | 97.2 | |
| No DefenseAttack Scenario=CI2026.05 | 98 | 97.2 | |
| No DefenseAttack Scenario=NA2026.05 | 99 | 99.2 |