Most property teams now use some form of automated screening. A score arrives, a recommendation attaches to it, a decision follows. Very few teams can explain, if asked, what produced the score.

That gap is the risk. Not the technology, and not anyone's intent. The exposure comes from making consequential decisions about who gets housing using a process you cannot describe, evidence, or defend.

This article covers what United States fair housing law covers when an algorithm is involved, where liability sits between a housing provider and a screening vendor, what has changed in federal enforcement and what has not, and what documentation actually protects a decision. For the broader capability picture, RIOO's overview of AI in property management platforms covers what these systems do.

Key takeaways

• The Fair Housing Act applies to housing decisions regardless of who makes them or what technology is used.

• Outsourcing screening does not outsource responsibility. The housing provider remains the decision-maker.

• Federal enforcement posture has shifted recently. State law, local ordinances, and private litigation have not.

• Credit score, eviction records, and criminal history are the three criteria carrying the most exposure.

• The thing that protects a decision is a written policy applied consistently, plus a record of what was weighed.

In this guide

• Can you use AI in tenant screening?

• Does fair housing law apply to algorithmic decisions?

• What changed in 2025 and 2026, and what did not

• How screening algorithms create risk without anyone intending it

• Which screening criteria carry the most risk?

• Who is liable, the housing provider or the screening vendor?

• What about algorithmic rent pricing?

• What documentation protects a screening decision

• Best practices for AI tenant screening

• What to ask a screening or AI vendor

• Common mistakes

• Frequently asked questions

Can you use AI in tenant screening?

Short answer: Yes. Automated screening is widely used and is not prohibited. What is regulated is the outcome, not the tool. If an automated process produces decisions that disadvantage applicants on the basis of a protected characteristic, the fact that an algorithm made the call is not a defence. The practical question is not whether you may use AI, but whether you can explain and defend the decisions it produces.

That reframing matters because most teams approach this as a technology question. It is a decision-making question. The same standards that applied when a leasing agent read an application by hand apply when a model scores it.

Does fair housing law apply to algorithmic decisions?

Short answer: Yes. In May 2024 HUD issued two guidance documents, one on tenant screening and one on advertising through digital platforms, making the position explicit: the Fair Housing Act applies to tenant screening and housing advertising including where algorithms and artificial intelligence perform those functions.

The analysis by consumer financial services counsel sets out the core points. HUD's position was that the Act applies to housing decisions regardless of who makes them and the technology used, that both housing providers and tenant screening companies have a responsibility to avoid discriminatory use of AI, and that housing providers remain responsible for ensuring their decisions comply, and may be vicariously liable, even where screening has been outsourced to a third party.

The guidance also set out principles for how screening should work: screening only for information relevant to the likelihood that an applicant will meet tenancy obligations, on the reasoning that more precise screening tends to produce less discriminatory outcomes. It addressed what screening companies should and should not offer, including that screening companies should help implement a housing provider's policy rather than effectively set it, and should offer customisation of criteria, standards, and weights rather than opaque recommendations.

Whatever happens to federal enforcement, those are sound operating principles. They describe a screening process that can be explained.

What changed in 2025 and 2026, and what did not

This is where teams are getting confused, so it is worth being precise and neutral about it.

Two conclusions follow, and neither depends on a view about whether the policy shift is right.

Federal enforcement is not the main exposure for most operators anyway. State attorneys general, state fair housing agencies, local ordinances, and private plaintiffs bring the majority of housing discrimination actions. A change in federal posture does not touch any of them.

The operational fix is identical either way. A screening process you can explain, evidence, and show was applied consistently protects you under any enforcement regime. A process nobody can describe is a liability under all of them.

How screening algorithms create risk without anyone intending it

Nobody builds a screening model on race. The risk comes from proxies and from data.

The last two are the ones most within your control and the ones most often ignored.

Which screening criteria carry the most risk?

Short answer: Credit score, eviction records, and criminal history. All three are standard, all three were specifically identified in HUD's guidance as capable of producing disparate impact, and the issue is not the criterion itself but whether you can justify the connection between it and tenancy performance.

Fair housing trainers summarising the guidance put the credit example plainly. A low credit score seems like common sense grounds for rejection, and many automated programs treat it that way, but the guidance indicated that without proper research behind the rejection there is no valid justification for it. The same reasoning was applied to eviction records and criminal history, with the observation that rejections based on eviction records were found to create disparate impact on minorities, families with children, and individuals with disabilities.

