The AI tenant score that priced out voucher holders: SafeRent paid $2.275M and agreed to stop scoring housing-voucher applicants for five years (Louis v. SafeRent, D. Mass., 2024)

The problem

Mary Louis, a Black woman who uses a federal housing voucher, applied for a Massachusetts apartment in spring 2021 and received an email saying a “third-party service” had denied her tenancy (source). That third-party service was an algorithm designed to score rental applicants, the SafeRent Score (source). Louis had sent two landlords’ references showing she had paid rent early or on time for 16 years, even though she did not have a strong credit history (source). She ultimately moved into an apartment she found on Facebook Marketplace that was $200 more expensive and in a less desirable area (source).

Louis and co-plaintiff Monica Douglas alleged they were denied rental housing because of their SafeRent Score, a score derived from SafeRent’s algorithm-based screening software (source). Their suit alleged the scores produce an unlawful disparate impact against Black and Hispanic applicants because the algorithm relies on factors such as credit history and non-tenancy-related debts that disproportionately disadvantage those groups, while failing to weight the fact that a HUD housing voucher makes such tenants more likely to pay their rent (source).

What was built

The SafeRent Score is a proprietary algorithm-based tenant-screening score that landlords use to accept or decline rental applicants (source). The plaintiffs’ central objection was not that a human made a biased call but that an opaque model did: the score drew on credit history and non-tenancy debts rather than the applicant’s actual rent-paying record (source).

The Justice Department made the legal stakes explicit. On 9 January 2023 the DOJ and HUD filed a Statement of Interest to explain that the Fair Housing Act applies to algorithm-based tenant-screening systems (source). “Housing providers and tenant screening companies that use algorithms and data to screen tenants are not absolved from liability when their practices disproportionately deny people of color access to fair housing opportunities,” Assistant Attorney General Kristen Clarke of the Civil Rights Division said (source). In the same filing the United States set out the standard for pleading disparate-impact claims under the Fair Housing Act and clarified that the Act applies to companies providing residential screening services (source). Independent legal-industry reporting confirmed the filing: Orrick’s InfoBytes noted that “on January 9, the DOJ and HUD announced they filed a joint statement of interest in a pending action alleging discrimination under the Fair Housing Act (FHA) against Black and Hispanic rental applicants based on the use of an algorithm-based tenant screening system” (source).

The outcome

The outcome is fixed by the Court’s own record. In its signed Final Approval Order (Doc 135) the Court records “Case 1:22-cv-10800-AK Document 135 Filed 11/20/24,” states that “the Motion for Final Approval of Class Action Settlement is GRANTED under Rule 23(e)(2),” and closes “Dated: BY THE COURT: November 20, 2024 /s/ Angel Kelley HONORABLE ANGEL KELLEY UNITED STATES DISTRICT JUDGE” (source). The executed Class Action Settlement Agreement approved by that order fixes the amount at “two million two hundred and seventy five thousand dollars ($2,275,000.00),” of which SafeRent “will establish a settlement fund in the amount of $1,175,000 in monetary relief” and pays a “separate award of attorneys’ fees and costs up to $1,100,000, with any unawarded funds to be added to the settlement fund” (source). Independent reporting corroborates the outcome while rounding the figure: Bloomberg Law reported that the renters “won final approval of their $2.3 million class settlement” (source), and the Associated Press, carried by Fortune, reported that “a federal judge approved a settlement in the lawsuit … on Wednesday, with the company behind the algorithm agreeing to pay over $2.2 million” (source).

The injunctive relief is what makes the case notable, and the Settlement Agreement sets it out precisely. SafeRent “agrees to comply with the following practices for a period of five (5) years” (source). First, SafeRent “will cease including a SafeRent Score or an accept/decline recommendation based on a tenant screening score in any of its score reports generated for housing providers who subscribe to or purchase SafeRent’s ‘Affordable Model’” (source). Any future tenant-screening score it wishes to use in those reports for voucher applicants must have “been validated for its use for applicants with housing vouchers by the National Fair Housing Alliance or any similar organization agreed to by the Parties” (source). Bloomberg Law summarised the bar in the same terms, reporting that during the next five years SafeRent will not include “a SafeRent Score or an accept/decline recommendation based on a tenant screening score” (source); the Court’s Final Approval Order in turn “expressly retains continuing and exclusive jurisdiction over the parties … for a period of five years from the date SafeRent confirms it has made the practice changes outlined in Section 3.5 of the Agreement” (source). SafeRent admitted no fault and said in a statement that while it “continues to believe the SRS Scores comply with all applicable laws, litigation is time-consuming and expensive” (source).

“There is no beating the system. The system is always going to beat us,” Louis said of automated screening (source). Her attorney Todd Kaplan framed the ruling as a warning to the industry: management companies and landlords “need to know that they’re now on notice, that these systems that they are assuming are reliable and good are going to be challenged” (source).

Weakest load-bearing source. The precise amount, judge, docket, date and injunctive terms now rest on the Court’s own signed Final Approval Order and the executed Settlement Agreement, both Tier 1 (source). The honest limit is the host, not the documents: these court records were retrieved from the plaintiffs’ counsel’s own web server (cohenmilstein.com) rather than from PACER or a court domain, so a skeptical reader is relying on that party to have posted the Court’s order and the executed agreement faithfully; the file contents (docket header, the judge’s signature block, and the “$2,275,000.00” figure) are the Court’s and SafeRent’s, not the firm’s characterisation (source). The DOJ position rests on two U.S. Department of Justice pages, which are authoritative primaries but not independent of each other, now backed by one independent report of the filing from Orrick’s InfoBytes (source).

