@inproceedings{ugandhar-etal-2026-reasoning,
title = "Can Reasoning Models Detect Changes to their Chains of Thought?",
author = "Ugandhar, Sathvik Manikantan Napa and
Singh, Utkarsh and
Xue, Chengyuan and
Wanner, Miriam and
Walden, William Gantt",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Junyi 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.1571/",
doi = "10.18653/2026.emnlp-main.1571",
pages = "35784--35801",
ISBN = "979-8-89176-459-0",
abstract = "There are many reasons one may want to edit a model{'}s chain of thought (CoT){---}e.g., to prefill it with reasoning from a stronger model or to remove steps that may yield unsafe outputs. The success of these edits plausibly depends on a model{'}s inability to notice them, as the model may alter its behavior if it suspects tampering. In this work, we study whether recent reasoning models are able to detect such edits on their CoTs under a variety of conditions: during reasoning and after it; when alerted to the possibility of tampering and when not alerted; and when prefilled both with their own CoTs and with those of other models. Broadly, we find that (i) models exhibit only very modest detection accuracy; (ii) models struggle to identify *where* their CoT was modified; and (iii) models are about as good at detecting changes to their own CoTs as to those of other models."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="ugandhar-etal-2026-reasoning">
<titleInfo>
<title>Can Reasoning Models Detect Changes to their Chains of Thought?</title>
</titleInfo>
<name type="personal">
<namePart type="given">Sathvik</namePart>
<namePart type="given">Manikantan</namePart>
<namePart type="given">Napa</namePart>
<namePart type="family">Ugandhar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Utkarsh</namePart>
<namePart type="family">Singh</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chengyuan</namePart>
<namePart type="family">Xue</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Miriam</namePart>
<namePart type="family">Wanner</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">William</namePart>
<namePart type="given">Gantt</namePart>
<namePart type="family">Walden</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">Junyi</namePart>
<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>There are many reasons one may want to edit a model’s chain of thought (CoT)—e.g., to prefill it with reasoning from a stronger model or to remove steps that may yield unsafe outputs. The success of these edits plausibly depends on a model’s inability to notice them, as the model may alter its behavior if it suspects tampering. In this work, we study whether recent reasoning models are able to detect such edits on their CoTs under a variety of conditions: during reasoning and after it; when alerted to the possibility of tampering and when not alerted; and when prefilled both with their own CoTs and with those of other models. Broadly, we find that (i) models exhibit only very modest detection accuracy; (ii) models struggle to identify *where* their CoT was modified; and (iii) models are about as good at detecting changes to their own CoTs as to those of other models.</abstract>
<identifier type="citekey">ugandhar-etal-2026-reasoning</identifier>
<identifier type="doi">10.18653/2026.emnlp-main.1571</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1571/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>35784</start>
<end>35801</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Can Reasoning Models Detect Changes to their Chains of Thought?
%A Ugandhar, Sathvik Manikantan Napa
%A Singh, Utkarsh
%A Xue, Chengyuan
%A Wanner, Miriam
%A Walden, William Gantt
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Junyi 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 ugandhar-etal-2026-reasoning
%X There are many reasons one may want to edit a model’s chain of thought (CoT)—e.g., to prefill it with reasoning from a stronger model or to remove steps that may yield unsafe outputs. The success of these edits plausibly depends on a model’s inability to notice them, as the model may alter its behavior if it suspects tampering. In this work, we study whether recent reasoning models are able to detect such edits on their CoTs under a variety of conditions: during reasoning and after it; when alerted to the possibility of tampering and when not alerted; and when prefilled both with their own CoTs and with those of other models. Broadly, we find that (i) models exhibit only very modest detection accuracy; (ii) models struggle to identify *where* their CoT was modified; and (iii) models are about as good at detecting changes to their own CoTs as to those of other models.
%R 10.18653/2026.emnlp-main.1571
%U https://aclanthology.org/2026.emnlp-main.1571/
%U https://doi.org/10.18653/2026.emnlp-main.1571
%P 35784-35801
Markdown (Informal)
[Can Reasoning Models Detect Changes to their Chains of Thought?](https://aclanthology.org/2026.emnlp-main.1571/) (Ugandhar et al., EMNLP 2026)
ACL
- Sathvik Manikantan Napa Ugandhar, Utkarsh Singh, Chengyuan Xue, Miriam Wanner, and William Gantt Walden. 2026. Can Reasoning Models Detect Changes to their Chains of Thought?. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 35784–35801, Budapest, Hungary. Association for Computational Linguistics.