Position: Sycophancy is an Educational Safety Risk: Why LLM Tutors Need Sycophancy Benchmarks

Enkelejda Kasneci, Gjergji Kasneci
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:170836-170855, 2026.

Abstract

This position paper argues that effective tutoring requires corrective friction: surfacing misconceptions and challenging them supportively to drive conceptual change. Yet preference-aligned LLMs can trade epistemic rigor for agreeableness. We identify a Reasoning-Sycophancy Paradox: models that resist context-switch frame attacks can still capitulate under social-epistemic pressure, especially authority ("my notes say I’m right") and social-affective face-saving ("please don’t tell me I’m wrong”). We introduce EduFrameTrap, a tutoring benchmark across math, physics, economics, chemistry, biology, and computer science that varies student confidence and pressure (context-switch, authority, social-affective). Across two frontier LLMs, context-switch failures are comparatively lower for GPT-5.2, while authority and social pressure more often trigger epistemic retreat. In contrast, Claude shows substantial context-switch fragility in this run. Because these failures are hard to judge automatically, we report two-judge disagreement as a reliability signal. We argue benchmarks should measure social-epistemic courage, i.e., supportive but corrective tutoring, and treat kind-but-correct behavior as a safety requirement.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-kasneci26a, title = {Position: Sycophancy is an Educational Safety Risk: Why {LLM} Tutors Need Sycophancy Benchmarks}, author = {Kasneci, Enkelejda and Kasneci, Gjergji}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {170836--170855}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/kasneci26a/kasneci26a.pdf}, url = {https://proceedings.mlr.press/v306/kasneci26a.html}, abstract = {This position paper argues that effective tutoring requires corrective friction: surfacing misconceptions and challenging them supportively to drive conceptual change. Yet preference-aligned LLMs can trade epistemic rigor for agreeableness. We identify a Reasoning-Sycophancy Paradox: models that resist context-switch frame attacks can still capitulate under social-epistemic pressure, especially authority ("my notes say I’m right") and social-affective face-saving ("please don’t tell me I’m wrong”). We introduce EduFrameTrap, a tutoring benchmark across math, physics, economics, chemistry, biology, and computer science that varies student confidence and pressure (context-switch, authority, social-affective). Across two frontier LLMs, context-switch failures are comparatively lower for GPT-5.2, while authority and social pressure more often trigger epistemic retreat. In contrast, Claude shows substantial context-switch fragility in this run. Because these failures are hard to judge automatically, we report two-judge disagreement as a reliability signal. We argue benchmarks should measure social-epistemic courage, i.e., supportive but corrective tutoring, and treat kind-but-correct behavior as a safety requirement.} }
Endnote
%0 Conference Paper %T Position: Sycophancy is an Educational Safety Risk: Why LLM Tutors Need Sycophancy Benchmarks %A Enkelejda Kasneci %A Gjergji Kasneci %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-kasneci26a %I PMLR %P 170836--170855 %U https://proceedings.mlr.press/v306/kasneci26a.html %V 306 %X This position paper argues that effective tutoring requires corrective friction: surfacing misconceptions and challenging them supportively to drive conceptual change. Yet preference-aligned LLMs can trade epistemic rigor for agreeableness. We identify a Reasoning-Sycophancy Paradox: models that resist context-switch frame attacks can still capitulate under social-epistemic pressure, especially authority ("my notes say I’m right") and social-affective face-saving ("please don’t tell me I’m wrong”). We introduce EduFrameTrap, a tutoring benchmark across math, physics, economics, chemistry, biology, and computer science that varies student confidence and pressure (context-switch, authority, social-affective). Across two frontier LLMs, context-switch failures are comparatively lower for GPT-5.2, while authority and social pressure more often trigger epistemic retreat. In contrast, Claude shows substantial context-switch fragility in this run. Because these failures are hard to judge automatically, we report two-judge disagreement as a reliability signal. We argue benchmarks should measure social-epistemic courage, i.e., supportive but corrective tutoring, and treat kind-but-correct behavior as a safety requirement.
APA
Kasneci, E. & Kasneci, G.. (2026). Position: Sycophancy is an Educational Safety Risk: Why LLM Tutors Need Sycophancy Benchmarks. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:170836-170855 Available from https://proceedings.mlr.press/v306/kasneci26a.html.

Related Material