How this was verified

Method: desk verification against the public record, 3 September 2026. The settlement amount, judge, docket, date and five-year injunctive terms were quoted verbatim from the Court’s own signed Final Approval Order (Doc 135, 11/20/24) and the executed Class Action Settlement Agreement (Ex. 1 to Doc 114, 03/28/24), both Tier 1, downloaded and text-extracted this session; independent reporting (Bloomberg Law, “$2.3 million”; the Associated Press via Fortune, “over $2.2 million”) corroborates the outcome while rounding the figure. The DOJ/HUD Statement of Interest and its Fair Housing Act position were quoted from two U.S. Department of Justice pages (Tier 1) and one independent report of the filing (Orrick InfoBytes, Tier 2). Mary Louis’s denial narrative was quoted from the Associated Press. Every cited page was archived to web.archive.org and saved locally. No confirmation was sought from SafeRent or any party: the public record either corroborates a figure or it does not.

Related case files

• Apple Card and Goldman Sachs, where a regulator found no unlawful discrimination in an algorithmic credit decision — the mirror image: a fair-lending algorithm that survived scrutiny, where SafeRent’s did not.
• Earnest Operations and the Massachusetts AG, an AI-underwriting fair-lending settlement — another U.S. settlement over an algorithm that scored people for a financial decision.
• The Netherlands childcare-benefits risk algorithm and its 2.75M euro fine — a government scoring system that discriminated against protected groups, ruled unlawful.
• The Hague District Court striking down the SyRI welfare-fraud algorithm — a court holding an opaque risk-scoring system unlawful for the people it profiled.

Sources

1. U.S. Department of Justice, Office of Public Affairs · “Justice Department Files Statement of Interest in Fair Housing Act Case Alleging Unlawful Algorithm-Based Tenant Screening Practices” · 9 January 2023 · Tier 1 · https://www.justice.gov/archives/opa/pr/justice-department-files-statement-interest-fair-housing-act-case-alleging-unlawful-algorithm (archived: http://web.archive.org/web/20250203060255/https://www.justice.gov/opa/pr/justice-department-files-statement-interest-fair-housing-act-case-alleging-unlawful-algorithm)
2. U.S. Department of Justice, Civil Rights Division · “Louis et al. v. SafeRent et al. (D. Mass.)” case page · 9 January 2023 · Tier 1 · https://www.justice.gov/crt/case/louis-et-al-v-saferent-et-al-d-mass (archived: http://web.archive.org/web/20260525020609/https://www.justice.gov/crt/case/louis-et-al-v-saferent-et-al-d-mass)
3. Bloomberg Law · “SafeRent’s $2.3 Million Deal in AI Screening Tool Suit Approved” · 21 November 2024 · Tier 2 · https://news.bloomberglaw.com/us-law-week/saferents-2-3-million-deal-in-ai-screening-tool-suit-approved (archived: http://web.archive.org/web/20260903113517/https://news.bloomberglaw.com/us-law-week/saferents-2-3-million-deal-in-ai-screening-tool-suit-approved)
4. Associated Press (via Fortune), Jesse Bedayn · “Renter scoring firm agrees to pay $2.2 million to settle case accusing its algorithm of discriminating on race and income” · 21 November 2024 · Tier 2 · https://fortune.com/2024/11/21/renter-scoring-saferent-million-settle-case-algorithm-discriminating-race-income (archived: http://web.archive.org/web/20260827045816/https://fortune.com/2024/11/21/renter-scoring-saferent-million-settle-case-algorithm-discriminating-race-income/)
5. Cohen Milstein Sellers & Toll (plaintiffs’ counsel, a party to the case) · “Louis, et al. v. SafeRent Solutions, et al.” case study · 20 November 2024 · Tier 2 · https://www.cohenmilstein.com/case-study/louis-et-al-v-saferent-solutions-et-al/ (archived: http://web.archive.org/web/20260903112956/https://www.cohenmilstein.com/case-study/louis-et-al-v-saferent-solutions-et-al/)
6. U.S. District Court, District of Massachusetts · “Final Approval Order,” Louis et al. v. SafeRent Solutions, No. 1:22-cv-10800-AK, Doc 135 (signed /s/ Angel Kelley) · 20 November 2024 · Tier 1 · https://www.cohenmilstein.com/wp-content/uploads/2022/05/SafeRent-Final-Approval-Order-11-20-2024.pdf (archived: http://web.archive.org/web/20260903140711/https://www.cohenmilstein.com/wp-content/uploads/2022/05/SafeRent-Final-Approval-Order-11-20-2024.pdf)
7. U.S. District Court, District of Massachusetts · “Class Action Settlement Agreement and Release,” Ex. 1 to Memorandum in Support of Unopposed Motion to Settle, No. 1:22-cv-10800-AK, Doc 114 · 28 March 2024 · Tier 1 · https://www.cohenmilstein.com/wp-content/uploads/2022/05/SafeRent-Memo-in-Support-of-Unopposed-Motion-to-Settle-and-Certify-the-Classes-and-Exhibits-March-28-2024.pdf (archived: http://web.archive.org/web/20260903140738/https://www.cohenmilstein.com/wp-content/uploads/2022/05/SafeRent-Memo-in-Support-of-Unopposed-Motion-to-Settle-and-Certify-the-Classes-and-Exhibits-March-28-2024.pdf)
8. Orrick, InfoBytes · “DOJ, HUD say Fair Housing Act extends to algorithm-based tenant screening” · 13 January 2023 · Tier 2 · https://infobytes.orrick.com/2023-01-13/doj-hud-say-fair-housing-act-extends-algorithm-based-tenant-screening/ (archived: http://web.archive.org/web/20260412055723/https://infobytes.orrick.com/2023-01-13/doj-hud-say-fair-housing-act-extends-algorithm-based-tenant-screening/)