I have been screened more times than I have been interviewed.
Three countries, more apartments than I care to count, a couple of jobs, one car loan. Every time the same ritual: application, fee, and the box you tick giving a company you have never heard of permission to go find out about you. And every time, nothing from me — I never asked who ran the check or what came back. Once I was turned down for a place I badly wanted and did what most people do: assumed the file was accurate, decided the problem was me, moved on.
It never occurred to me that the document deciding my housing might have counted the same thing twice.
The nine-day-old fact that changes the tense of this story
On July 9, 2026, the Federal Trade Commission announced that a tenant screening company called RentGrow would pay $2.25 million to settle Federal Trade Commission allegations that it violated the Fair Credit Reporting Act and the FTC Act. Most coverage treated that as the ending. It was a proposed order.
Nine days ago it stopped being one. On August 12 the Justice Department announced that a federal court has entered a stipulated order resolving the case the FTC investigated and referred; it sits on the District of Columbia docket as case 1:26-cv-02415. The verb changed from would to did, and almost nobody noticed.
One thing to fence off first. Everything the government says here is an allegation: the order records that RentGrow neither admits nor denies any of the complaint, and the Commission files a case when it has "reason to believe" the law is being broken — a threshold, not a finding. No judge weighed the evidence. What follows is what the United States alleged, in writing, and what the company agreed to be bound by without conceding a word of it.
Three lines on a landlord's screen, two cases in real life
A tenant screening report is not a description of you. It is a verdict about you. The complaint says these reports pull credit, criminal, eviction, rental and address history, then return a determination of whether the applicant meets, conditionally meets, or does not meet the landlord's criteria. RentGrow furnishes them to thousands of client rental property owners and property management companies.
The core allegation is narrower, and better, than the headline: that the company failed to follow reasonable procedures to prevent the inclusion of duplicative case records and multiple entries for the same criminal or eviction action.
Why would that happen? One court case leaves more than one footprint — the same conviction in a county and a state database, and a single eviction that, as the complaint puts it, "might involve different events such as a new case filing and a judgment of possession." Two rows. One eviction.
Here is the part that stopped me: the upstream data was often fine. Records arrived from other agencies "with clear headers and with different events in a particular record grouped together," and the complaint says RentGrow "disregarded that formatting in favor of its own." Nothing was invented. The software took clean input and printed it in a shape that read as more.
The specifics are worse than the abstraction. One report showed three eviction events as three records when the applicant "actually had only two records — two of the displayed events were separate parts of the same eviction case." Another, in small fonts and split across pages, "listed six records… In reality, the consumer only had two records." A third listed 15 separate criminal charges across seven criminal records, on which basis RentGrow determined the applicant did not meet the criteria. He disputed it, saying none were his. All 15 came off, and the determination "was reversed from denial to acceptance."
The answer changed from no to yes — not because anything about him changed, but because he knew to push back. The complaint also alleges the company "knew it was reporting these duplicative records and failed to implement any procedures at all" to stop it, and fixed that only after the FTC started investigating.
And one allegation I cannot stop thinking about: when a renter successfully disputed a record and had it changed or deleted, RentGrow told the renter it had notified the property manager — "but RentGrow instead told those property owners that there was no change."
You win. Nobody tells the person holding the keys.
Two and a quarter million dollars, and where it lands
The order enters a $2,250,000 civil penalty judgment and directs payment to the Treasurer of the United States within seven days. I searched the eighteen pages for "redress." It appears zero times: no consumer fund, no notice to past applicants, no mechanism that reopens a denial. The Commission's stated reasons for settling are future compliance, deterrence, consistency with past orders and litigation cost. Nobody's lost apartment is on that list. A right-of-center outlet called it a $2.25 million civil penalty judgment — a penalty, which is a different object from a payment to a harmed person.
