CiteGuard: Conformal False-Discovery Control for Faithful Retrieval-Augmented Generation

Xiangyu Jiang
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:52157-52187, 2026.

Abstract

Large language models increasingly rely on retrieval-augmented generation (RAG) to ground responses in external corpora. Yet, even with strong retrievers, generated statements can remain unsupported, and the resulting citations are often not reliable indicators of evidence. We introduce CiteGuard, a RAG decoding layer that treats sentence-level factuality as a multiple-testing problem and combines conformal calibration with false-discovery-rate control. CiteGuard converts claim–evidence scores into p-values for the null hypothesis "unsupported" and uses BH/BY procedures to decide which claims to keep (with citations) and which to abstain on. On FEVER and Natural Questions, CiteGuard reduces the false-discovery rate among accepted claims from 28–31% (vanilla RAG) to below 10% at $\alpha$=0.10, while retaining 86–92% of supported claims. This yields a user-controlled risk budget: practitioners can trade off faithfulness and coverage via $\alpha$, with finite-sample guarantees under standard exchangeability assumptions.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-jiang26f, title = {{C}ite{G}uard: Conformal False-Discovery Control for Faithful Retrieval-Augmented Generation}, author = {Jiang, Xiangyu}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {52157--52187}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/jiang26f/jiang26f.pdf}, url = {https://proceedings.mlr.press/v306/jiang26f.html}, abstract = {Large language models increasingly rely on retrieval-augmented generation (RAG) to ground responses in external corpora. Yet, even with strong retrievers, generated statements can remain unsupported, and the resulting citations are often not reliable indicators of evidence. We introduce CiteGuard, a RAG decoding layer that treats sentence-level factuality as a multiple-testing problem and combines conformal calibration with false-discovery-rate control. CiteGuard converts claim–evidence scores into p-values for the null hypothesis "unsupported" and uses BH/BY procedures to decide which claims to keep (with citations) and which to abstain on. On FEVER and Natural Questions, CiteGuard reduces the false-discovery rate among accepted claims from 28–31% (vanilla RAG) to below 10% at $\alpha$=0.10, while retaining 86–92% of supported claims. This yields a user-controlled risk budget: practitioners can trade off faithfulness and coverage via $\alpha$, with finite-sample guarantees under standard exchangeability assumptions.} }
Endnote
%0 Conference Paper %T CiteGuard: Conformal False-Discovery Control for Faithful Retrieval-Augmented Generation %A Xiangyu Jiang %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-jiang26f %I PMLR %P 52157--52187 %U https://proceedings.mlr.press/v306/jiang26f.html %V 306 %X Large language models increasingly rely on retrieval-augmented generation (RAG) to ground responses in external corpora. Yet, even with strong retrievers, generated statements can remain unsupported, and the resulting citations are often not reliable indicators of evidence. We introduce CiteGuard, a RAG decoding layer that treats sentence-level factuality as a multiple-testing problem and combines conformal calibration with false-discovery-rate control. CiteGuard converts claim–evidence scores into p-values for the null hypothesis "unsupported" and uses BH/BY procedures to decide which claims to keep (with citations) and which to abstain on. On FEVER and Natural Questions, CiteGuard reduces the false-discovery rate among accepted claims from 28–31% (vanilla RAG) to below 10% at $\alpha$=0.10, while retaining 86–92% of supported claims. This yields a user-controlled risk budget: practitioners can trade off faithfulness and coverage via $\alpha$, with finite-sample guarantees under standard exchangeability assumptions.
APA
Jiang, X.. (2026). CiteGuard: Conformal False-Discovery Control for Faithful Retrieval-Augmented Generation. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:52157-52187 Available from https://proceedings.mlr.press/v306/jiang26f.html.

Related Material