@inproceedings{zhang-etal-2026-helpfulness,
title = "When Helpfulness Becomes Scientific Sycophancy: Triggering and Mitigating {LLM} Failures on Flawed Scientific Premises",
author = "Zhang, Yiyuan and
Hua, Cailong and
Salapaka, Murti and
Kang, Dongyeop",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.1501/",
doi = "10.18653/2026.findings-emnlp.1501",
pages = "31958--31987",
ISBN = "979-8-89176-462-0",
abstract = "Large language models (LLMs) are increasingly used as scientific assistants, but their helpfulness can become harmful when they validate or build upon a user{'}s flawed scientific premise instead of challenging it. We study this failure mode as scientific sycophancy and introduce SciSycEval, a trap-based benchmark that tests whether LLMs reject, ambiguously qualify, or sycophantically operationalize flawed but plausible premises. Using trap assertions constructed from evidence abstracts across six domains and three reasoning-error types, we evaluate high-performing models under controlled task and authority settings. We find that scientific sycophancy decreases sharply across stronger GPT models, but persists as a trigger-dependent failure mode: generative tasks, high-authority framing, domain-specific evidential gaps, and overgeneralized premises substantially amplify sycophantic behavior. Critical-evaluation prompts reduce but do not eliminate the failures, showing that scientific LLMs must be evaluated and safeguarded for epistemic robustness under the interaction pressures that trigger sycophancy, rather than for factual accuracy alone."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhang-etal-2026-helpfulness">
<titleInfo>
<title>When Helpfulness Becomes Scientific Sycophancy: Triggering and Mitigating LLM Failures on Flawed Scientific Premises</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yiyuan</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Cailong</namePart>
<namePart type="family">Hua</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Murti</namePart>
<namePart type="family">Salapaka</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dongyeop</namePart>
<namePart type="family">Kang</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>Large language models (LLMs) are increasingly used as scientific assistants, but their helpfulness can become harmful when they validate or build upon a user’s flawed scientific premise instead of challenging it. We study this failure mode as scientific sycophancy and introduce SciSycEval, a trap-based benchmark that tests whether LLMs reject, ambiguously qualify, or sycophantically operationalize flawed but plausible premises. Using trap assertions constructed from evidence abstracts across six domains and three reasoning-error types, we evaluate high-performing models under controlled task and authority settings. We find that scientific sycophancy decreases sharply across stronger GPT models, but persists as a trigger-dependent failure mode: generative tasks, high-authority framing, domain-specific evidential gaps, and overgeneralized premises substantially amplify sycophantic behavior. Critical-evaluation prompts reduce but do not eliminate the failures, showing that scientific LLMs must be evaluated and safeguarded for epistemic robustness under the interaction pressures that trigger sycophancy, rather than for factual accuracy alone.</abstract>
<identifier type="citekey">zhang-etal-2026-helpfulness</identifier>
<identifier type="doi">10.18653/2026.findings-emnlp.1501</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1501/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>31958</start>
<end>31987</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T When Helpfulness Becomes Scientific Sycophancy: Triggering and Mitigating LLM Failures on Flawed Scientific Premises
%A Zhang, Yiyuan
%A Hua, Cailong
%A Salapaka, Murti
%A Kang, Dongyeop
%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 zhang-etal-2026-helpfulness
%X Large language models (LLMs) are increasingly used as scientific assistants, but their helpfulness can become harmful when they validate or build upon a user’s flawed scientific premise instead of challenging it. We study this failure mode as scientific sycophancy and introduce SciSycEval, a trap-based benchmark that tests whether LLMs reject, ambiguously qualify, or sycophantically operationalize flawed but plausible premises. Using trap assertions constructed from evidence abstracts across six domains and three reasoning-error types, we evaluate high-performing models under controlled task and authority settings. We find that scientific sycophancy decreases sharply across stronger GPT models, but persists as a trigger-dependent failure mode: generative tasks, high-authority framing, domain-specific evidential gaps, and overgeneralized premises substantially amplify sycophantic behavior. Critical-evaluation prompts reduce but do not eliminate the failures, showing that scientific LLMs must be evaluated and safeguarded for epistemic robustness under the interaction pressures that trigger sycophancy, rather than for factual accuracy alone.
%R 10.18653/2026.findings-emnlp.1501
%U https://aclanthology.org/2026.findings-emnlp.1501/
%U https://doi.org/10.18653/2026.findings-emnlp.1501
%P 31958-31987
Markdown (Informal)
[When Helpfulness Becomes Scientific Sycophancy: Triggering and Mitigating LLM Failures on Flawed Scientific Premises](https://aclanthology.org/2026.findings-emnlp.1501/) (Zhang et al., Findings 2026)
ACL