The practical translation is uncomfortable but clear. If you reject on credit, eviction history, or criminal record, you should be able to say what threshold you use, why that threshold rather than another, and what connects it to whether someone pays rent and keeps to the lease. "The system flagged it" is not an answer to any of those.

The same summary notes a second shift worth planning for: private market housing moving toward the appeals practice that already applies in federally funded housing, meaning applicants are given the reasoning behind a rejection and the decision can be revisited. Whether or not that becomes a formal requirement in your jurisdiction, a process that can explain and revisit a denial is the same process that survives a challenge.

Who is liable, the housing provider or the screening vendor?

Short answer: Both may be, but the housing provider cannot transfer the risk by outsourcing. HUD's 2024 position was that housing providers remain responsible for ensuring their decisions comply with the Act, and may be vicariously liable, even where the screening function has been contracted out.

The vendor's position on this is worth understanding, because it is contested. Screening companies have argued they are not housing providers and therefore not subject to the Fair Housing Act. Housing advocates argue that as agents of landlords and management companies they can be found liable. HUD's guidance took the advocates' side. Federal enforcement policy has since shifted, but the argument itself remains unresolved and continues in private litigation.

The operational implication is unaffected by how that argument resolves. If a screening decision is challenged, you are the party who denied the applicant. Your vendor's legal theory about its own status is not your defence.

Three things to establish in the contract, before you need them:

• Indemnification for claims arising from the vendor's process

• A right to the reasoning behind any adverse recommendation, in a form you can hand to an applicant

• Documentation of validation, meaning evidence the vendor has tested its model for disparate outcomes

What about algorithmic rent pricing?

Short answer: It is a separate exposure from screening, and it has moved faster. A number of cities and states have introduced or passed restrictions on the use of algorithmic pricing tools that draw on non-public competitor data to set rents, and there is active litigation in the area. The regulatory position varies significantly by jurisdiction and is changing.

The concern is different from screening. Screening raises discrimination questions. Algorithmic pricing raises competition questions, specifically whether landlords using a shared pricing tool fed by non-public data are effectively coordinating rents, even where no explicit agreement exists between them.

If you use a revenue management tool, three questions are worth answering now rather than later:

1. Does the tool use non-public data from other landlords, or only your own data and publicly available market information? This distinction is central to most of the restrictions.

2. Which jurisdictions are you operating in, and what have they enacted? Requirements differ by city as well as by state.

3. Can you show the pricing decision was yours, meaning the tool recommends and a person decides, with a record of that.

For how these tools work operationally, see RIOO's guide to AI revenue management and rent optimisation. This article covers the exposure rather than the capability.

What documentation protects a screening decision

The defensible position is not "our vendor said no." It is "here is our written policy, here is how this application was assessed against it, and here is the same assessment applied to everyone else."

One clarification on the adverse action notice, since two separate regimes touch it. Fair housing law concerns whether the decision discriminates. Where the screening relies on a consumer report, the Fair Credit Reporting Act separately governs what the applicant must be told, including notice that a consumer report was used, the identity of the agency that supplied it, and the applicant's right to dispute the information and obtain a free copy. Meeting one obligation does not satisfy the other, and the notice that satisfies both is more informative than the one that satisfies neither.

The applicant data behind all of this carries its own obligations around retention and deletion, which RIOO's guide to security, compliance, and data privacy covers.

One point worth stressing: a rejected applicant's file contains a Social Security number, a credit report, and income documentation for someone who never became a customer. Holding it indefinitely serves no operational purpose and creates exposure. Set a retention schedule.

Best practices for AI tenant screening

Short answer: Write the policy first, keep a person in the decision, and record what happened. Everything else follows from those three.

1. Define written screening standards before automating anything. The tool implements a policy. If no policy exists, the tool becomes the policy, and you cannot explain a decision you did not design.

2. Confirm the criteria relate to tenancy performance. For each rejection reason, be able to state what connects it to paying rent and keeping to the lease.

3. Publish the criteria to applicants before they apply, so the standard demonstrably predates the decision.

4. Review vendor validation. Ask for evidence the model has been tested for disparate outcomes, and read it rather than filing it.

5. Keep human oversight on adverse decisions. A recommendation reviewed by a person is defensible in a way an automatic denial is not.

6. Record every override with a reason. Undocumented exceptions are the weakest point in any screening process.

7. Audit outcomes on a fixed cycle. Approval and denial rates checked for unexplained disparity, so a pattern surfaces internally first.

