@inproceedings{zhengyuan-etal-2026-ldede,
title = "{LDEDE}: {LRP}-Driven Efficient Detection and Editing Framework for {LLM} Privacy Neurons",
author = "Zhengyuan, Zhao and
Lifeng, Cao and
Sunhaodong and
Haotian, Shi and
Xuehui, Du and
Aodi, Liu and
Lanjie, Niu and
Xiaocheng, Yang",
editor = "Liakata, Maria and
Moreira, Viviane P. and
Zhang, Jiajun and
Jurgens, David",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {ACL} 2026",
month = jul,
year = "2026",
address = "San Diego, California, United States",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-acl.1666/",
doi = "10.18653/v1/2026.findings-acl.1666",
pages = "33315--33329",
ISBN = "979-8-89176-395-1",
abstract = "The rapid advancement of large language models (LLMs) has significantly propelled downstream innovation, yet pervasive sensitive information in training data and the models' memory characteristics pose severe privacy leakage risks. This contravenes core requirements of the General Data Protection Regulation (GDPR) and the right to be forgotten, becoming a critical bottleneck for secure and compliant deployment. Existing privacy protection methods have notable limitations: data preprocessing fails to cover context-dependent sensitive information; differential privacy (DP) and homomorphic encryption (HE) degrade model performance and increase computational overhead; traditional machine unlearning may cause catastrophic collapse; and neuron editing methods struggle with the accuracy-efficiency trade-off in privacy neuron localization, alongside privacy seesaw phenomena and general performance degradation. To address these challenges, this paper proposes LDEDE, a Layer-wise Relevance Propagation (LRP)-driven framework for efficient privacy neuron detection and editing. It offers three core advantages: 1) Precise multi-scale privacy localization via LRP-based relevance backpropagation and multi-token attention aggregation, achieving over 80{\%} higher efficiency than gradient attribution methods; 2) First reveals the existence of ``coupled privacy neurons'' in LLMs, which are the key cause of the privacy seesaw phenomenon{---}mitigated by Polarity-Aware Neuron Editing (PANE) with differentiated logic; 3) Enhanced robustness and generalization for batch processing via privacy neuron aggregation. Experiments on Enron and MIMIC datasets demonstrate that compared to baselines, LDEDE maintains comparable general performance while reducing leakage risks of Phone, Email, and medical privacy by 42.7{\%}{--}73.5{\%} on average and cutting computational time by 60{\%}{--}90{\%}. It also exhibits stable performance across GPT-2, BERT-base, and LLAMA-7B, providing an efficient, lightweight solution for post-deployment dynamic LLM privacy protection."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhengyuan-etal-2026-ldede">
<titleInfo>
<title>LDEDE: LRP-Driven Efficient Detection and Editing Framework for LLM Privacy Neurons</title>
</titleInfo>
<name type="personal">
<namePart type="given">Zhao</namePart>
<namePart type="family">Zhengyuan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Cao</namePart>
<namePart type="family">Lifeng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name>
<namePart>Sunhaodong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shi</namePart>
<namePart type="family">Haotian</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Du</namePart>
<namePart type="family">Xuehui</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Liu</namePart>
<namePart type="family">Aodi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Niu</namePart>
<namePart type="family">Lanjie</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yang</namePart>
<namePart type="family">Xiaocheng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: ACL 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Maria</namePart>
<namePart type="family">Liakata</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Viviane</namePart>
<namePart type="given">P</namePart>
<namePart type="family">Moreira</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiajun</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">David</namePart>
<namePart type="family">Jurgens</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">San Diego, California, United States</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-395-1</identifier>
</relatedItem>
