Draft — pending counsel review. This AI/ML Use Disclaimer is part of the LoanIQ contractual framework. Counsel familiar with Fannie Mae LL-2026-04, ECOA / Regulation B, FCRA adverse-action requirements, and CFPB guidance on AI in credit decisioning should review before adoption.

AI/ML Use Disclaimer

Effective: [DATE] · Last updated: [DATE] · Version 1.1

This Disclaimer governs your use of any artificial intelligence ("AI") or machine learning ("ML") feature inside the LoanIQ platform operated by Mortgage Dude Holdings, LLC ("MDH"). It is incorporated by reference into the Terms of Service. Read together with our Data Processing Addendum and the AI Governance Policy (the published mirror of MDH's information security program).

1. What "AI/ML" means here

The LoanIQ platform uses third-party large language models (currently Anthropic Claude Sonnet 4.6) and a managed OCR service (Google Cloud Vision) to power four product surfaces:

  • Scenario Desk (Lola) — guideline retrieval and scenario triage. The model receives your prompt + retrieved guideline chunks and returns an informational summary with citations.
  • Ask Lola — Income Calculator right-rail — context-aware Q&A while you size qualifying income. The model receives your question + a sanitized snapshot of the calculator state (no borrower names, no bank/last-4) and returns an interpretation grounded in agency rules.
  • Income OCR (Tax Return / Paystub / Bank Statement) — structured-data extraction from borrower PDFs. The pipeline runs Cloud Vision OCR → server-side PII sanitization → Anthropic Claude on sanitized text only. Anthropic does not receive borrower PDFs or unredacted OCR text. See Section 4 below for the data flow.
  • Email drafting — borrower-facing or AE-facing draft emails based on a scenario you have already approved. Drafts are seeded by you and edited by you before sending.

2. AI output is informational. You are the licensed professional.

Every AI/ML output in the platform is provided as informational assistance. It is not, and may not be relied upon as:

  • A loan commitment, pre-approval, or Loan Estimate under Regulation Z.
  • An underwriting decision, condition, or final qualifying-income determination.
  • A credit decision, adverse-action determination, or denial.
  • Legal, compliance, regulatory, accounting, or tax advice.
  • A substitute for the licensed mortgage professional's independent judgment.

You are the licensed mortgage professional responsible for every scenario decision, every qualifying-income figure used in underwriting, every guideline interpretation, and every borrower-facing communication generated by the platform. The Service places "informational only" framing on every AI/ML output so the decisional authority is unambiguous.

3. AI output may be wrong. Verify against the source.

Large language models can hallucinate, miss context, or summarize a guideline incorrectly. Cloud Vision can misread a low-quality scan or an unusual layout. The platform mitigates by (i) retrieval-grounded prompting on the Scenario Desk, (ii) tool-schema-forced structured responses on Income OCR, (iii) confidence pills on Scenario Desk responses, and (iv) explicit "review and approve" gates on every Income OCR extraction. None of those mitigations replaces your verification:

  • Verify any guideline citation against the live source (Fannie Selling Guide, Freddie Seller/Servicer Guide, FHA 4000.1, VA Handbook, USDA HB-1-3555, or the lender matrix of record).
  • Verify every Income OCR extraction against the source PDF before clicking Approve. The platform will not let you push extracted values into the calculator until you have approved them.
  • Treat AI-drafted emails as drafts. Edit before sending.

4. Data flow and sub-processors (Income OCR pipeline)

For Income OCR specifically, MDH has structured the pipeline so that borrower non-public personal information (NPI) does not reach Anthropic. The data flow is:

  1. You upload a borrower PDF to Firebase Storage. The PDF is encrypted at rest and readable only by your authenticated Firebase user account.
  2. A server-side Cloud Function downloads the PDF and submits it to Google Cloud Vision API for OCR. Cloud Vision is a Google-internal call covered by the existing Google Cloud DPA. Cloud Vision does not retain the document beyond the operational window required to return a response.
  3. The OCR text returns to the Cloud Function, where a server-side sanitization layer redacts SSN, EIN, full account numbers, ABA routing numbers, driver license numbers, dates of birth (in DOB context), credit-card-shaped numbers, phone numbers, and email addresses. Borrower names and addresses are tokenized for restoration after extraction.
  4. A residual-PII check refuses the call if any high-confidence PII pattern remains in the redacted text. If the check fails, Anthropic is never invoked and the failure is logged.
  5. Anthropic Claude receives ONLY the sanitized text plus a tool schema that itself excludes PII fields from the response. Anthropic returns a structured JSON object.
  6. Server-side restoration replaces the borrower-name and address tokens with the original values. The result is persisted to your owner-scoped Firestore tree.

For Scenario Desk and Ask Lola, your prompts are sent to Anthropic as you typed them. The system prompt instructs Lola to refuse borrower-identifying inputs (full SSN, full account number, etc.) and to treat any such inputs as a request error. You are responsible for not pasting NPI into free-text prompts. See Section 3 of the Acceptable Use Policy for specific restrictions on AI prompt content.

5. Sub-processor retention

Anthropic's standard API retains prompts and completions for up to 30 days for trust-and-safety review, then auto-deletes. For the Income OCR path, this retention does not present an NPI-disclosure concern because the prompts contain no borrower NPI (PII has been redacted). For Scenario Desk and Ask Lola, prompts may contain scenario details (loan amounts, FICO scores, occupancy types, etc.); see Section 8 of the AI Governance Policy for the documented risk frame.

Cloud Vision does not retain document content beyond the operational window required to return a response.

Pinecone (the Scenario Desk retrieval index) holds only investor guideline text MDH has uploaded; no borrower data is indexed in Pinecone.

6. Fair lending and prohibited use

You may not use AI/ML output in the platform to make adverse-action decisions, to produce or justify denials, or to introduce disparate treatment or impact in a manner inconsistent with ECOA / Regulation B, the Fair Housing Act, or any equivalent law. ECOA-protected attributes (race, color, religion, national origin, sex, marital status, age, source of public assistance) are not extracted by the OCR tool schemas and must not be entered into prompts. See Section 3 of the Acceptable Use Policy for specific restrictions.

7. Fannie Mae LL-2026-04 governance

Customer LOs that sell or service loans to Fannie Mae are subject to Lender Letter LL-2026-04 ("Governance framework on use of artificial intelligence and machine learning"). MDH supports your compliance by documenting the AI/ML systems in use (Section 3 of the AI Governance Policy), maintaining the audit trail (Section 7D), tracking sub-processors (Section 8), and operating change-management procedures (Section 7G). LL-2026-04 obligations attach to you as the seller/servicer; MDH supplies the disclosure-ready records on request.

8. Adverse action and FCRA

Where AI/ML output influences a credit decision that results in adverse action, FCRA adverse-action disclosure obligations apply to your decision, not to the AI. You must produce specific and accurate adverse-action reasons under FCRA and ECOA / Regulation B. The platform's audit trail and citation-bearing outputs are designed to support your reason-giving; they are not adverse-action notices in themselves.

9. Changes to this Disclaimer

MDH may update this Disclaimer from time to time. Material changes will be announced via email to the address on your account and posted with a new Effective Date. Continued use after the change constitutes acceptance.

10. Contact

Mortgage Dude Holdings, LLC
Attn: AI Governance
Austin, Texas
Email: austen@austensmith.com