Performative Privacy: When Differential Privacy Maximizes Utility
Abstract
Privacy-preserving learning is often motivated by the idea that protecting users' data can preserve trust and thus participation, improving utility in the long term. However, this claim has not been formalized so far. In parallel, performative learning provides a framework for studying learning systems whose deployment affects the data they later observe. In this work, we bring these two perspectives together and introduce performative privacy, where data leakage reduces future participation. We study a simple model where agents repeatedly contribute data for mean estimation but may leave the system when their data is leaked. Privacy is implemented through differentially private mechanisms, creating a trade-off between estimation noise and future participation. We show, through a theoretical study of the dynamics and numerical experiments, that a finite privacy budget can outperform non-private estimation in the long term when the feedback loop between leakage and participation is sufficiently strong. This provides first evidence that differential privacy can be optimal not only as a protection mechanism, but also from the perspective of long-term utility.
Related
- The Right to Be Forgotten under Correlation: Privacy-Utility Tradeoffs
- Diffusion Sensitivity Geometry and Strong Duality for Optimal Privacy-Utility Tradeoffs
- On Privacy Mechanism Design Under a Bounded α-Lift Leakage Constraint with χ2-Utility
- Prophet Inequalities under Local Differential Privacy
- BCFL-DAI-PUB: Blockchain-integrated federated learning with dynamic adaptive incentives for privacy-utility balanced credit risk modelling
BibTeX
@misc{mukherjee2026performative,
title = {{Performative Privacy: When Differential Privacy Maximizes Utility}},
author = {Uddalak Mukherjee and Edwige Cyffers and Yann Chevaleyre},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2608.28198},
eprint = {2608.28198},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2608.28198},
note = {CC BY-NC-SA 4.0},
}
RIS
TY - UNPB
TI - Performative Privacy: When Differential Privacy Maximizes Utility
AU - Uddalak Mukherjee
AU - Edwige Cyffers
AU - Yann Chevaleyre
PY - 2026
DA - 2026/09/28
PB - arXiv
DO - 10.48550/arxiv.2608.28198
UR - https://arxiv.org/abs/2608.28198
AB - Privacy-preserving learning is often motivated by the idea that protecting users' data can preserve trust and thus participation, improving utility in the long term. However, this claim has not been formalized so far. In parallel, performative learning provides a framework for studying learning systems whose deployment affects the data they later observe. In this work, we bring these two perspectives together and introduce performative privacy, where data leakage reduces future participation. We study a simple model where agents repeatedly contribute data for mean estimation but may leave the system when their data is leaked. Privacy is implemented through differentially private mechanisms, creating a trade-off between estimation noise and future participation. We show, through a theoretical study of the dynamics and numerical experiments, that a finite privacy budget can outperform non-private estimation in the long term when the feedback loop between leakage and participation is sufficiently strong. This provides first evidence that differential privacy can be optimal not only as a protection mechanism, but also from the perspective of long-term utility.
ER -