Before you conclude the agency chose not to bother: in 2021 the Supreme Court held that Section 13(b) does not authorize the Commission to seek, or a court to award, equitable monetary relief such as restitution or disgorgement. The civil-penalty route that remains pays the Treasury by statute. The FTC did not refuse to get renters their money back — the remedy it has does not reach the people the case is about.
Which we know, because three years ago it did. In October 2023 the FTC brought a near-identical allegation — "multiple entries for the same eviction case" — against Trans Union's rental screening arm, with the Consumer Financial Protection Bureau as co-plaintiff. That case produced $11 million, which will be used to compensate consumers. Same problem, same industry, different arithmetic — decided by who was in the room.
The FTC's standing guidance to this industry has warned for years that screening reports with multiple entries for the same offense raise compliance concerns. Nobody had to discover this. And RentGrow is no garage startup: the CFPB's registry lists it as a wholly owned subsidiary of Yardi Systems, one of the largest property-management software companies in the country.
While that was happening, twenty-one documents left the building
Now the timing, which I would not believe if it weren't in the Federal Register.
Two notices this year withdrew a combined twenty-one HUD guidance documents: eight in an April 6, 2026 Fair Housing and Equal Opportunity notice with an effective date of withdrawal of September 17, 2025, thirteen more in a July 17, 2026 Office of General Counsel notice. They have been "removed from active use and should not be relied upon as authoritative." One of the thirteen is the 2016 guidance titled Application of Fair Housing Act Standards to Use of Criminal Records — the document telling landlords a criminal record is not, by itself, a lawful reason to slam the door. Its 2022 implementation memo went in the April batch.
To be precise: HUD repealed nothing. The Fair Housing Act is untouched, and the notices say violations "continue to be subject to enforcement by the Department." What changed is the interpretive scaffolding, withdrawn under criteria including, in HUD's words, that "guidance imposes additional compliance burdens."
Both sides read the same paperwork and agreed on what it meant. The National Low Income Housing Coalition called the result heightened uncertainty for tenants; the National Apartment Association called the same withdrawals promising developments for the rental housing industry. One event, from opposite ends of a lease.
And there is more. In January, HUD proposed a rule that would eliminate all HUD regulations governing disparate impact liability — the doctrine that catches a neutral-looking rule with a lopsided result, which is exactly the shape of a scoring system. Meanwhile the other agency that might have shown up here withdrew its data broker rulemaking in May 2025, deciding it was not necessary or appropriate at this time.
The industry's answer deserves a hearing, and it has one
Let me put the strongest case against my own argument, because it is not stupid.
The Consumer Data Industry Association told the FTC that accuracy complaints here are vanishingly rare: only 0.35% of one CDIA member company's resident screening reports ever receive a consumer inquiry about content accuracy, and fewer still would have changed the outcome. Its second argument points at the courthouse rather than the vendor: courts must supply "adequate matching identifiers at scale," and where they don't, reports suffer.
That second point is not spin, and a neutral referee confirms it. The Urban Institute took a decade of Pennsylvania court records, tried to count how many people have an eviction record, and found the number varied by hundreds of thousands, depending on our assumptions; the records lack unique identifiers. Matching a court file to a human being is hard, and anyone who pretends otherwise is selling something.
But notice what that defense does not cover. The allegation here is not a matching error; it is that clean, correctly grouped records came in and were printed as separate ones. And the honest answer to "only 0.35%" is that a renter who is never shown the report cannot inquire about it. A survey of tenant lawyers found the most common landlord response to a dispute was to ignore the existence of the dispute and reject the renter — 86% of respondents reported seeing it. Those are practitioner observations, not an audited error rate, drawn from 253 responses from 35 states and the District of Columbia. Scope them that way. They still describe a market where complaining is a bad strategy — a strange foundation for a complaint rate.
The Dutch made the rejection letter explain itself
Cross the North Sea and the same problem sits inside a different set of rules.
