Responsible AI use

AI may assist; it does not receive a licence

AI can help retrieve governed information, organize a borrower's facts, explain concepts and invoke approved deterministic calculations. It does not become a licensed mortgage professional and must not present uncertain policy, lender discretion or regulated recommendations as settled fact.

No guessing when confidence is insufficient

• A knowledge gap should be acknowledged rather than filled with fabricated information.
• Missing borrower facts should be identified before a conclusion is attempted.
• Lender-discretion or policy ambiguity should be surfaced as ambiguity and, where needed, confirmed with the lender or a licensed professional.
• Legal, tax, appraisal and other professional conclusions should be routed to the appropriate professional.

Human escalation preserves context

When licensed review is required, the system should preserve the structured context already collected so the borrower does not need to repeat the same story. The handoff should include the relevant borrower profile, question, calculations, evidence, unresolved issue, confidence level and reason for escalation where those elements are available and appropriate to share.

AI retrieval is permission-aware

Not every internal record is eligible for public AI answers. Content can be classified as primary, supporting, evidence-only, navigation-only or excluded, and access controls can prevent internal lender intelligence, client-specific information or restricted compliance material from entering public answer synthesis.

Policy-sensitive answers can require a freshness check. Case evidence must not be silently converted into a universal lender rule, and deterministic calculator outputs should come from the governed calculation engine rather than model arithmetic when an approved engine exists.

Source-grounded answers and citations

Being searchable by AI does not automatically make information eligible for a borrower-facing answer. Policy-sensitive material can remain indexed for governance while being blocked from synthesis when its source is missing, stale, superseded, outside the requested jurisdiction or too broadly cited for the risk of the claim.

Where an AI-assisted answer relies on governed policy evidence, the system should build citations from registered sources rather than inventing links or authorities. A funded case may illustrate what occurred in one file, but it must not be converted into a universal lender rule without separate policy evidence.

Answer decisions are governed

The system can answer, qualify an answer, request missing information, require a governed calculation, escalate, refuse, or refuse to infer. Attempts to bypass source controls, obtain restricted internal information, or turn uncertain evidence into a definitive policy answer do not override those boundaries.

Auditability without duplicating sensitive data

AI-assisted decision paths are designed to preserve evidence, calculation, citation, confidence and escalation lineage for review. The AI-governance audit record is privacy-minimized: it can record hashes and field names rather than duplicating the raw question or raw borrower fact values into a separate governance log.

Privacy and data use

Personal information used in an AI-enabled workflow remains subject to applicable privacy and security controls. HopeWell's public website does not treat a visitor's mortgage inquiry as permission to expose that information to advertisers, and optional analytics or advertising tools are controlled separately from the ability to use the website or submit an inquiry.