List your company

End-to-End Defense in RAG on HotpotQA

0Attack Success Rate (ASR)TrustRAG
-3.8822.3148.574.69May 1, 2026May 5, 2026May 9, 2026May 13, 2026May 17, 2026May 21, 2026May 26, 2026

Evaluation results

MethodLinks
TrustRAGAttack Scenario=GL2026.0500
CleanBaseAttack Scenario=GL2026.0500
CleanBase + MetaSecAlignAttack Scenario=GL2026.0500
Cordon-MASSeed=422026.050—
PromptGuardAttack Scenario=FC2026.0530
CleanBaseAttack Scenario=NA2026.0530
CleanBase + MetaSecAlignAttack Scenario=NA2026.0530
CleanBase + MetaSecAlignAttack Scenario=CI2026.0530
PromptGuardAttack Scenario=NA2026.0540
PromptGuardAttack Scenario=CI2026.0540
PromptGuardAttack Scenario=CA2026.0540
PromptGuardAttack Scenario=MA2026.0540
PromptGuardAttack Scenario=PB2026.0540
CleanBaseAttack Scenario=CI2026.0540
CleanBaseAttack Scenario=CA2026.0540
CleanBase + MetaSecAlignAttack Scenario=FC2026.0540.6
CleanBase + MetaSecAlignAttack Scenario=CA2026.0540
CleanBase + MetaSecAlignAttack Scenario=PB2026.0541
TrustRAGAttack Scenario=CA2026.0550.2
PromptGuardAttack Scenario=GL2026.0552
TrustRAGAttack Scenario=FC2026.0560
TrustRAGAttack Scenario=MA2026.0560.6
TrustRAGAttack Scenario=PT2026.0560.2
CleanBaseAttack Scenario=FC2026.0560.6
CleanBaseAttack Scenario=PB2026.0561
CleanBase + MetaSecAlignAttack Scenario=MA2026.0561.2
RobustRAGSeed=422026.056.1—
CleanBaseAttack Scenario=MA2026.0571.2
TrustRAGAttack Scenario=NA2026.05899
TrustRAGAttack Scenario=PW2026.0580.8
TrustRAGAttack Scenario=PB2026.0580
TrustRAGAttack Scenario=CI2026.0590.2
CleanBaseAttack Scenario=PT2026.05114
PromptGuardAttack Scenario=PT2026.05135
CleanBase + MetaSecAlignAttack Scenario=PT2026.05134
DebateSeed=422026.0516.3—
PromptGuardAttack Scenario=PW2026.05187.8
DataSentinelAttack Scenario=CA2026.052211.4
TrustRAGSeed=422026.0524.5—
CleanBase + MetaSecAlignAttack Scenario=PW2026.052723.8
CleanBaseAttack Scenario=PW2026.052823.8
Vanilla RAGSeed=422026.0528.6—
ParaphraseSeed=422026.0530.6—
DataSentinelAttack Scenario=CI2026.053521.2
DataSentinelAttack Scenario=FC2026.053826.6
MetaSecAlignAttack Scenario=NA2026.053999.8
MetaSecAlignAttack Scenario=FC2026.053997.6
DataSentinelAttack Scenario=MA2026.054427.4
DataSentinelAttack Scenario=GL2026.054531
MetaSecAlignAttack Scenario=GL2026.0550100
MetaSecAlignAttack Scenario=MA2026.055397.2
MetaSecAlignAttack Scenario=CA2026.055795.8
MetaSecAlignAttack Scenario=PT2026.056098.4
MetaSecAlignAttack Scenario=CI2026.056198.2
No DefenseAttack Scenario=PT2026.057598.4
DataSentinelAttack Scenario=NA2026.057776.8
DataSentinelAttack Scenario=PT2026.057897.4
MetaSecAlignAttack Scenario=PW2026.057999.2
DataSentinelAttack Scenario=PW2026.058289.8
MetaSecAlignAttack Scenario=PB2026.058399.8
No DefenseAttack Scenario=GL2026.0585100
No DefenseAttack Scenario=CA2026.058995.8
No DefenseAttack Scenario=FC2026.059297.6
DataSentinelAttack Scenario=PB2026.059294.4
No DefenseAttack Scenario=MA2026.059397.2
No DefenseAttack Scenario=PW2026.059499.2
No DefenseAttack Scenario=NA2026.059599.8
No DefenseAttack Scenario=CI2026.059798.2
No DefenseAttack Scenario=PB2026.059799.8