90 incidents on record · 2026 Headlights Incident reports by Ellie Harris · Melbourne
10 new this week Library last updated 13 July 2026
← The incident library
HD-INC-031
Housing & real estate · United States · 2024 · Algorithmic discrimination

A tenant-screening algorithm allegedly scored renters with housing vouchers out of homes, and a $2.275M settlement stopped it scoring them in Massachusetts

By Ellie Harris · Filed Application denied May 2021; suit filed May 2022

Alleged: SafeRent Solutions (formerly CoreLogic Rental Property Solutions) developed or deployed the AI system implicated in this incident. Details are drawn from public reports; parties are presumed innocent of any wrongdoing not established by an official finding.

A tenant-screening algorithm allegedly scored renters with housing vouchers out of homes, and a $2.275M settlement stopped it scoring them in Massachusetts

What happened

To rent an apartment in much of the United States, an applicant is scored before they are met. SafeRent Solutions, formerly CoreLogic Rental Property Solutions, sold landlords a screening product that turned an applicant into a single number, the SafeRent Score, on a scale from 200 to 800, and a recommendation to accept or decline. The score leaned heavily on credit history, debts, and records that, the plaintiffs argued, have little to do with whether someone will pay rent, and, the case alleged, it did not properly account for the one thing that most directly guaranteed the rent would be paid: a government housing voucher, which on average covers more than 70 percent of the monthly rent, paid directly to the landlord.

In May 2021, Mary Louis, a Black woman in Massachusetts, applied to rent an apartment using a housing voucher. Her score came back low and her application was denied, despite a voucher that would have covered most of the rent. In May 2022 she brought a class action, joined by a second named plaintiff, Monica Douglas, also a Black woman in Massachusetts, and by the Community Action Agency of Somerville, a non-profit whose work is placing voucher holders in housing and which said the tool made that work impossible. They argued that SafeRent’s scoring had a disproportionate, adverse effect on Black and Hispanic applicants and on voucher holders, in violation of the federal Fair Housing Act and Massachusetts law. The institutional plaintiff mattered: it turned two individual denials into evidence of a pattern.

The case did not go to a verdict. In November 2024, US District Judge Angel Kelley approved a settlement of $2,275,000, of which $1,175,000 went to cash payments for class members. The more consequential term was not the money. For five years, in Massachusetts, SafeRent agreed that for applicants using housing vouchers it would not display a SafeRent Score and would not issue an accept-or-decline recommendation, unless its scoring were independently validated as non-discriminatory. The injunction was not nationwide, which sharpens rather than softens the precedent: a screening vendor whose product is used by landlords across the country was made to change its practice for one state’s voucher holders, with independent validation required before any new scoring model could be used on them. The product was allowed to keep scoring everyone else; for voucher holders in Massachusetts, the number that had been deciding their housing was switched off.

What an auditable version would have shown

A screening score is a verdict delivered without an explanation. Louis was told no; she was not told that a model had weighed her credit record above her voucher, or how applicants like her fared compared with everyone else. The disparate impact the case turned on was a property of the system as a whole, visible only in the distribution of its scores across groups, and that distribution was not something an individual applicant, or the landlord clicking accept or decline, could see. An auditable version would record each scoring decision, the inputs, the model version, the score, the recommendation, as a signed entry, and would compute signed aggregate metrics over those entries: approval and denial rates for voucher holders against non-voucher applicants, and across race where lawfully measurable, bound cryptographically to the underlying records. A MetricRecord of that kind is what turns “we believe the score is fair” into a claim a regulator or court can check, and what would have surfaced the impact on voucher holders as a measurable pattern rather than a litigated allegation.

Where the gap was

The underlying gap was that a model could deny someone housing without ever having to account for whether it denied their group at a higher rate. The score relied on factors correlated with race and income while discounting the voucher that de-risked the tenancy, and nothing in the product measured the result. A neutral-looking formula that falls harder on a protected class is the textbook shape of disparate impact, and it is invisible until someone measures the outputs by group. A ConductRecord captures each scoring decision as it happens; a MetricRecord aggregates those decisions into signed, verifiable measures of who is approved and who is turned away. Together they let a screening company demonstrate fairness, or detect the absence of it, before a class action does the measuring for them. The settlement’s core remedy, no score and no recommendation for voucher holders absent validation, is in effect an admission that the impact had never been adequately measured.

What governance should have looked like

A score that decides access to housing should be held to the same standard as any other consequential automated decision: its maker should be able to show, from signed records, how it performs across protected groups, and should withhold a recommendation it cannot stand behind rather than issue one it has never tested. The factors that feed the score matter too; a model that discounts a guaranteed subsidy while leaning on credit history is making a choice with a predictable distributional effect, and that choice should be measured, not assumed neutral.

SafeRent reduced people to a number and sold landlords the verdict. What it lacked was a verifiable record of those verdicts and a signed measure of how they fell across voucher holders and racial groups. Producing both is ordinary in a mature governance framework, and a great deal cheaper than a settlement that ends with the product being switched off for an entire class of applicants.

The reference implementation of ConductRecord and MetricRecord is open source. It lives at github.com/saffronandindia/headlights-oss, Apache 2.0 licensed and free to install. The repository is public now.

Sources

The mailing list

Fresh incident reports every week. One email to match.

We add new incidents to the library regularly, and send a single short email each week with what's new. The library stays free and open; this is just how you keep up with it.

No tracking. Unsubscribe in one click.

The record

An auditable system would have produced a signed, tamper-evident record the moment this happened: what the system did, the version that did it, the basis it acted on, and the action taken, and SafeRent Solutions (formerly CoreLogic Rental Property Solutions) could have produced it on demand.

This is the record the system as deployed did not produce in a signed, auditable form.

What this teaches
Capture what happened when it happens
What the system did, the version that did it, the basis it acted on, and the action taken, recorded at the moment, not reconstructed after.
Sign it, so no one has to trust the record-keeper
A tamper-evident entry. Edit it later and the signature breaks. The record does not ask for the benefit of the doubt.
Make it verifiable by anyone
A court, a regulator, a customer's lawyer can check the record themselves, without taking the company, or us, at our word.

Headlights summarises publicly reported AI incidents. All summaries are independently written, attributed to their original sources, and intended for research and educational purposes. Allegations are identified as such until established through official findings.

Last reviewed June 2026. This report is based on the sources listed above and reflects information available at the time of review; later developments may not be captured. Where a person is described as charged with or alleged to have done something, that allegation is unproven unless a conviction or a court or regulatory finding is stated. Headlights publishes journalism and commentary, not legal advice.

Want to write back?

Direct to my inbox.

ellie@useheadlights.com →