8. Retain the decision record, and only for as long as required. The reasoning, inputs, and notice sent, held to a documented retention schedule.

9. Review state and local requirements annually. This area moves, and it moves at state and city level more than federal.

What to ask a screening or AI vendor

1. What data sources feed the decision, and how current are they?

2. Can we configure the criteria, standards, and weights, or is the model fixed?

3. Will you provide the specific reasons behind an adverse recommendation, in a form we can give an applicant?

4. Have you tested the model for disparate outcomes across protected classes, and will you share that testing?

5. What is your process when an applicant disputes an underlying record?

6. What does the contract say about indemnification if a decision is challenged?

7. Can we export our decision records, including the inputs, at any time?

8. Which jurisdictions' requirements have you built for?

Question two is the one that separates a tool from a decision-maker. A screening product that cannot be configured is setting your policy rather than implementing it.

Common mistakes

Frequently asked questions

1. Can you use AI for tenant screening?
Yes. Automated screening is not prohibited. What is regulated is the outcome rather than the tool. If the process produces decisions that disadvantage applicants on the basis of a protected characteristic, the involvement of an algorithm is not a defence.

2. Can AI automatically reject a rental applicant?
It can produce a denial recommendation, and some systems are configured to act on it without review. Whether that is advisable is a separate question from whether it is possible. An automatic denial nobody reviewed is harder to explain, harder to appeal, and harder to defend than the same denial with a documented human decision behind it. Some jurisdictions are also moving toward requiring human review of consequential automated decisions.

3. Does the Fair Housing Act apply to AI?
HUD's May 2024 guidance stated that the Act applies to tenant screening and housing advertising including where algorithms and artificial intelligence perform those functions, and that the Act applies to housing decisions regardless of who makes them or what technology is used. The statute itself has not changed, though federal enforcement policy has shifted since.

4. Which tenant screening criteria create the most fair housing risk?
Credit score, eviction records, and criminal history. All three were specifically identified in HUD's guidance as capable of producing disparate impact where the rejection is not supported by research connecting the criterion to tenancy performance. Rejections based on eviction records were noted as creating disparate impact on minorities, families with children, and individuals with disabilities.

5. Is the landlord or the screening company liable if an algorithm discriminates?
HUD's position was that housing providers remain responsible for their decisions and may be vicariously liable even where screening is outsourced. Whether screening companies are themselves subject to the Act is contested and continues in litigation. Practically, the housing provider is the party that denied the applicant.

6. Has the law on AI tenant screening changed?
The Fair Housing Act has not changed. Federal enforcement policy has, following a Department of Justice rule issued in December 2025 that civil rights organisations have criticised as substantially narrowing disparate impact liability. State fair housing statutes, local ordinances, and private litigation are unaffected.

7. Is algorithmic rent pricing legal?
It varies by jurisdiction and is changing. A number of cities and states have introduced or passed restrictions, particularly on tools that use non-public competitor data to set rents, and there is active litigation. Confirm the position for each jurisdiction you operate in.

8. What should you tell an applicant who was denied by an automated system?
The specific reasons for the decision, in terms they can act on and dispute. Where a consumer report was used, the Fair Credit Reporting Act separately governs what the notice must contain. A composite score with no breakdown does not meet either standard.

9. What documentation protects a screening decision?
Published criteria available before application, a written policy naming what is weighed, an adverse action notice with specific reasons, a record of the data relied on, a log of any overrides with reasons, and periodic review of approval and denial outcomes.

10. How long should you keep rejected applicant files?
Only as long as a documented retention schedule requires. The file contains a Social Security number, a credit report, and income documentation for someone who never became a customer. Requirements vary by jurisdiction, so confirm yours and delete on schedule.

The useful way to think about this is that automation did not change the standard. It changed how easy it is to fail the standard without noticing.

A leasing agent applying a bad rule inconsistently is a problem you can see. A model applying the same bad rule perfectly consistently across ten thousand applications is a problem you will not see until someone else finds it. The protection is the same as it always was: a written policy, applied the same way every time, with a record of what was decided and why.

This article provides general information and is not legal advice. Fair housing obligations, screening requirements, consumer reporting rules, and restrictions on algorithmic pricing vary by state and locality. The federal enforcement position changed in December 2025 and state and local rules continue to move, so confirm the current position for your jurisdictions with qualified counsel before relying on anything here.