@inproceedings{maraia-etal-2026-sycophancy,
title = "The Sycophancy Matrix: Measuring and Steering Ethos-Driven Sycophancy Across Languages and Cultures",
author = "Maraia, Gabriele and
Menchi, Micol and
Fidelibus, Alessia and
Magnante, Luana and
Orlando, Irene and
Palazzi, Gaia and
Zanzotto, Fabio Massimo",
editor = "Ranaldi, Leonardo and
Pucci, Giulia and
Ranaldi, Federico and
Valentino, Marco and
Birch, Alexandra and
Zanzotto, Fabio Massimo",
booktitle = "The Proceedings of the Workshop on Open Reasoning Across Cultures Languages ({ORACLE})",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.oracle-main.2/",
pages = "22--39",
ISBN = "979-8-89176-483-5",
abstract = "Large Language Models (LLMs) exhibit sycophancy, aligning with user assertions even when they conflict with facts. We frame sycophancy as a sociolinguistic phenomenon, disentangling explicit authority (credentials) from implicit authority (linguistic register), and adopt the Sycophancy Matrix, an adversarial evaluation framework that isolates these variables. Evaluating open-weight models on a controlled TruthfulQA subset across English, Spanish, and Portuguese cultural variants, we find that models exhibit statistically significant cultural variation and show stable sycophancy fingerprints intrinsic to model family, independent of language. We then recast the Matrix as a defence framework, evaluating two activation steering methods, CAA and NoLAI, on whether methods trained on English transfer to Italian, Spanish, and Portuguese. Both methods reduce the Weighted Sycophancy Rate across all four attack styles, and the defence generalises cross-lingually: sociolinguistic stance, its emergence, and its correction are encoded in a largely language-agnostic geometric space."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="maraia-etal-2026-sycophancy">
<titleInfo>
<title>The Sycophancy Matrix: Measuring and Steering Ethos-Driven Sycophancy Across Languages and Cultures</title>
</titleInfo>
<name type="personal">
<namePart type="given">Gabriele</namePart>
<namePart type="family">Maraia</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Micol</namePart>
<namePart type="family">Menchi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alessia</namePart>
<namePart type="family">Fidelibus</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Luana</namePart>
<namePart type="family">Magnante</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Irene</namePart>
<namePart type="family">Orlando</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gaia</namePart>
<namePart type="family">Palazzi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Fabio</namePart>
<namePart type="given">Massimo</namePart>
<namePart type="family">Zanzotto</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>The Proceedings of the Workshop on Open Reasoning Across Cultures Languages (ORACLE)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Leonardo</namePart>
<namePart type="family">Ranaldi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Giulia</namePart>
<namePart type="family">Pucci</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Federico</namePart>
<namePart type="family">Ranaldi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Marco</namePart>
<namePart type="family">Valentino</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alexandra</namePart>
<namePart type="family">Birch</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Fabio</namePart>
<namePart type="given">Massimo</namePart>
<namePart type="family">Zanzotto</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">979-8-89176-483-5</identifier>
</relatedItem>
<abstract>Large Language Models (LLMs) exhibit sycophancy, aligning with user assertions even when they conflict with facts. We frame sycophancy as a sociolinguistic phenomenon, disentangling explicit authority (credentials) from implicit authority (linguistic register), and adopt the Sycophancy Matrix, an adversarial evaluation framework that isolates these variables. Evaluating open-weight models on a controlled TruthfulQA subset across English, Spanish, and Portuguese cultural variants, we find that models exhibit statistically significant cultural variation and show stable sycophancy fingerprints intrinsic to model family, independent of language. We then recast the Matrix as a defence framework, evaluating two activation steering methods, CAA and NoLAI, on whether methods trained on English transfer to Italian, Spanish, and Portuguese. Both methods reduce the Weighted Sycophancy Rate across all four attack styles, and the defence generalises cross-lingually: sociolinguistic stance, its emergence, and its correction are encoded in a largely language-agnostic geometric space.</abstract>
<identifier type="citekey">maraia-etal-2026-sycophancy</identifier>
<location>
<url>https://aclanthology.org/2026.oracle-main.2/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>22</start>
<end>39</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T The Sycophancy Matrix: Measuring and Steering Ethos-Driven Sycophancy Across Languages and Cultures
%A Maraia, Gabriele
%A Menchi, Micol
%A Fidelibus, Alessia
%A Magnante, Luana
%A Orlando, Irene
%A Palazzi, Gaia
%A Zanzotto, Fabio Massimo
%Y Ranaldi, Leonardo
%Y Pucci, Giulia
%Y Ranaldi, Federico
%Y Valentino, Marco
%Y Birch, Alexandra
%Y Zanzotto, Fabio Massimo
%S The Proceedings of the Workshop on Open Reasoning Across Cultures Languages (ORACLE)
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-483-5
%F maraia-etal-2026-sycophancy
%X Large Language Models (LLMs) exhibit sycophancy, aligning with user assertions even when they conflict with facts. We frame sycophancy as a sociolinguistic phenomenon, disentangling explicit authority (credentials) from implicit authority (linguistic register), and adopt the Sycophancy Matrix, an adversarial evaluation framework that isolates these variables. Evaluating open-weight models on a controlled TruthfulQA subset across English, Spanish, and Portuguese cultural variants, we find that models exhibit statistically significant cultural variation and show stable sycophancy fingerprints intrinsic to model family, independent of language. We then recast the Matrix as a defence framework, evaluating two activation steering methods, CAA and NoLAI, on whether methods trained on English transfer to Italian, Spanish, and Portuguese. Both methods reduce the Weighted Sycophancy Rate across all four attack styles, and the defence generalises cross-lingually: sociolinguistic stance, its emergence, and its correction are encoded in a largely language-agnostic geometric space.
%U https://aclanthology.org/2026.oracle-main.2/
%P 22-39
Markdown (Informal)
[The Sycophancy Matrix: Measuring and Steering Ethos-Driven Sycophancy Across Languages and Cultures](https://aclanthology.org/2026.oracle-main.2/) (Maraia et al., ORACLE 2026)
ACL