We study the problem of monitoring model performance in dynamic environments where labeled data are limited. To this end, we propose prediction-powered risk monitoring (PPRM), a semi-supervised risk-monitoring approach based on prediction-powered inference (PPI). PPRM constructs anytime-valid lower bounds on the running risk by combining synthetic labels with a small set of true labels. Harmful shifts are detected via a threshold-based comparison with an upper bound on the nominal risk, satisfying assumption-free finite-sample guarantees in the probability of false alarm. We demonstrate the effectiveness of PPRM through extensive experiments on image classification, large language model (LLM), and telecommunications monitoring tasks.
@article{arxiv.2602.02229,
title = {Prediction-Powered Risk Monitoring of Deployed Models for Detecting Harmful Distribution Shifts},
author = {Guangyi Zhang and Yunlong Cai and Guanding Yu and Osvaldo Simeone},
journal= {arXiv preprint arXiv:2602.02229},
year = {2026}
}