Generative AI for Lending: Adverse Action, Borrower Comms, and Underwriting Support

A lender deploys a commercial AI tool to help with borrower communication. Within weeks, several issues surface. The AI drafts adverse action notices with reason codes that don't match the actual ECOA-compliant codes the underwriting system produced. The AI drafts borrower letters about loan modifications that include language violating Reg Z disclosure requirements. The AI answers borrower questions about rate lock with statements that create potential UDAAP exposure because the language is technically accurate but easily misunderstood. Each is a consequence of deploying generic AI in a context where language has specific regulatory meaning. Lender compliance teams can't rely on a tool that sounds fluent and produces content that fails regulatory review.

Lender generative AI done right is grounded in the lender's approved language, compliance-reviewed templates, and current regulatory requirements. Adverse action notice drafting that pulls actual reason codes from the decisioning system and formats them using ECOA-compliant templates. Borrower communication using Reg Z-compliant disclosure language and state-law-specific phrasing. Underwriting support that summarizes applicant data for underwriter review without making credit decisions. With explicit refusal for UDAAP-sensitive content, rate guarantees, and any content requiring specific compliance-approved phrasing. Done with this discipline, generative AI reduces compliance workload. Done casually, it becomes a regulatory finding.