Start with the statute. Article 2 of the Wet goed verhuurderschap — the Good Landlordship Act — makes a landlord run a transparent selection procedure, publish objective criteria, and then do something no American law asks of anyone: het motiveren van de keuze voor de gekozen huurder aan de afgewezen kandidaat-huurders. Justify the choice of tenant — to the candidates you rejected. The government's explainer repeats it: criteria must be non-discriminatory, and the landlord explains to rejected candidates why another tenant was chosen. The rules took effect July 1, 2023, and they have teeth: a municipality can pull a rental permit, order the violation undone, or levy a fine up to EUR 90,000 for a repeated violation.
Behind that sits a court decision. On February 5, 2020, the Hague District Court ruled that the legislation authorizing SyRI — a government system that scored citizens for likely benefits fraud — breached Article 8 of the European Convention on Human Rights, because the application of SyRI is insufficiently transparent and verifiable. Note what the court did not say: it did not find that SyRI discriminated. It found a risk that it might, then held that "it cannot be assessed whether this risk is sufficiently neutralised" — the inability to check was itself the violation. It had only ever been aimed at what Human Rights Watch called so-called "problem" neighborhoods, a potential proxy for socio-economic and immigration status. And the leading analysis is careful about how far it travels: the ruling is limited to the circumstances of the case.
Europe has also decided that a private scoring bureau cannot hide behind its clients. In December 2023 the Court of Justice held that a credit bureau's score is an "automated individual decision" prohibited in principle under the GDPR, in so far as the bureau's clients attribute to it a determining role. The bureau answers for the number, not just the bank. And while the EU AI Act's own reasoning acknowledges that credit-scoring systems determine access to financial resources or essential services such as housing, its operative high-risk list stops short of naming tenant screening. Even the good example has a gap in exactly our shape.
Now the part I refuse to leave out, because a contrast that flatters one side is marketing. Dutch algorithm transparency is real and mostly unbuilt. The government's Algorithm Register launched in December 2022, and its own about page admits that supplying information to it is nu nog niet verplicht — not yet required. The data protection authority publishes its own failure rate: more than half of municipalities have registered nothing, and a fundamental rights assessment has only been done in 5 per cent of cases, with "little insight into algorithm registration" in sectors including the housing market. Parliament voted in April 2022 to make human rights impact assessments mandatory before using algorithms by public institutions — but that was a motion asking the government to act, and the current framing ties the instrument to Article 27 of the AI Act rather than a standalone Dutch duty. Dutch digital rights groups call the successor to the struck-down statute Super SyRI, and they are not being affectionate.
Amsterdam is the humbling one. The city consulted experts, ran bias tests, built safeguards, asked affected people — every recommendation in the ethical-AI playbook — and bias reemerged in the live pilot anyway. The alderman stopped the pilot in late November 2023; in the fall of 2024 the city shelved the project altogether.
So the Netherlands is not a paradise. It is something more useful: a place where this failure produces a court judgment, a published compliance rate, a regulator asking for more power, and a project that gets canceled. Here it produces a penalty paid to the Treasury and a press release.
Run the tape forward five years
Here is what worries me, and it is not a robot uprising.
The order tells one company it may not report a case more than once. It binds RentGrow. Housing advocates put the tenant screening market above $1.3 billion, with roughly 2,000 companies offering screening software. The other 1,999 were not in that courtroom.
So picture 2031. Deduplication is solved — genuinely solved, because it turned out to be a data-engineering problem and data engineering gets better. The verdict comes back in ninety seconds: an agent assembles the file, applies the criteria, drafts the adverse action notice and closes the ticket before the applicant has left the parking lot. Landlords trust it more than their own judgment, and the fair housing guidance that used to complicate the criminal-records question was removed from active use six years earlier.
Every eviction filing is still a row in a database — over a million in 2024 in the jurisdictions Eviction Lab tracks — and cases the tenant won still generate a filing. With the duplicates gone, the file finally looks authoritative. That is the trap: accuracy is not fairness, and a clean number is more persuasive than a messy one. We will have built a machine that is right about a record that was never a fair description of a person, and we will believe it precisely because we fixed the arithmetic.
