Fair Housing Act AI Rules for Leasing at Affordable Properties

A prospect texts your leasing line at 10pm asking why your bot told them "nothing available" while their neighbor got a tour booked the same night. Fair Housing Act AI rules don't care that a bot sent that message instead of a person. The law cares what the message said and who it said it to.

HUD spent part of 2026 pulling a stack of guidance documents off its website. That's made a lot of operators nervous about what still applies to an AI leasing tool. HUD pulled guidance, and the underlying law stayed in place. For a look at where the line sits at an income-restricted property, see AI leasing in affordable housing.

Does the Fair Housing Act Apply to an AI Leasing Tool?

Yes, the Fair Housing Act applies to an AI leasing tool the same way it applies to a leasing agent. The statute bans any "notice, statement, or advertisement" about a rental that shows a preference or limitation based on a protected class, and it specifically covers a statement someone else "caused to be made" on your behalf. A text your software sends is a statement made on your behalf.

HUD's own rule spells this out further: the prohibitions apply to "all written or oral notices or statements by a person engaged in the sale or rental of a dwelling." That list includes conveying availability to some groups and not others, and choosing which channels or audiences get which information. None of that language cares whether a human or a script typed the reply. For the basics on how the Act applies day to day, our fair housing overview is a good starting point.

Protected classes under the federal Act are race, color, religion, sex, handicap, familial status, and national origin. Source of income isn't a federal class, though plenty of states and cities have their own source-of-income laws, which is a separate check worth running with counsel.

What Did HUD Withdraw, and What Is Still in Force?

HUD withdrew a set of guidance documents in 2026, not the Fair Housing Act itself. An April 2026 Federal Register notice pulled eight FHEO documents, including the 2024 guidance on AI and digital-advertising and both assistance-animal guidance documents, with the withdrawal effective back to September 17, 2025. A July 2026 notice pulled thirteen more Office of General Counsel documents, covering everything from criminal-records guidance to a 2016 language-access memo, effective September 25, 2025. That's 21 documents total.

HUD's own words on why matter here: "these documents have been removed from active use and should not be relied upon as authoritative," and agency guidance "does not carry the force and effect of law" the way a statute does. But the same notice adds the line that answers the operator's question: "Any actions that do not comply with the text of the Fair Housing Act continue to be subject to enforcement by the Department."

One document sits in a gray zone. HUD's 2024 tenant-screening guidance isn't named in either withdrawal notice, but its page is no longer live on HUD's site. Treat it as gone from active use, not as something HUD formally withdrew on the record. Reporting on a separate September 2025 HUD memo says the agency told staff to prioritize cases with clear evidence of intentional discrimination over disparate-impact ones. HUD hasn't published that memo itself, so take it as reported, not primary-sourced.

Is Disparate Impact Still a Risk in 2026?

Yes, disparate impact is still a live risk. The rule at 24 CFR 100.500 says liability can be established from a practice's discriminatory effect, even without discriminatory intent, and that rule is still current in the federal code. HUD proposed removing it in a January 2026 rule and followed up with a supplemental proposal in August 2026, with comments closing October 9, 2026. A proposal is not the same thing as a repeal.

Even if that regulation eventually goes away, the underlying legal theory doesn't disappear with it. The Supreme Court held in 2015, in Texas Department of Housing and Community Affairs v. Inclusive Communities Project, that disparate-impact claims are cognizable under the Fair Housing Act. HUD can change its own regulation, but a regulation change alone can't undo a Supreme Court ruling. An operator who treats "HUD is rolling back a rule" as "the risk is gone" is the one most likely to get caught flat-footed by a court that reads it differently.

What Does Steering Look Like When a Bot Does It?

Steering looks like anything that restricts or discourages someone's housing choice by word or conduct because of who they are, and a bot can do all of that through what it says or leaves out. HUD's rule lists specific patterns: discouraging someone from touring or renting a unit, exaggerating drawbacks or leaving out desirable features for some prospects, telling someone they wouldn't be "comfortable" near certain residents, or steering a person toward a particular floor or section of a building.

A chat script doesn't have to say anything overtly discriminatory to steer. It can do it by which amenities it mentions to which prospects, or by consistently pointing one demographic toward the units near the back stairwell. That's why Leslie is built with no steering, no discounts, and no holds as a hard rule rather than a setting someone could switch off.

What Happened in the Two AI Fair Housing Cases?

Two cases from 2024 are still the clearest look at how this plays out, and our state of AI in leasing post covers both briefly. This section goes a step deeper.

In Louis et al. v. SafeRent Solutions, the Department of Justice filed a statement of interest in January 2023 arguing the Fair Housing Act's text and case law support applying it to screening companies as well as landlords. A Massachusetts court denied SafeRent's motion to dismiss in July 2023 on that theory. The case, over a scoring model used on voucher applicants, settled in November 2024 for up to $2.275 million, with SafeRent agreeing not to issue approve-or-decline recommendations for voucher applicants unless the model has been validated for fairness. The court keeps jurisdiction over that settlement for five years. This case involved a screening score rather than a leasing chatbot, and SafeRent made no admission of liability.

