End-to-End Defense in RAG on ArguAna
0Attack Success Rate (ASR)TrustRAG
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TrustRAGAttack Scenario=GL2026.05 | 0 | 0 | |
| TrustRAGAttack Scenario=PT2026.05 | 0 | 6.2 | |
| PromptGuardAttack Scenario=GL2026.05 | 0 | 1 | |
| CleanBaseAttack Scenario=GL2026.05 | 0 | 0 | |
| CleanBase + MetaSecAlignAttack Scenario=GL2026.05 | 0 | 0 | |
| CleanBase + MetaSecAlignAttack Scenario=PT2026.05 | 1 | 18.2 | |
| PromptGuardAttack Scenario=PT2026.05 | 2 | 43.4 | |
| CleanBaseAttack Scenario=PT2026.05 | 2 | 18.2 | |
| MetaSecAlignAttack Scenario=FC2026.05 | 3 | 79.8 | |
| MetaSecAlignAttack Scenario=PT2026.05 | 3 | 74.4 | |
| CleanBase + MetaSecAlignAttack Scenario=CA2026.05 | 3 | 0 | |
| CleanBase + MetaSecAlignAttack Scenario=PB2026.05 | 3 | 0 | |
| CleanBase + MetaSecAlignAttack Scenario=NA2026.05 | 4 | 0 | |
| CleanBase + MetaSecAlignAttack Scenario=CI2026.05 | 4 | 0 | |
| CleanBase + MetaSecAlignAttack Scenario=MA2026.05 | 4 | 0 | |
| PromptGuardAttack Scenario=CA2026.05 | 5 | 0.4 | |
| CleanBaseAttack Scenario=MA2026.05 | 5 | 0 | |
| CleanBase + MetaSecAlignAttack Scenario=FC2026.05 | 5 | 0 | |
| CleanBase + MetaSecAlignAttack Scenario=PW2026.05 | 5 | 0.2 | |
| No DefenseAttack Scenario=PT2026.05 | 6 | 74.4 | |
| TrustRAGAttack Scenario=NA2026.05 | 6 | 0 | |
| PromptGuardAttack Scenario=NA2026.05 | 6 | 0 | |
| DataSentinelAttack Scenario=PT2026.05 | 6 | 55.6 | |
| CleanBaseAttack Scenario=CI2026.05 | 6 | 0 | |
| CleanBaseAttack Scenario=FC2026.05 | 6 | 0 | |
| CleanBaseAttack Scenario=CA2026.05 | 6 | 0 | |
| CleanBaseAttack Scenario=PW2026.05 | 6 | 0.2 | |
| CleanBaseAttack Scenario=PB2026.05 | 6 | 0 | |
| MetaSecAlignAttack Scenario=NA2026.05 | 7 | 79.8 | |
| TrustRAGAttack Scenario=FC2026.05 | 7 | 0 | |
| CleanBaseAttack Scenario=NA2026.05 | 7 | 0 | |
| MetaSecAlignAttack Scenario=PB2026.05 | 8 | 79.8 | |
| TrustRAGAttack Scenario=MA2026.05 | 8 | 0 | |
| TrustRAGAttack Scenario=PW2026.05 | 8 | 0 | |
| PromptGuardAttack Scenario=CI2026.05 | 8 | 0.2 | |
| PromptGuardAttack Scenario=FC2026.05 | 8 | 0 | |
| MetaSecAlignAttack Scenario=CA2026.05 | 9 | 79.8 | |
| PromptGuardAttack Scenario=MA2026.05 | 9 | 0.2 | |
| DataSentinelAttack Scenario=PB2026.05 | 9 | 79.8 | |
| No DefenseAttack Scenario=PB2026.05 | 10 | 79.8 | |
| MetaSecAlignAttack Scenario=CI2026.05 | 10 | 79.8 | |
| TrustRAGAttack Scenario=CI2026.05 | 10 | 0 | |
| TrustRAGAttack Scenario=CA2026.05 | 10 | 0 | |
| MetaSecAlignAttack Scenario=PW2026.05 | 11 | 79.4 | |
| MetaSecAlignAttack Scenario=MA2026.05 | 12 | 79.8 | |
| PromptGuardAttack Scenario=PW2026.05 | 12 | 79.4 | |
| DataSentinelAttack Scenario=PW2026.05 | 12 | 79.4 | |
| No DefenseAttack Scenario=PW2026.05 | 14 | 79.4 | |
| TrustRAGAttack Scenario=PB2026.05 | 14 | 4.8 | |
| No DefenseAttack Scenario=GL2026.05 | 15 | 80 | |
| MetaSecAlignAttack Scenario=GL2026.05 | 15 | 80 | |
| PromptGuardAttack Scenario=PB2026.05 | 15 | 79.8 | |
| DataSentinelAttack Scenario=GL2026.05 | 15 | 74 | |
| DataSentinelAttack Scenario=CA2026.05 | 16 | 49.6 | |
| No DefenseAttack Scenario=FC2026.05 | 17 | 79.8 | |
| DataSentinelAttack Scenario=CI2026.05 | 18 | 54.4 | |
| DataSentinelAttack Scenario=MA2026.05 | 19 | 54.4 | |
| No DefenseAttack Scenario=CA2026.05 | 20 | 79.8 | |
| DataSentinelAttack Scenario=NA2026.05 | 21 | 66.6 | |
| No DefenseAttack Scenario=NA2026.05 | 22 | 79.8 | |
| No DefenseAttack Scenario=MA2026.05 | 22 | 79.8 | |
| DataSentinelAttack Scenario=FC2026.05 | 22 | 75 | |
| No DefenseAttack Scenario=CI2026.05 | 25 | 79.8 |