Overview
- U.S. banking supervisors have increased hands‑on questioning of lenders’ AI use in routine exams, pressing firms to map where models run, show data limits, and prove who can stop a system if it misbehaves.
- Freddie Mac has required explicit AI and machine‑learning governance in its seller/servicer guide since March 3, 2026, creating a concrete compliance baseline for conforming lending.
- The Mortgage Bankers Association released a white paper calling for a unified, principles‑based AI risk framework for mortgage that would standardize rules on model validation, vendor oversight and consumer protections.
- Across regulators and industry guidance four operational pillars have emerged — explainability, fairness testing, human‑in‑the‑loop oversight, and continuous audit documentation — that lenders must embed before scaling AI.
- Lenders are already seeing large productivity gains from assistive AI in underwriting and condition clearing, but heavy vendor reliance, state laws and consumer‑protection rules mean borrowers must still be tied to a named, licensed human originator and easy access to human review.