RAGCharacter localizes poisoned character spans in RAG evidence via prompt-conditioned counterfactual masking and achieves the best accuracy-over-attribution trade-off across tested attacks and models.
In: Findings of the Associa- tion for Computational Linguistics: EMNLP 2025
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CR 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Needle-in-RAG: Prompt-Conditioned Character-Level Traceback of Poisoned Spans in Retrieved Evidence
RAGCharacter localizes poisoned character spans in RAG evidence via prompt-conditioned counterfactual masking and achieves the best accuracy-over-attribution trade-off across tested attacks and models.