A $2.275M class settlement takes the algorithm's score away: Louis v. SafeRent and the five-year rollback of tenant-screening scores for voucher applicants

The problem

In May 2021, Mary Louis, a Black housing-voucher holder in Massachusetts, was told by a property management company that “the third party service we utilize to screen all prospective tenants has denied your tenancy,” and that she needed a score of at least 443 for her application to be accepted, with no explanation and no way to appeal (source).

The service was SafeRent Solutions’ SafeRent Score, a product that uses rental-applicant data and proprietary scoring models to assign a three-digit value and an approve/decline recommendation to the housing provider (source). On 25 May 2022, Mary Louis, Monica Douglas, and the Community Action Agency of Somerville filed a class action in the District of Massachusetts alleging the score discriminated against housing-voucher applicants by race and income under the Fair Housing Act and Massachusetts law (source). Those were allegations, never adjudicated findings (source).

What the court approved (20 November 2024)

Judge Angel Kelley granted final approval of the class settlement and certified two settlement classes of Massachusetts voucher applicants denied housing because of their SafeRent Score, in Louis v. SafeRent Solutions, No. 1:22-cv-10800 (source). The signed orders and the settlement agreement sit on the federal docket (source).

The money reconciles inside the record: a total consideration cap split into a class fund plus a fees-and-costs cap, with service awards to each named plaintiff (source).

The outcome

Total settlement consideration, capped: $2,275,000. That is a $1,175,000 class Settlement Fund plus a $1,100,000 fees-and-costs cap (source). A separate order awards the fees-and-costs to the cent, $1,077,547.68 in fees and $22,452.32 in costs, plus $10,000 service awards to each named plaintiff (source).

The score is switched off for voucher applicants, for five years. SafeRent may not put a SafeRent Score, or any accept/decline recommendation based on a tenant-screening score, on reports for housing-voucher applicants; the landlord instead gets a report with the underlying information and decides (source). Any successor score for voucher-holders must first be validated by the National Fair Housing Alliance or an agreed equivalent (source).

What this is NOT

This is not a finding of discrimination: the settlement carries an express non-admission clause and no court ever ruled that SafeRent discriminated, and SafeRent states it continues to believe the SafeRent Scores comply with all applicable laws (source). It is also not, in the court’s own words, an “artificial intelligence” case: the court papers say “algorithm,” and the AI label is the press’s (source).

How this was verified

Checked live on 2026-08-14. Every headline figure was re-confirmed against independent reporting (Fortune) and plaintiffs’-counsel case summaries (Cohen Milstein, National Consumer Law Center), and the underlying orders, settlement agreement, and fee order sit on the public CourtListener docket for No. 1:22-cv-10800. The cent-level fee split ($1,077,547.68 + $22,452.32) comes from the court’s separate fee order on that docket; the $443 score threshold and the “third party service” quote come from the Guardian’s reporting, read here via an archived full-text relay because the Guardian is not directly fetchable from this host, and that limit is stated where the figure appears. No deployer-client exists to confirm an outcome with, and none was contacted: this is an independent read of the public court record, which is the correct standard for a court-origin case file.

Related case files

The five-year ban here mirrors the FTC’s court-ordered five-year prohibition on Rite Aid’s facial-recognition system, another automated tool switched off by an order rather than a finding of intent. The screen-out pattern also runs through the EEOC’s consent decree over iTutorGroup’s age-screening software, where an automated filter, not a human, made the exclusion. And the “algorithm, not AI” framing tension recurs in the UK DWP’s machine-learning fraud model, a system scoring low-income benefit claimants where the label mattered as much as the math.

Sources

Tiers: Tier 1 is the primary court record; Tier 2 is independent reputable press naming the parties; Tier 3 is a party’s own counsel or PR. Figures checked live 2026-08-14.

1. CourtListener, “Louis v. SafeRent Solutions, LLC, 1:22-cv-10800 (D. Mass.)” docket (orders, settlement agreement, and fee order), accessed 2026-08-14 (Tier 1, primary federal court record). https://www.courtlistener.com/docket/63335697/louis-v-saferent-solutions-llc/
2. Fortune, “Renter-scoring firm agrees to pay $2.2 million to settle case accusing its algorithm of discriminating on race and income,” 2024-11-21 (Tier 2, independent press naming the parties and figures). https://fortune.com/2024/11/21/renter-scoring-saferent-million-settle-case-algorithm-discriminating-race-income/
3. Cohen Milstein (plaintiffs’ counsel), “Louis, et al. v. SafeRent Solutions, et al.” case study, accessed 2026-08-14 (Tier 3, party’s own counsel; used for the settlement terms it states on record). https://www.cohenmilstein.com/case-study/louis-et-al-v-saferent-solutions-et-al/
4. AI Commission relay of The Guardian, “She didn’t get an apartment because of an AI-generated score,” 2024-12 (Tier 2 underlying reporting via archived relay; the Guardian original is not fetchable from this host, gap recorded). Used for the 443 score threshold and the “third party service” quote. https://www.aicommission.org/2024/12/she-didnt-get-an-apartment-because-of-an-ai-generated-score-and-sued-to-help-others-avoid-the-same-fate/