<abstract>The rapid advancement of large language models (LLMs) has significantly propelled downstream innovation, yet pervasive sensitive information in training data and the models’ memory characteristics pose severe privacy leakage risks. This contravenes core requirements of the General Data Protection Regulation (GDPR) and the right to be forgotten, becoming a critical bottleneck for secure and compliant deployment. Existing privacy protection methods have notable limitations: data preprocessing fails to cover context-dependent sensitive information; differential privacy (DP) and homomorphic encryption (HE) degrade model performance and increase computational overhead; traditional machine unlearning may cause catastrophic collapse; and neuron editing methods struggle with the accuracy-efficiency trade-off in privacy neuron localization, alongside privacy seesaw phenomena and general performance degradation. To address these challenges, this paper proposes LDEDE, a Layer-wise Relevance Propagation (LRP)-driven framework for efficient privacy neuron detection and editing. It offers three core advantages: 1) Precise multi-scale privacy localization via LRP-based relevance backpropagation and multi-token attention aggregation, achieving over 80% higher efficiency than gradient attribution methods; 2) First reveals the existence of “coupled privacy neurons” in LLMs, which are the key cause of the privacy seesaw phenomenon—mitigated by Polarity-Aware Neuron Editing (PANE) with differentiated logic; 3) Enhanced robustness and generalization for batch processing via privacy neuron aggregation. Experiments on Enron and MIMIC datasets demonstrate that compared to baselines, LDEDE maintains comparable general performance while reducing leakage risks of Phone, Email, and medical privacy by 42.7%–73.5% on average and cutting computational time by 60%–90%. It also exhibits stable performance across GPT-2, BERT-base, and LLAMA-7B, providing an efficient, lightweight solution for post-deployment dynamic LLM privacy protection.</abstract>
<identifier type="citekey">zhengyuan-etal-2026-ldede</identifier>
<identifier type="doi">10.18653/v1/2026.findings-acl.1666</identifier>
<location>
<url>https://aclanthology.org/2026.findings-acl.1666/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>33315</start>
<end>33329</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T LDEDE: LRP-Driven Efficient Detection and Editing Framework for LLM Privacy Neurons
%A Zhengyuan, Zhao
%A Lifeng, Cao
%A Haotian, Shi
%A Xuehui, Du
%A Aodi, Liu
%A Lanjie, Niu
%A Xiaocheng, Yang
%Y Liakata, Maria
%Y Moreira, Viviane P.
%Y Zhang, Jiajun
%Y Jurgens, David
%A Sunhaodong
%S Findings of the Association for Computational Linguistics: ACL 2026
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, California, United States
%@ 979-8-89176-395-1
%F zhengyuan-etal-2026-ldede
%X The rapid advancement of large language models (LLMs) has significantly propelled downstream innovation, yet pervasive sensitive information in training data and the models’ memory characteristics pose severe privacy leakage risks. This contravenes core requirements of the General Data Protection Regulation (GDPR) and the right to be forgotten, becoming a critical bottleneck for secure and compliant deployment. Existing privacy protection methods have notable limitations: data preprocessing fails to cover context-dependent sensitive information; differential privacy (DP) and homomorphic encryption (HE) degrade model performance and increase computational overhead; traditional machine unlearning may cause catastrophic collapse; and neuron editing methods struggle with the accuracy-efficiency trade-off in privacy neuron localization, alongside privacy seesaw phenomena and general performance degradation. To address these challenges, this paper proposes LDEDE, a Layer-wise Relevance Propagation (LRP)-driven framework for efficient privacy neuron detection and editing. It offers three core advantages: 1) Precise multi-scale privacy localization via LRP-based relevance backpropagation and multi-token attention aggregation, achieving over 80% higher efficiency than gradient attribution methods; 2) First reveals the existence of “coupled privacy neurons” in LLMs, which are the key cause of the privacy seesaw phenomenon—mitigated by Polarity-Aware Neuron Editing (PANE) with differentiated logic; 3) Enhanced robustness and generalization for batch processing via privacy neuron aggregation. Experiments on Enron and MIMIC datasets demonstrate that compared to baselines, LDEDE maintains comparable general performance while reducing leakage risks of Phone, Email, and medical privacy by 42.7%–73.5% on average and cutting computational time by 60%–90%. It also exhibits stable performance across GPT-2, BERT-base, and LLAMA-7B, providing an efficient, lightweight solution for post-deployment dynamic LLM privacy protection.
%R 10.18653/v1/2026.findings-acl.1666
%U https://aclanthology.org/2026.findings-acl.1666/
%U https://doi.org/10.18653/v1/2026.findings-acl.1666
%P 33315-33329
Markdown (Informal)
[LDEDE: LRP-Driven Efficient Detection and Editing Framework for LLM Privacy Neurons](https://aclanthology.org/2026.findings-acl.1666/) (Zhengyuan et al., Findings 2026)
ACL
- Zhao Zhengyuan, Cao Lifeng, Sunhaodong, Shi Haotian, Du Xuehui, Liu Aodi, Niu Lanjie, and Yang Xiaocheng. 2026. LDEDE: LRP-Driven Efficient Detection and Editing Framework for LLM Privacy Neurons. In Findings of the Association for Computational Linguistics: ACL 2026, pages 33315–33329, San Diego, California, United States. Association for Computational Linguistics.