Who is saying what, and from where
The renter's-eye view arrived years before the FTC did. ProPublica reported in 2022 on an applicant screened by RentGrow who could not find out what happened: "You don't know why you got denied or if you were ever considered." That is the whole experience in one sentence.
Upturn's filing to the FTC makes the point no accuracy law fixes: the FCRA gives a reporting agency thirty days to investigate a dispute, which is not quick enough for consumers who are likely to be denied housing inside that window. The National Low Income Housing Coalition's summary of the same research found 46% of respondents saying private landlords rarely or never reviewed underlying information behind a report — sometimes the score is all that is conveyed. And Wonyoung So has shown with simulated reports that landlords conflate tenant record with outcomes — a filing reads as an executed eviction — exactly the confusion duplicate rows amplify.
The other side deserves its own words. Writing for the R Street Institute, Steven Greenhut has defended credit-based insurance scores as effective predictors of risk under automobile policies, arguing that banning such tools punishes people who behaved responsibly. I accept the core of that: a system that cannot tell anyone apart has costs too, and they land on renters as higher deposits and slower answers. The disagreement worth having is not scoring or no scoring. It is who has to explain the output.
Where does anyone actually get paid? Not from the regulator. In a private class action against a different screening company, a federal judge granted final approval of a $2.275 million settlement in November 2024 — almost the identical sum to the penalty here, except that money went to renters. And in November 2025 a DC Superior Court judge let a consumer-protection suit against RentGrow and Yardi proceed, holding the claims are not preempted by federal credit-reporting law. That is where the next round of this fight is.
What does this mean for you?
In rough order of how soon it matters:
Ask for your file before you need it, not after you are denied. The order now makes the company disclose, on request, the vendors behind your address history and middle names, "including, but not limited to, LexisNexis Accurint." Do it in a quiet month.
When you dispute something, close the loop yourself. A renter can win a dispute and the landlord still be told nothing changed. Send the corrected report to the property manager under your own name.
Count the rows. One eviction case or one conviction: how many lines does it occupy? A filing and a judgment in the same case are one event. Two lines is a question worth asking in writing.
Ask the landlord what they actually looked at. "Did you read the report, or the recommendation?" Four awkward seconds, and roughly half of practitioners say private landlords rarely or never go past the summary.
Find out whether your state seals eviction records. The same reporting cites PolicyLink counting around 17 states and six jurisdictions with sealing policies, and more debating them. Sealing is the only fix that stops the row existing at all.
Watch the disparate impact rulemaking, not the settlement. A $2.25 million penalty is a headline. A rule that repeals HUD's disparate-impact regulations — the machinery behind the theory that reaches neutral-looking criteria with lopsided results — decides the next decade.
What I would tell the version of me who never asked
The lesson I take is not that tenant screening software is evil. It is smaller and more uncomfortable than that.
A company was accused of printing one court case as several, for years, knowingly. The government caught it, wrote a good injunction, collected two and a quarter million dollars for the Treasury, and closed the file. The person whose application came back "does not meet" in 2022 because a single eviction appeared twice will never be told, never be contacted, never get the deposit back. The remedy and the harm are in separate buildings.
That is not a scandal about algorithms. It is a design choice about who our enforcement is for. The Dutch answer — justify the rejection to the person you rejected — is not sophisticated technology policy. It is a manners rule, written down, with a fine attached. We could copy it tomorrow.
Until we do, the only person guaranteed to check whether the file describing you is true is you. I spent a decade assuming someone somewhere was. Nobody was.
If somebody you know is filling in a rental application this month — and somebody always is — send them this before they tick the box. That is the whole ask. The HAIA Foundation works on the documents that decide things about people who were never shown them, and the rest of that work lands here.