AI Disclosure

1. Loanwright Uses Artificial Intelligence

Loanwright's Guideline Search feature uses artificial intelligence to help mortgage professionals locate provisions within published agency mortgage guidelines. The feature ships in two surfaces with different AI behavior:

• Public Guideline Search (no account, at /tools/guideline-search) uses AI only to embed your query for vector similarity search. It returns the most relevant agency excerpts verbatim, with section citations. No generative model is invoked, and no AI summary is produced.
• Signed-in Guideline Search (Pro) runs the same retrieval and additionally uses a large language model (currently OpenAI GPT-4o) to summarize the matched excerpts into a cited answer tuned to your question.

2. What the AI Does

Across both surfaces, the system:

• Searches a knowledge base of published agency guidelines (currently: Fannie Mae Selling Guide, Freddie Mac Seller/Servicer Guide, FHA/HUD Handbook 4000.1, USDA HB-1-3555, and the VA Loan Guaranty corpus consisting of 38 CFR Part 36, the active VA Loan Guaranty Circulars, and the VA Lender's Handbook / Pamphlet 26-7 as published in VA's KnowVA self-service portal; additional sources are added over time)
• Returns numbered source citations so you can locate and verify the original text

Additionally, on the signed-in surface only, the system:

• Generates a plain-language summary of the most relevant guideline excerpts
• May serve a cached AI summary when a substantially similar question has been previously answered by another signed-in user. Cached answers are clearly labeled and display the original processing time for transparency. Cached answers are automatically invalidated when the underlying knowledge base is updated, ensuring cached results always reflect the same guideline versions as freshly generated answers. The cache is scoped to signed-in queries only; public queries do not write to or read from this cache.

4. Accuracy and Limitations

On the signed-in surface, AI-generated summaries may be inaccurate, incomplete, or outdated. The knowledge base is refreshed periodically but may not reflect the latest agency updates at all times. You are solely responsible for verifying all AI-generated content against the original source documents before making any decisions. AI-generated content is not a substitute for professional judgment.

On the public surface, the response is the agency text itself (verbatim excerpts), so the accuracy concern is the same as reading the original published guideline: the excerpts are only as current as the most recently ingested version. The lookup is matched by semantic similarity, so a less-relevant passage can be returned when no closely matching excerpt exists; verify the section reference against the published guideline before relying on a passage for a file decision.

5. Data and Privacy

When you use Guideline Search on either surface, your question text is sent to OpenAI's API for processing. The public surface sends it only to the embedding endpoint (text-embedding-3-small) for vector search. The signed-in surface additionally sends the matched excerpts and your query to a generative model (currently GPT-4o) to produce a cited summary. No personal information, borrower data, or loan file data is included in any of these requests. OpenAI does not use API data for model training. For full details, see our Privacy Policy.

6. Safeguards

Loanwright implements multiple safeguards to reduce the risk of harm from AI-generated content. The first and most important is structural: AI-generated answers are limited to authenticated users in the signed-in app. The public Guideline Search returns the source passages themselves, with no generative model in the response path, so an anonymous user cannot be served a hallucinated answer or a poisoned cache entry from an attempted prompt injection.

For the signed-in surface, additional safeguards include:

• A per-session acknowledgment confirming you understand the content is AI-generated and requires verification
• Every answer is labeled with an “AI Generated” badge
• Source citations accompany every answer
• A “Report Inaccuracy” button is available on every answer
• The AI is instructed not to provide advice, recommendations, or eligibility determinations
• Cached answers are labeled and show the original processing duration so you know the answer was not freshly generated. The cache is automatically invalidated whenever the knowledge base is updated, so cached results stay current with the latest ingested guidelines. The cache pool is scoped to signed-in queries only; public queries cannot read or write entries

Usage rate limits apply to both surfaces to prevent misuse.

7. Risk Management

Loanwright maintains an internal AI Risk Management Policy and Impact Assessment that document how risks from AI use are identified, mitigated, and monitored. These documents are reviewed annually. If you would like more information about our AI risk management practices, contact us at the address below.

8. Standards Alignment

Loanwright voluntarily aligns its AI governance practices with two recognized frameworks:

• NIST AI Risk Management Framework (AI RMF 1.0) published by the National Institute of Standards and Technology. Loanwright maintains a NIST AI RMF Profile mapping the framework's GOVERN, MAP, MEASURE, and MANAGE functions to specific controls and documentation.
• ISO/IEC 42001:2023, the international standard for AI Management Systems. Loanwright maintains a Statement of Applicability mapping ISO 42001 Annex A controls to its AI practices. Loanwright is not certified to ISO 42001; alignment is voluntary.

For details on our framework alignment, contact us at hello@loanwright.io.

9. Contact Us

If you have questions or concerns about Loanwright's use of artificial intelligence, or if you believe AI-generated content has caused you harm, contact us at: privacy@loanwright.io

10. Related Documents

• Terms of Service (Section 4: AI-Generated Content)
• Privacy Policy (Section 6: Third-Party Services)