@inproceedings{bao-etal-2026-loki,
title = "{LOKI}: Locating Optimal Knock-In Points for Backdoor Attacks on Dense Retrievers",
author = "Bao, Qiqi and
Feng, Linzhi and
Qian, Yaguan and
Guo, Yanru and
Wang, Bin and
Gu, Zhaoquan and
Ji, Shouling and
Lei, Zhen",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.1242/",
pages = "28309--28325",
ISBN = "979-8-89176-459-0",
abstract = "Dense retrievers are a core component of Retrieval-Augmented Generation (RAG) systems, yet their growing adoption introduces critical security risks in the Al supply chain. Existing retrieval backdoor attacks primarily rely on coarse-grained parameter updates, which often degrade benign retrieval behavior and require substantial computational cost, limiting their stealth and practicality across deployment settings. In this paper, we propose LOKI (Locating Optimal Knock-In), a framework that reformulates stealthy retrieval backdoor injection as a neuron-level local editing problem. Following a Locate-then-Edit paradigm, LOKI first identifies pivotal neurons that are highly sensitive to triggers yet decoupled from benign retrieval semantics via a multi-dimensional diagnostic analysis. And then LOKI implants backdoor logic through constrained local editing that balances trigger-induced alignment with semantic manifold preservation. Extensive experiments show that LOKI achieves near 100{\%} attack success while modifying fewer than 1{\%} of model parameters, with only limited impact on clean retrieval performance."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="bao-etal-2026-loki">
<titleInfo>
<title>LOKI: Locating Optimal Knock-In Points for Backdoor Attacks on Dense Retrievers</title>
</titleInfo>
<name type="personal">
<namePart type="given">Qiqi</namePart>
<namePart type="family">Bao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Linzhi</namePart>
<namePart type="family">Feng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yaguan</namePart>
<namePart type="family">Qian</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yanru</namePart>
<namePart type="family">Guo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Bin</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhaoquan</namePart>
<namePart type="family">Gu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shouling</namePart>
<namePart type="family">Ji</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhen</namePart>
<namePart type="family">Lei</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Dense retrievers are a core component of Retrieval-Augmented Generation (RAG) systems, yet their growing adoption introduces critical security risks in the Al supply chain. Existing retrieval backdoor attacks primarily rely on coarse-grained parameter updates, which often degrade benign retrieval behavior and require substantial computational cost, limiting their stealth and practicality across deployment settings. In this paper, we propose LOKI (Locating Optimal Knock-In), a framework that reformulates stealthy retrieval backdoor injection as a neuron-level local editing problem. Following a Locate-then-Edit paradigm, LOKI first identifies pivotal neurons that are highly sensitive to triggers yet decoupled from benign retrieval semantics via a multi-dimensional diagnostic analysis. And then LOKI implants backdoor logic through constrained local editing that balances trigger-induced alignment with semantic manifold preservation. Extensive experiments show that LOKI achieves near 100% attack success while modifying fewer than 1% of model parameters, with only limited impact on clean retrieval performance.</abstract>
<identifier type="citekey">bao-etal-2026-loki</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1242/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>28309</start>
<end>28325</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T LOKI: Locating Optimal Knock-In Points for Backdoor Attacks on Dense Retrievers
%A Bao, Qiqi
%A Feng, Linzhi
%A Qian, Yaguan
%A Guo, Yanru
%A Wang, Bin
%A Gu, Zhaoquan
%A Ji, Shouling
%A Lei, Zhen
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F bao-etal-2026-loki
%X Dense retrievers are a core component of Retrieval-Augmented Generation (RAG) systems, yet their growing adoption introduces critical security risks in the Al supply chain. Existing retrieval backdoor attacks primarily rely on coarse-grained parameter updates, which often degrade benign retrieval behavior and require substantial computational cost, limiting their stealth and practicality across deployment settings. In this paper, we propose LOKI (Locating Optimal Knock-In), a framework that reformulates stealthy retrieval backdoor injection as a neuron-level local editing problem. Following a Locate-then-Edit paradigm, LOKI first identifies pivotal neurons that are highly sensitive to triggers yet decoupled from benign retrieval semantics via a multi-dimensional diagnostic analysis. And then LOKI implants backdoor logic through constrained local editing that balances trigger-induced alignment with semantic manifold preservation. Extensive experiments show that LOKI achieves near 100% attack success while modifying fewer than 1% of model parameters, with only limited impact on clean retrieval performance.
%U https://aclanthology.org/2026.emnlp-main.1242/
%P 28309-28325
Markdown (Informal)
[LOKI: Locating Optimal Knock-In Points for Backdoor Attacks on Dense Retrievers](https://aclanthology.org/2026.emnlp-main.1242/) (Bao et al., EMNLP 2026)
ACL
- Qiqi Bao, Linzhi Feng, Yaguan Qian, Yanru Guo, Bin Wang, Zhaoquan Gu, Shouling Ji, and Zhen Lei. 2026. LOKI: Locating Optimal Knock-In Points for Backdoor Attacks on Dense Retrievers. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 28309–28325, Budapest, Hungary. Association for Computational Linguistics.