Prediction-Powered Risk Monitoring of Deployed Models for Detecting Harmful Distribution Shifts

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 on the type-I error. We demonstrate the effectiveness of PPRM through extensive experiments on image classification, large language model (LLM), and telecommunications monitoring tasks.

(1) Machine learning models can become unreliable when real-world data changes, but monitoring their performance is difficult when human-labeled data are scarce. (2) We developed PPRM, a method that combines synthetic labels with a small number of true labels to monitor risk over time. (3) This helps detect harmful distribution shifts early and makes deployed AI systems safer.