Separately, a fair housing nonprofit ran a six-month investigation into a leasing chatbot that was issuing a blanket denial to Housing Choice Voucher applicants across more than 100 properties. The case against the property's operator, Harbor Group, resolved in January 2024 with a consent decree: no more source-of-income denials, two years of monitoring, and nationwide fair housing training for staff. The chatbot's vendor isn't named here, because the mechanism matters more than the vendor: one wrong scripted answer, repeated automatically, scaled a single policy error across a whole portfolio at once. No federal enforcement action against an AI leasing tool had been filed in 2025 or 2026 as of this writing, though a law firm covering this space flagged that automated platforms are an efficient environment for fair housing testers to run through.

What Happens When an Accommodation Request Comes in by Text?

An accommodation request that arrives by text is still a request, and it still needs a prompt answer from a person. The Fair Housing Act's accommodation duty doesn't require any specific wording, and a request doesn't even have to come from the person themselves; it can come from a family member or someone acting on their behalf. A 2004 joint statement from HUD and the Department of Justice, still hosted on DOJ's site, adds that an undue delay in responding to a reasonable accommodation request can itself be treated as a denial.

Federally assisted properties carry an extra layer here under Section 504, which requires effective communication with anyone requesting it. A text-only channel can't be the only door in if a prospect asks for a phone call or another format instead.

A message like "can I keep my cat, it helps my anxiety" doesn't say "reasonable accommodation" anywhere in it, and it still counts as one. That's why Leslie's guardrail on fair housing and accommodation topics has no toggle: every site, every time, straight to a person.

Do You Have to Answer Prospects in Other Languages?

You don't have one clean federal answer to that question anymore. HUD's guidance on language access has been withdrawn, and the executive order it leaned on for language-access policy was revoked in 2025. What hasn't moved is Title VI's ban on national-origin discrimination at any property receiving federal financial assistance, and national origin remains a protected class under the Fair Housing Act on its own. Our language access at affordable housing post covers what that obligation looks like day to day.

What that means in practice depends on your program documents, your state's rules, and your own counsel, not on a blog post. Replying to a prospect in the language they wrote in is a reasonable service to offer at an income-restricted property. It's not a substitute for whatever your specific program requires, and it doesn't earn you compliance credit on its own. Leslie replies in Spanish when a prospect writes in Spanish, from the same live data it always uses.

What Records Should You Keep of AI Conversations?

Keep the transcript and the reason for every closed conversation, because the filing windows on a fair housing claim run for a while. A complainant has one year to file with HUD and two years to sue in court after an alleged violation. HUD-assisted owners already have to keep waitlist removal records, including the reason someone came off the list, for three years under Handbook 4350.3. There's no single federal rule that sets a retention period for AI chat logs specifically, so match your practice to those windows and confirm the exact number with counsel. Our post on preparing affordable housing files for review covers the broader recordkeeping habit this fits into.

If your property carries an affirmative fair housing marketing plan, check that your automated outreach still matches who and how that plan says you market. A plan that describes a human process on paper, while a bot runs a different one in practice, is a gap worth closing before an auditor finds it.

What Should You Demand From Any AI Leasing Vendor?

Ask a vendor these questions before you sign anything, and get the answers in writing.

• Does it answer only from your live data, and what does it say when it doesn't know something?
• Which topics get hard-routed to a person, and can any site turn that routing off?
• How fast does a flagged conversation reach a named person, and what does the prospect see in the meantime?
• Does it ever recommend a specific unit, building, or floor based on anything the prospect said about themselves?
• Does every prospect get the same specials, fees, and availability information?
• Can you export full transcripts and the reason for every closed conversation, and how long are they kept?
• Can a prospect reach a human by phone or another format on request?
• Which languages does it reply in, and what happens with the ones it doesn't cover?
• Who tested it for fair housing risk, and can you see the test scripts?
• Does the contract give you audit rights and notice when the underlying model changes?

Ask for the boring screens in the demo as well as the polished conversation. A vendor who can pull up a flagged-conversation queue and an audit log without stalling is showing you the part that matters most.

When Should You Not Turn on an AI Leasing Tool?

Don't turn on an AI leasing tool if nobody is assigned to work flagged conversations within a business day, because a fast bot reply followed by a slow human follow-up on a sensitive topic is worse than no bot at all. Don't turn it on if your listing data, rents, openings, and fees aren't current in the system it reads from, because a wrong "nothing available" answer is a 3604(d) problem waiting to happen. And don't turn it on before your counsel has looked at how it interacts with your written tenant selection plan or affirmative marketing plan.

This doesn't mean AI leasing is unsafe, only that the tool is as careful as the process built around it. Leslie flags fair housing, qualifying, and accommodation topics to your team at every site, and it texts only prospects and applicants, never residents. No software can make your property fair housing compliant on its own. The housing provider carries that responsibility, and the vendor doesn't take it off your plate.

This post explains what the law says. It isn't legal advice, and it can't be, because your program documents, your state's rules, and your specific facts all matter here. Talk to your counsel before you change how your property markets, screens, or communicates with prospects.