Q&A: Monthly Q&A Briefing
Answer:
Q: What recent enforcement should QC teams know about?
A: Massachusetts reached a $2.5 million settlement tied to alleged failures in AI underwriting governance, including lack of disparate‑impact testing, reliance on arbitrary human selections in training, and inadequate adverse‑action notices. Mass.gov
Q: Are other states taking action or issuing rules?
A: Yes. New Jersey and New York have strengthened state‑level expectations for explainability and consumer protection around automated decisioning, and several states have issued guidance or enforcement actions targeting biased algorithmic outcomes. This creates a fragmented, multi‑state compliance landscape for multi‑state originators. Lexology
Q: How are GSEs and investors responding?
A: Fannie Mae and Freddie Mac have issued AI governance expectations for seller/servicers that require inventories, documented controls, and auditable governance programs; investors increasingly condition purchase eligibility on model documentation. QC failures tied to model explainability can create representation, warranty, and buy‑back risk. HousingWire
Q: What specific model governance gaps create QC exposure?
A: The common deficiencies are insufficient ongoing validation, missing documentation tying algorithm outputs to explainable reason codes for adverse‑action notices, and poor tracking of human overrides that may introduce inconsistent exceptions, all of which QC reviews can and should detect. National Mortgage Professional
Q: Q: What specific model governance gaps create QC exposure?e QC exposure?
A: Prioritize these checkpoints:
a) Model inventory and owner verification;
b) Periodic fair‑lending and performance testing (drift, disparate impact);
c) Adverse‑action traceability from model score to reason code;
d) Override consistency reviews and exception pattern analysis;
e) Vendor governance evidence (contracts, testing results). National Mortgage Professional National Mortgage Professional
Quick implementation guide (key considerations)
a) Scope: Map all AI/algorithmic tools touching origination and decisioning.
b) Data: Ensure test datasets include protected‑class proxies for disparate‑impact analysis.
c) Documentation: Capture versioning, training inputs, validation reports, and adverse‑action logic.
d) QC integration: Add pre‑funding fair‑lending screens and post‑closing audits that verify decision logic.
Risks and timeline
a) Risk: State enforcement and investor rejection can follow QC findings; adverse‑action deficiencies are a high‑risk area. Mass.gov HousingWire
b) Timing: Expect heightened state scrutiny through late‑2026; building a defensible governance program takes months; start immediately. Lexology
