@inproceedings{jin-etal-2026-mirage,
title = "The Mirage of Explainability: A Survey on Chain-of-Thought Faithfulness in Large Language Models",
author = "Jin, Xi and
Qi, LeChen and
Wang, Houfeng and
Pan, Liangming",
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.614/",
pages = "14122--14136",
ISBN = "979-8-89176-459-0",
abstract = "Chain-of-Thought (CoT) reasoning appears to provide explainability, leading users to trust that verbalized rationales reflect the model{'}s underlying computation. However, substantial evidence indicates that CoT often fails to reflect the model{'}s actual decision-making process, leading to a surge of research into the faithfulness of these explanations. This paper presents a comprehensive survey of CoT faithfulness. We first unify the definition of faithfulness by integrating internal alignment with external consistency and synthesize key failure phenomena, such as post-hoc rationalization and sycophancy. Furthermore, we systematize evaluation metrics, benchmarks, and critically review current mitigation strategies. We conclude by outlining open challenges and advocating for architectural innovations to achieve genuinely faithful reasoning for credible AI."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="jin-etal-2026-mirage">
<titleInfo>
<title>The Mirage of Explainability: A Survey on Chain-of-Thought Faithfulness in Large Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Xi</namePart>
<namePart type="family">Jin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">LeChen</namePart>
<namePart type="family">Qi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Houfeng</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Liangming</namePart>
<namePart type="family">Pan</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>Chain-of-Thought (CoT) reasoning appears to provide explainability, leading users to trust that verbalized rationales reflect the model’s underlying computation. However, substantial evidence indicates that CoT often fails to reflect the model’s actual decision-making process, leading to a surge of research into the faithfulness of these explanations. This paper presents a comprehensive survey of CoT faithfulness. We first unify the definition of faithfulness by integrating internal alignment with external consistency and synthesize key failure phenomena, such as post-hoc rationalization and sycophancy. Furthermore, we systematize evaluation metrics, benchmarks, and critically review current mitigation strategies. We conclude by outlining open challenges and advocating for architectural innovations to achieve genuinely faithful reasoning for credible AI.</abstract>
<identifier type="citekey">jin-etal-2026-mirage</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.614/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>14122</start>
<end>14136</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T The Mirage of Explainability: A Survey on Chain-of-Thought Faithfulness in Large Language Models
%A Jin, Xi
%A Qi, LeChen
%A Wang, Houfeng
%A Pan, Liangming
%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 jin-etal-2026-mirage
%X Chain-of-Thought (CoT) reasoning appears to provide explainability, leading users to trust that verbalized rationales reflect the model’s underlying computation. However, substantial evidence indicates that CoT often fails to reflect the model’s actual decision-making process, leading to a surge of research into the faithfulness of these explanations. This paper presents a comprehensive survey of CoT faithfulness. We first unify the definition of faithfulness by integrating internal alignment with external consistency and synthesize key failure phenomena, such as post-hoc rationalization and sycophancy. Furthermore, we systematize evaluation metrics, benchmarks, and critically review current mitigation strategies. We conclude by outlining open challenges and advocating for architectural innovations to achieve genuinely faithful reasoning for credible AI.
%U https://aclanthology.org/2026.emnlp-main.614/
%P 14122-14136
Markdown (Informal)
[The Mirage of Explainability: A Survey on Chain-of-Thought Faithfulness in Large Language Models](https://aclanthology.org/2026.emnlp-main.614/) (Jin et al., EMNLP 2026)
ACL