How RLHF Amplifies Sycophancy

Itai Shapira, Gerdus Benade, Ariel D. Procaccia
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:110417-110441, 2026.

Abstract

Large language models often exhibit increased sycophantic behavior after preference-based post-training, showing a stronger tendency to affirm a user’s stated or implied belief even when this conflicts with factual accuracy or sound judgment. We present a formal analysis of how alignment from human feedback can increase this failure mode by identifying an explicit amplification mechanism that causally links optimization against a learned reward to bias in the human preference data used for alignment. We show that the direction of behavioral drift is determined by a covariance under the base policy between endorsing the belief signal in the prompt and the learned reward, and that the first-order effect reduces to a simple mean-gap condition. We then analyze reward learning from pairwise comparisons under random utility models like Bradley–Terry and characterize when bias in human annotators’ preferences induces this reward gap. Next, we propose a training-time intervention designed to neutralize the amplification mechanism itself. Among all post-trained policies that prevent sycophantic behavior from increasing, we characterize the unique policy closest in KL divergence to the unconstrained post-trained policy, and derive the corresponding minimal reward correction as a closed-form agreement penalty. Computational experiments find that reward gaps are common and cause behavioral drift in all the configurations considered.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-shapira26a, title = {How {RLHF} Amplifies Sycophancy}, author = {Shapira, Itai and Benade, Gerdus and Procaccia, Ariel D.}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {110417--110441}, 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/shapira26a/shapira26a.pdf}, url = {https://proceedings.mlr.press/v306/shapira26a.html}, abstract = {Large language models often exhibit increased sycophantic behavior after preference-based post-training, showing a stronger tendency to affirm a user’s stated or implied belief even when this conflicts with factual accuracy or sound judgment. We present a formal analysis of how alignment from human feedback can increase this failure mode by identifying an explicit amplification mechanism that causally links optimization against a learned reward to bias in the human preference data used for alignment. We show that the direction of behavioral drift is determined by a covariance under the base policy between endorsing the belief signal in the prompt and the learned reward, and that the first-order effect reduces to a simple mean-gap condition. We then analyze reward learning from pairwise comparisons under random utility models like Bradley–Terry and characterize when bias in human annotators’ preferences induces this reward gap. Next, we propose a training-time intervention designed to neutralize the amplification mechanism itself. Among all post-trained policies that prevent sycophantic behavior from increasing, we characterize the unique policy closest in KL divergence to the unconstrained post-trained policy, and derive the corresponding minimal reward correction as a closed-form agreement penalty. Computational experiments find that reward gaps are common and cause behavioral drift in all the configurations considered.} }
Endnote
%0 Conference Paper %T How RLHF Amplifies Sycophancy %A Itai Shapira %A Gerdus Benade %A Ariel D. Procaccia %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-shapira26a %I PMLR %P 110417--110441 %U https://proceedings.mlr.press/v306/shapira26a.html %V 306 %X Large language models often exhibit increased sycophantic behavior after preference-based post-training, showing a stronger tendency to affirm a user’s stated or implied belief even when this conflicts with factual accuracy or sound judgment. We present a formal analysis of how alignment from human feedback can increase this failure mode by identifying an explicit amplification mechanism that causally links optimization against a learned reward to bias in the human preference data used for alignment. We show that the direction of behavioral drift is determined by a covariance under the base policy between endorsing the belief signal in the prompt and the learned reward, and that the first-order effect reduces to a simple mean-gap condition. We then analyze reward learning from pairwise comparisons under random utility models like Bradley–Terry and characterize when bias in human annotators’ preferences induces this reward gap. Next, we propose a training-time intervention designed to neutralize the amplification mechanism itself. Among all post-trained policies that prevent sycophantic behavior from increasing, we characterize the unique policy closest in KL divergence to the unconstrained post-trained policy, and derive the corresponding minimal reward correction as a closed-form agreement penalty. Computational experiments find that reward gaps are common and cause behavioral drift in all the configurations considered.
APA
Shapira, I., Benade, G. & Procaccia, A.D.. (2026). How RLHF Amplifies Sycophancy. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:110417-110441 Available from https://proceedings.mlr.press/v306/shapira26a.html.

Related Material