@inproceedings{takenami-etal-2026-mechanistically,
title = "Mechanistically Understanding and Mitigating the Anchoring Effect in {LLM}s via Attention Head Intervention",
author = "Takenami, Yoshiki and
Huang, Yin Jou and
Murawaki, Yugo and
Takeuchi, Koh and
Chu, Chenhui",
editor = "Zhang, Yue and
Xu, Ruifeng and
Purwarianti, Ayu and
Rong, Tong and
Suzuki, Jun",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {IJCNLP}-{AACL} 2026",
month = nov,
year = "2026",
address = "Hengqin, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-aacl.33/",
pages = "630--644",
ISBN = "979-8-89176-444-6",
abstract = "Similarly to humans, LLMs are susceptible to the anchoring effect in price negotiations, which can lead to unfavorable or unintended outcomes. To ensure LLM safety in dialogue tasks such as negotiation, a mechanistic understanding and mitigation methods are essential. We propose four variants of a method to identify Anchoring Heads, which are the attention heads responsible for the anchoring effect, while preserving those essential for general tasks. Experimental results show that suppressing these heads mitigates the anchoring effect by over 16{\%}, achieving complete mitigation in certain models without compromising general performance. Furthermore, the observation that suppressing random heads fails to mitigate the anchoring effect suggests that specific attention heads drive the anchoring effect by paying excessive attention to the anchor."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="takenami-etal-2026-mechanistically">
<titleInfo>
<title>Mechanistically Understanding and Mitigating the Anchoring Effect in LLMs via Attention Head Intervention</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yoshiki</namePart>
<namePart type="family">Takenami</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yin</namePart>
<namePart type="given">Jou</namePart>
<namePart type="family">Huang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yugo</namePart>
<namePart type="family">Murawaki</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Koh</namePart>
<namePart type="family">Takeuchi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chenhui</namePart>
<namePart type="family">Chu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: IJCNLP-AACL 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yue</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruifeng</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ayu</namePart>
<namePart type="family">Purwarianti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tong</namePart>
<namePart type="family">Rong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jun</namePart>
<namePart type="family">Suzuki</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Hengqin, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-444-6</identifier>
</relatedItem>
<abstract>Similarly to humans, LLMs are susceptible to the anchoring effect in price negotiations, which can lead to unfavorable or unintended outcomes. To ensure LLM safety in dialogue tasks such as negotiation, a mechanistic understanding and mitigation methods are essential. We propose four variants of a method to identify Anchoring Heads, which are the attention heads responsible for the anchoring effect, while preserving those essential for general tasks. Experimental results show that suppressing these heads mitigates the anchoring effect by over 16%, achieving complete mitigation in certain models without compromising general performance. Furthermore, the observation that suppressing random heads fails to mitigate the anchoring effect suggests that specific attention heads drive the anchoring effect by paying excessive attention to the anchor.</abstract>
<identifier type="citekey">takenami-etal-2026-mechanistically</identifier>
<location>
<url>https://aclanthology.org/2026.findings-aacl.33/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>630</start>
<end>644</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Mechanistically Understanding and Mitigating the Anchoring Effect in LLMs via Attention Head Intervention
%A Takenami, Yoshiki
%A Huang, Yin Jou
%A Murawaki, Yugo
%A Takeuchi, Koh
%A Chu, Chenhui
%Y Zhang, Yue
%Y Xu, Ruifeng
%Y Purwarianti, Ayu
%Y Rong, Tong
%Y Suzuki, Jun
%S Findings of the Association for Computational Linguistics: IJCNLP-AACL 2026
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-444-6
%F takenami-etal-2026-mechanistically
%X Similarly to humans, LLMs are susceptible to the anchoring effect in price negotiations, which can lead to unfavorable or unintended outcomes. To ensure LLM safety in dialogue tasks such as negotiation, a mechanistic understanding and mitigation methods are essential. We propose four variants of a method to identify Anchoring Heads, which are the attention heads responsible for the anchoring effect, while preserving those essential for general tasks. Experimental results show that suppressing these heads mitigates the anchoring effect by over 16%, achieving complete mitigation in certain models without compromising general performance. Furthermore, the observation that suppressing random heads fails to mitigate the anchoring effect suggests that specific attention heads drive the anchoring effect by paying excessive attention to the anchor.
%U https://aclanthology.org/2026.findings-aacl.33/
%P 630-644
Markdown (Informal)
[Mechanistically Understanding and Mitigating the Anchoring Effect in LLMs via Attention Head Intervention](https://aclanthology.org/2026.findings-aacl.33/) (Takenami et al., Findings 2026)
ACL