This article is general information for operational planning, not legal advice. Fair housing enforcement and AI-specific regulation are both moving targets in 2026 — confirm current requirements with qualified counsel before changing screening policy.
On May 2, 2024, the U.S. Department of Housing and Urban Development issued guidance addressing how the Fair Housing Act applies to artificial intelligence in two specific areas of multifamily operations: tenant screening, and the advertising of housing opportunities through platforms that use targeted, algorithmic ad delivery.
The guidance made one point unambiguous: a housing provider is accountable for a discriminatory outcome regardless of whether it came from an internal process or from a third-party vendor's algorithm. Outsourcing the screening decision to a software company does not outsource the liability. HUD was explicit that the Fair Housing Act reaches both intentional discrimination and practices that produce an unjustified discriminatory effect — meaning an algorithm never designed to discriminate can still create fair housing exposure if its outcomes do.
Separately, the SafeRent Fair Housing settlement has become one of the most-cited enforcement examples in this space — part of a broader pattern of scrutiny directed at opaque, score-based tenant screening algorithms that applicants and regulators cannot meaningfully inspect or contest.
Why "the vendor's algorithm did it" is not a defense
This is the operational reality HUD's guidance establishes: if a third-party screening tool scores an applicant in a way that produces a disparate impact on a protected class, the property management company that used that tool carries the exposure — not just the software vendor that built it. That reframes vendor selection and screening-tool configuration from a procurement decision into a governance decision.
Practically, that means an organization needs to be able to answer:
- What factors does our screening tool actually weigh, and can we explain them in plain language if challenged?
- Who reviews a borderline or denied application before the denial goes out?
- Do we have a documented, consistent process — applied the same way across every property — or does practice vary by region or individual leasing agent?
- If asked, could we produce a record showing a human reviewed the basis for an automated denial?
The regulatory posture is unsettled — plan for that, not around it
It is worth being direct about the uncertainty here rather than overstating it: HUD's enforcement posture around AI and disparate-impact liability has shifted since the 2024 guidance was issued, and organizations should not assume the exact 2024 framework remains untouched in its current form. What has not become less relevant is the underlying exposure — fair housing liability for screening outcomes predates this guidance by decades and does not depend on any single administration's enforcement priorities remaining fixed.
The operationally sound position is to build a screening governance process that would hold up regardless of which way federal guidance moves next: documented criteria, human review of adverse outcomes, and a consistent process applied identically across every property. That is a defensible posture under disparate-impact theory whether HUD's 2024 guidance is actively enforced, revised, or superseded.
What a governed screening process looks like
It does not mean abandoning AI-assisted screening. It means the screening tool operates inside a documented structure: clear criteria, a human checkpoint before an adverse decision is finalized, consistent application across the portfolio, and a record that shows the organization can explain and defend every denial — not just process it quickly.
That structure is what Sharpe Group's governance framework is built to put in place around the tools an operator already uses, so that "the algorithm denied them" is never the end of the explanation an organization can give — to a resident, a regulator, or a court.