SafeRent Tenant Screening: What the Settlement Actually Says

The SafeRent case is cited constantly and described wrongly almost as often. It was not a regulator’s enforcement action. It was a private class action that settled, and the terms it settled on are more specific — and narrower — than the summaries suggest.

What was filed, and by whom

The case is Louis v. SafeRent Solutions LLC, filed in 2022 in the United States District Court for the District of Massachusetts, docket number 1:22-cv-10800. The named plaintiffs were rental applicants who used Housing Choice Vouchers. Their allegation was that the SafeRent Score — a numerical tenant-screening output sold to landlords and property managers — produced disproportionately low scores for Black and Hispanic applicants, and that it did so partly because it weighted credit history and non-tenancy debt while giving no weight to the fact that a voucher would pay a large share of the rent.

The legal theory was disparate impact under the federal Fair Housing Act, together with Massachusetts anti-discrimination law (M.G.L. c. 151B) and the state consumer-protection statute (c. 93A). Disparate impact matters here: the plaintiffs did not have to allege that anyone intended to discriminate. The claim was that a facially neutral scoring practice produced a disproportionate effect and was not justified by a substantial, legitimate interest that could not be served a less discriminatory way.

In 2023 the court declined to dismiss the Fair Housing Act claims against SafeRent, allowing the case to proceed against a screening company rather than only against the landlords who used its output. A ruling on a motion to dismiss assumes the allegations are true and decides only that they state a claim if proved. It is not a finding that the score discriminated.

The settlement terms

The parties settled and the court approved the settlement in November 2024. Two components, and the second is the one that matters operationally.

• Money. A settlement fund of approximately $2.275 million for the class. As with every class settlement, the payment is not an admission of liability, and the agreement says so.
• A five-year restriction on the product. For five years, SafeRent agreed to stop providing a SafeRent Score, and to stop issuing an accept-or-decline style recommendation, for applicants using housing vouchers in the covered jurisdictions. If it wants to reintroduce a score for that population, the agreement requires the model to be validated by an independent third party acceptable to the plaintiffs first.

That second term is the interesting one, because it is a remedy shaped around a specific population rather than around the model as a whole. The settlement did not require SafeRent to retrain, retire or open up the score generally. It carved out the group whose treatment was in issue and put a validation gate in front of any attempt to bring the score back for them. Anyone reading this as “the algorithm was banned” has the shape wrong.

Where the government actually appears

The United States Department of Justice and the Department of Housing and Urban Development jointly filed a statement of interest in the case in January 2023. A statement of interest is a brief: it tells the court what the United States thinks the legal standard is. Here the position was that the Fair Housing Act’s disparate-impact framework applies to algorithmic screening tools, and that a company that supplies a score used in tenancy decisions can be within the statute’s reach. The DOJ Civil Rights Division lists its housing filings publicly, and this one sits there among them — the Justice Department’s Housing and Civil Enforcement Section case list is the primary place to read it.

That is the whole of the federal involvement. Neither agency was a party, neither imposed a penalty, and neither settled anything. The separate action people often merge into this one is the Federal Trade Commission and Consumer Financial Protection Bureau case against TransUnion Rental Screening Solutions, announced in October 2023 and resolved with a $15 million payment over alleged Fair Credit Reporting Act failures — a different company, a different statute, and an accuracy claim rather than a discrimination claim. The FTC’s cases and proceedings library holds that record.

What the settlement does not decide

A settlement decides nothing about the law. There is no judicial holding in this case that the SafeRent Score violated the Fair Housing Act, no finding on the statistical evidence, and no ruling on whether the business justification SafeRent advanced was substantial or whether a less discriminatory alternative existed. Those questions were live when the parties settled and they stayed live.

Nor does it resolve the broader question that every algorithmic screening defendant raises: whether a vendor that supplies a score, rather than the landlord who acts on it, is the right defendant. The court’s refusal to dismiss suggests one district judge thought the claim could proceed on those facts. That is persuasive at best, binding nowhere, and the same argument is being run in employment screening — see the Workday hiring litigation for the same vendor-liability question under Title VII and the ADEA.

What carries to other screening models

Three things generalise, and they generalise because they are about model design rather than about housing law.

• An input that proxies for a protected characteristic is the exposure. The allegation was not that race was a feature. It was that credit and debt history correlate with race and that the model leaned on them. Removing the protected attribute from the feature set does not remove the claim.
• Ignoring a mitigating fact is a design choice. The complaint made a great deal of the score not accounting for the voucher — a guaranteed subsidy directly relevant to whether rent gets paid. Omissions are as reviewable as inclusions.
• A single number invites reliance. A score with an accept/decline recommendation attached is much harder to defend as decision support than a set of underlying facts a human weighs, and the settlement’s remedy targeted exactly that output.

If you are looking at the same problem from the employment side rather than the housing side, the statutory frame differs but the model critique is identical; the AI hiring law overview and the EEOC’s iTutorGroup settlement cover where an automated screen becomes a regulator’s problem rather than only a plaintiff’s.