Audit Protocol Detects Supervision Drift in AI Credit-Risk Models
Researchers propose a multi-signal audit framework to detect drift in proxy-labeled credit-risk prediction models, addressing base-rate shifts and feature-label relationship changes under deployment.