@inproceedings{zeng-etal-2025-privacyrestore,
title = "{P}rivacy{R}estore: Privacy-Preserving Inference in Large Language Models via Privacy Removal and Restoration",
author = "Zeng, Ziqian and
Wang, Jianwei and
Yang, Junyao and
Lu, Zhengdong and
Li, Haoran and
Zhuang, Huiping and
Chen, Cen",
editor = "Che, Wanxiang and
Nabende, Joyce and
Shutova, Ekaterina and
Pilehvar, Mohammad Taher",
booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
month = jul,
year = "2025",
address = "Vienna, Austria",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.acl-long.532/",
doi = "10.18653/v1/2025.acl-long.532",
pages = "10821--10855",
ISBN = "979-8-89176-251-0",
abstract = "The widespread usage of online Large Language Models (LLMs) inference services has raised significant privacy concerns about the potential exposure of private information in user inputs. Existing privacy protection methods for LLMs suffer from either insufficient privacy protection with performance degradation, or large inference time overhead. To address these limitations, we propose PrivacyRestore, a plug-and-play method to protect the privacy of user inputs during LLM inference for the client-server scenario. The server first trains restoration vectors for each privacy span type offline and then releases them to the clients. During inference, the client aggregates restoration vectors of all privacy spans in the user query into a meta restoration vector, which is later sent to the server to restore information. Before transmission, the client removes all privacy spans in the user query and applies $d_\chi$-privacy mechanism to the meta vector for privacy protection. We prove that our method can inherently prevent the linear growth of the privacy budget. We conduct extensive experimental, covering the medical and legal domains, and demonstrate that PrivacyRestore effectively protects private information and maintains acceptable levels of performance and inference efficiency"
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zeng-etal-2025-privacyrestore">
<titleInfo>
<title>PrivacyRestore: Privacy-Preserving Inference in Large Language Models via Privacy Removal and Restoration</title>
</titleInfo>
<name type="personal">
<namePart type="given">Ziqian</namePart>
<namePart type="family">Zeng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jianwei</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Junyao</namePart>
<namePart type="family">Yang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhengdong</namePart>
<namePart type="family">Lu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Haoran</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Huiping</namePart>
<namePart type="family">Zhuang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Cen</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Wanxiang</namePart>
<namePart type="family">Che</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Joyce</namePart>
<namePart type="family">Nabende</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ekaterina</namePart>
<namePart type="family">Shutova</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mohammad</namePart>
<namePart type="given">Taher</namePart>
<namePart type="family">Pilehvar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Vienna, Austria</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-251-0</identifier>
</relatedItem>
<abstract>The widespread usage of online Large Language Models (LLMs) inference services has raised significant privacy concerns about the potential exposure of private information in user inputs. Existing privacy protection methods for LLMs suffer from either insufficient privacy protection with performance degradation, or large inference time overhead. To address these limitations, we propose PrivacyRestore, a plug-and-play method to protect the privacy of user inputs during LLM inference for the client-server scenario. The server first trains restoration vectors for each privacy span type offline and then releases them to the clients. During inference, the client aggregates restoration vectors of all privacy spans in the user query into a meta restoration vector, which is later sent to the server to restore information. Before transmission, the client removes all privacy spans in the user query and applies d_χ-privacy mechanism to the meta vector for privacy protection. We prove that our method can inherently prevent the linear growth of the privacy budget. We conduct extensive experimental, covering the medical and legal domains, and demonstrate that PrivacyRestore effectively protects private information and maintains acceptable levels of performance and inference efficiency</abstract>
<identifier type="citekey">zeng-etal-2025-privacyrestore</identifier>
<identifier type="doi">10.18653/v1/2025.acl-long.532</identifier>
<location>
<url>https://aclanthology.org/2025.acl-long.532/</url>
</location>
<part>
<date>2025-07</date>
<extent unit="page">
<start>10821</start>
<end>10855</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T PrivacyRestore: Privacy-Preserving Inference in Large Language Models via Privacy Removal and Restoration
%A Zeng, Ziqian
%A Wang, Jianwei
%A Yang, Junyao
%A Lu, Zhengdong
%A Li, Haoran
%A Zhuang, Huiping
%A Chen, Cen
%Y Che, Wanxiang
%Y Nabende, Joyce
%Y Shutova, Ekaterina
%Y Pilehvar, Mohammad Taher
%S Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2025
%8 July
%I Association for Computational Linguistics
%C Vienna, Austria
%@ 979-8-89176-251-0
%F zeng-etal-2025-privacyrestore
%X The widespread usage of online Large Language Models (LLMs) inference services has raised significant privacy concerns about the potential exposure of private information in user inputs. Existing privacy protection methods for LLMs suffer from either insufficient privacy protection with performance degradation, or large inference time overhead. To address these limitations, we propose PrivacyRestore, a plug-and-play method to protect the privacy of user inputs during LLM inference for the client-server scenario. The server first trains restoration vectors for each privacy span type offline and then releases them to the clients. During inference, the client aggregates restoration vectors of all privacy spans in the user query into a meta restoration vector, which is later sent to the server to restore information. Before transmission, the client removes all privacy spans in the user query and applies d_χ-privacy mechanism to the meta vector for privacy protection. We prove that our method can inherently prevent the linear growth of the privacy budget. We conduct extensive experimental, covering the medical and legal domains, and demonstrate that PrivacyRestore effectively protects private information and maintains acceptable levels of performance and inference efficiency
%R 10.18653/v1/2025.acl-long.532
%U https://aclanthology.org/2025.acl-long.532/
%U https://doi.org/10.18653/v1/2025.acl-long.532
%P 10821-10855
Markdown (Informal)
[PrivacyRestore: Privacy-Preserving Inference in Large Language Models via Privacy Removal and Restoration](https://aclanthology.org/2025.acl-long.532/) (Zeng et al., ACL 2025)
ACL