<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="li-etal-2026-long-short">
    <titleInfo>
        <title>Long-Short Memory Steering to Mitigate Unfaithful Self-Brainwashing</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Mingjie</namePart>
        <namePart type="family">Li</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Michael</namePart>
        <namePart type="family">Backes</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Yang</namePart>
        <namePart type="family">Zhang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Yisen</namePart>
        <namePart type="family">Wang</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>Findings of the Association for Computational Linguistics: EMNLP 2026</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">979-8-89176-462-0</identifier>
    </relatedItem>
    <abstract>Although recent large reasoning models (LRMs) achieve strong performance, their reasoning processes remain vulnerable to external manipulation. We observe that LRMs can gradually rationalize incorrect user-provided hints during multi-step reasoning. Instead of rejecting false hints, models adapt their intermediate states to accommodate them and eventually produce seemingly coherent yet incorrect conclusions. We refer to this phenomenon as self-brainwashing and analyze it through empirical and theoretical analyses. We find that unfaithful reasoning steps are often locally detectable when models only focus on local reasoning context. However, this ability degrades over long reasoning trajectories, allowing unfaithful steps to accumulate and bias the final answer toward the hint. Motivated by these findings, we propose Long-Short Memory Steering (LSMS), a training-free inference-time framework that uses short-term reasoning hidden states to enhance faithfulness and long-term reasoning hidden states to maintain consistency. Across three open-source reasoning models and four reasoning datasets, LSMS consistently reduces unfaithful answer shifts while preserving benign accuracy. On average, it lowers the shifted ratio by more than 20 percentage points under three real-world hint-based threats.</abstract>
    <identifier type="citekey">li-etal-2026-long-short</identifier>
    <location>
        <url>https://aclanthology.org/2026.findings-emnlp.205/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>4314</start>
            <end>4325</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Long-Short Memory Steering to Mitigate Unfaithful Self-Brainwashing
%A Li, Mingjie
%A Backes, Michael
%A Zhang, Yang
%A Wang, Yisen
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F li-etal-2026-long-short
%X Although recent large reasoning models (LRMs) achieve strong performance, their reasoning processes remain vulnerable to external manipulation. We observe that LRMs can gradually rationalize incorrect user-provided hints during multi-step reasoning. Instead of rejecting false hints, models adapt their intermediate states to accommodate them and eventually produce seemingly coherent yet incorrect conclusions. We refer to this phenomenon as self-brainwashing and analyze it through empirical and theoretical analyses. We find that unfaithful reasoning steps are often locally detectable when models only focus on local reasoning context. However, this ability degrades over long reasoning trajectories, allowing unfaithful steps to accumulate and bias the final answer toward the hint. Motivated by these findings, we propose Long-Short Memory Steering (LSMS), a training-free inference-time framework that uses short-term reasoning hidden states to enhance faithfulness and long-term reasoning hidden states to maintain consistency. Across three open-source reasoning models and four reasoning datasets, LSMS consistently reduces unfaithful answer shifts while preserving benign accuracy. On average, it lowers the shifted ratio by more than 20 percentage points under three real-world hint-based threats.
%U https://aclanthology.org/2026.findings-emnlp.205/
%P 4314-4325

Markdown (Informal)

[Long-Short Memory Steering to Mitigate Unfaithful Self-Brainwashing](https://aclanthology.org/2026.findings-emnlp.205/) (Li et al., Findings 2026)

• Long-Short Memory Steering to Mitigate Unfaithful Self-Brainwashing (Li et al., Findings 2026)

ACL

• Mingjie Li, Michael Backes, Yang Zhang, and Yisen Wang. 2026. Long-Short Memory Steering to Mitigate Unfaithful Self-Brainwashing. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 4314–4325, Budapest, Hungary. Association for Computational Linguistics.