Responsible AI Policy

Latchstead uses AI for narrow text and information-assistance tasks described below. AI output can be incomplete, inaccurate, or misleading. A result is a starting point for review, not proof that an underlying fact is true.

Our policy is that AI output must not itself make a final housing eligibility, approval, denial, rent, deposit, lease, renewal, or eviction decision. A housing provider remains responsible for its criteria, review, notices, and final actions. The legal duties that apply depend on the decision, data, location, and program.

Before using a suggestion or extracted field, the responsible person should check the relevant source information, correct any error, and consider context the model did not receive. For the current prior-rental eviction parser, the intake contains no eviction narrative or screening report, so the AI response is not evidence of an eviction or disclosure. Obtain and review an appropriate source independently before relying on an eviction-related result. If evidence is missing, ambiguous, or disputed, leave it unresolved and use a human review process.

The screening code separates AI extraction from a deterministic evaluation of configured criteria. A rule result can still affect a landlord's workflow, so it should not be treated as an AI decision or as an automatic denial. A qualified human decision-maker must review the underlying information and applicable policy before taking consequential housing action.

• Do not use an AI response as the sole reason to deny housing or change a material housing term.
• Do not ask AI to infer protected traits or use personal characteristics as shortcuts for a housing decision.
• If an AI feature is unavailable or its result cannot be checked, use the applicable manual process.

As voluntary operating practices, housing providers can state their criteria in advance, apply them consistently, explain what information is requested and how it is used, and review source records before acting. Any legally required notice depends on the specific report and decision. AI-generated language or a parsed field may be wrong; review the source information and explain the actual basis for a decision through the provider's process.

The code and public evidence reviewed for this policy do not establish a comprehensive bias audit, measured outcome study, or fairness certification for these AI features. A prompt that asks a model to ignore personal characteristics is not proof that outputs are unbiased or that a workflow has no discriminatory effect. Customers should review criteria and outcomes with qualified counsel when concerns arise.

The current app routes AI requests through the Polsia AI service. The fields sent depend on the feature: saved property facts for listing assistance; the prior landlord name, phone number, and rental address for prior-rental parsing; the submitted issue description for maintenance urgency; and a waitlist email plus a recorded unit count for reply drafting. The current prior-rental flow sends no free-text rental-history narrative or screening report and tells the model no narrative was provided. Text entered into an issue description may include personal details, even though the maintenance prompt asks the model to ignore them.

The reviewed app code does not establish the AI provider's retention, model training, or deletion settings for these requests. We therefore make no promise here that provider-side data is not retained, used for training, or immediately deleted. Use only information needed for the task, avoid adding sensitive details unnecessarily, and consult the Privacy Notice and Vendors & Sub-processors for their current evidence and limits.

If a suggestion or extracted field appears inaccurate, biased, or unrelated to the task, do not rely on it as the basis for action. Check the source, correct the record where appropriate, and route the issue to the responsible landlord or administrator for human review. For urgent safety or maintenance needs, use the applicable emergency or maintenance channel; this policy is not an emergency service.

To raise a concern about a Latchstead AI feature, use the contact page. Include the feature and a description of the issue, but do not send unnecessary resident or applicant information. Contact does not promise a response time, investigation outcome, or legal review.

The Fair Housing Act, 42 U.S.C. § 3604 addresses specified discriminatory practices in covered sale and rental transactions, including conduct based on race, color, religion, sex, familial status, and national origin; subsection (f) addresses disability. The statute is subject to its coverage rules and exceptions. This page does not decide how it applies to a specific property, party, or decision.

Under FCRA § 615, 15 U.S.C. § 1681m(a), a person taking adverse action based in whole or in part on information in a consumer report has specified notice duties. Whether a particular rental report or decision is covered depends on the facts. An AI feature alone does not establish that the FCRA applies. The FTC's landlord guide explains common tenant-screening report and notice scenarios; it is agency guidance, while the statute controls.

HUD's April 29, 2024 tenant-screening guidance discussed AI-assisted screening and recommended practices. It is archived agency guidance, not a statute or regulation. HUD's April 6, 2026 FHEO withdrawal notice lists guidance documents withdrawn effective September 17, 2025, and says FHEO is continuing to review other existing guidance. The table does not name this particular 2024 screening publication, and we have not confirmed its current status. We cite it as policy background only and do not treat its recommendations as current legal requirements. We do not present HUD recommendations as universal duties for private landlords. Public-housing and HUD-assisted program rules apply to the participants and programs they govern; federal statutes such as the Fair Housing Act and FCRA may separately apply to private providers when their coverage requirements are met.

The NIST AI Risk Management Framework is a voluntary resource, not housing law. Latchstead does not claim certification or conformance to it.

This page is not a jurisdiction-by-jurisdiction survey of state or local housing, privacy, consumer-reporting, or automated-decision laws. Customers should ask qualified counsel about coverage, permissible screening criteria, required notices, accommodation processes, data use, and any HUD-assisted or public-housing program rules that apply to their facts.