The Rent Algorithm Case Ended With No Fine and No Admission. Read the Terms Anyway.
No damages, no admission — and one clause about twelve-month-old training data that concedes what the software was doing. Plus the September 1 deadline most renters will miss.
Back in March I wrote a sentence I would like to have back.
A judge, I said, had finally signed off on the RealPage settlement — the big algorithmic rent-pricing case — and that was that: no fine, no admission, nothing much to see. I put it in writing, to people relying on me to have actually read the file.
It was wrong twice.
Wrong on a plain fact, first. A federal judge did enter a final judgment on March 2, 2026 — against Greystar, one of the landlord defendants. Against RealPage itself, the docket still says "proposed": nine months on, the Justice Department's case page lists a proposed final judgment, a competitive impact statement and a stipulation, and no entered judgment for the software company at the center of it. Every confident search result telling you otherwise has swapped one defendant for another, exactly as I did.
Wrong a second time in a way that bothers me more. "Nothing much to see" was a verdict on a document I had skimmed for a penalty and closed when I didn't find one. There is no penalty. There is something stranger: a paragraph about how old the data is allowed to be. Once you understand why that number is twelve months, you are reading a description of what the software was doing — written by the government, and accepted by a company that says it did nothing wrong.
Two piles of paper, and only one of them has money in it
There are two entirely separate proceedings here, and nearly everyone mixes them. I did.
Pile one is private. Renters sued in federal court in Nashville, and the case has been settling in waves: thirty-seven settlements with a combined value of $359,925,000 now partially resolve the litigation, subject to a final approval hearing on October 15, 2026. They came in batches — 26 settlements providing $141,800,000 preliminarily approved on November 21, 2025, then a second round of 14 class settlements worth $218 million filed on May 14, 2026, in which eleven landlords "do not admit fault or liability." (I won't pretend those counts add cleanly to thirty-seven; the sources count them differently and I can't reconcile it.)
That is real money — the only money in this story. And even there nobody concedes anything: the settling defendants deny wrongdoing, and the court has decided nothing.
Pile two is the government's, and it contains no money at all.
On November 24, 2025 the Justice Department filed a proposed consent judgment against RealPage. Watch the verb in the department's own summary: if approved by the court, the proposed consent judgment would require the company to:
stop having its software use competitors' nonpublic, competitively sensitive information to set rental prices in runtime operation;
stop training the models on active lease data, limiting training to backward-looking nonpublic data aged at least twelve months;
stop using models that determine geographic effects narrower than at a state level;
remove or redesign features that limited price decreases or aligned pricing between competing users;
accept a court-appointed monitor.
Assistant Attorney General Abigail Slater framed it simply: "Competing companies must make independent pricing decisions." Notice what is not on the list. No fines or admissions of wrongdoing — and RealPage says its solutions remain available and already incorporate many of the required changes.
Two clocks sit inside the proposed judgment, and coverage keeps welding them together. The monitor serves three years from approval of his workplan, extendable by eighteen months at the government's discretion; the judgment expires seven years after entry and can be switched off after four. Antitrust counsel reading the decree for corporate clients came away with a compliance checklist rather than a scandal.
Nor did the case end in November. In July the department filed a proposed settlement with a fifth defendant, Willow Bridge — companies, Associate Attorney General Stanley Woodward said, "cannot share sensitive data and manipulate AI tools or algorithms to produce market aligned pricing" — and it would also have to cooperate against the remaining defendants.
Why twelve months? Because that is how long a lease lasts
Here is where it gets interesting.
The operative language is Paragraph IV.A.3. RealPage may not use current, forward-looking or historical data from unaffiliated properties to train its revenue-management models — except for historical data at least 12 months old and not from Active Leases. The competitive impact statement explains where twelve came from: per the Bureau of Labor Statistics, twelve months is the most common lease length, with only about 7% of leases running longer. Age the data sixteen months and virtually all active leases drop out — so data aged sixteen months is deemed not to come from an active lease. Sixteen isn't a stricter rule; it is the presumption that makes the twelve-month rule work.
Read that again slowly. The remedy is calibrated to the life cycle of a lease.
The Justice Department said the quiet part out loud in May 2026, answering public comments on its own deal: the data aging requirements effectively eliminate active leases — a unit with a rental agreement in effect — from the training process, so that RealPage's models "cannot be trained using nonpublic data that represent the current competitive conditions in the market."
That is the whole case in one clause. Not "the software was inaccurate." The concern was that models were learning from what your neighbors' landlords were charging right now, in leases currently in force, and pricing your renewal off it. The fix isn't to make the software better. It is to make its information old.
Which tells you what freshness was worth. DOJ's complaint alleges — and an allegation is not a finding; no court has ruled here — that a RealPage executive noted the need to use competitors' information "while it was fresh," and that RealPage and landlords used the phrase "a rising tide rises all ships" for how the software would move prices at the top and bottom of the market.
None of this came from nowhere: the 2022 investigation that started all of it, Heather Vogell's ProPublica piece on YieldStar, reported that RealPage acknowledged feeding clients' internal rent data into its pricing software, giving landlords an aggregated, anonymous look at what nearby competitors were charging. And what the lawyers told their corporate clients afterward was not stop. It was: age the data. A compliance instruction, not a moral one — which is what a consent decree produces.
The case against my own reading
I have just told you a tidy story, so let me try to break it.
Aging data is an old antitrust convention, not a smoking gun. For decades a safe harbor in the government's health-care policy statements turned on the data being more than three months old and aggregated. Twelve months is that convention with a bigger number. Except — and this cuts both ways — the Antitrust Division withdrew those statements in February 2023, arguing that the premise that historical or aggregated data is not competitively sensitive is "undermined by the rise of data aggregation, machine learning, and pricing algorithms that can increase the competitive value of historical data." Aging is therefore less of a confession than I claimed, and less of a fix than the decree implies.
The company says the software made rents lower. RealPage's counsel Stephen Weissman said there has been a great deal of misinformation about how the software works, and that its historical use of aggregated and anonymized nonpublic data has led to lower rents, less vacancies, and more procompetitive effects. Settling a separate matter with Nevada's attorney general, the company denies the allegations and calls settling a way to avoid cost and distraction. I don't have to agree with any of that to concede the point that matters: no court ever tested it.
The outside commenters contradict each other, which is the most honest thing in the file. Eight public comments arrived under the Tunney Act — recorded in the same May filing quoted above — attacking the deal from opposite ends. One argued the plaintiffs assume harm rather than demonstrating it. Another worried that RealPage gets to keep all the old data it gathered unlawfully and build models on top of it. Both cannot be right.
So the honest version of my thesis is narrower than the headline. The twelve-month rule is not a confession squeezed out of a defendant. It is the government's theory of the harm, made operational — and written down where anyone can read it.
Spain answered the same complaint by regulating the price instead of the pipeline
Now change countries, because one jurisdiction looked at unaffordable rent and never argued about training data at all.
Spain's 2023 right-to-housing law lets housing authorities declare zonas de mercado residencial tensionado — stressed residential market areas, where supply risks being insufficient at prices people can afford. Inside one, when the landlord is a gran tenedor (a large holder — generally an owner of more than ten urban properties), the rent may not exceed the maximum set by the state's reference price index. Not a nudge. A ceiling. The ministry must publish the roster quarterly, and does: the resolution listing the zones declared in the first quarter of 2024 sits in the state gazette with a document number.
The index doing the capping is what Americans should look at hardest. SERPAVI is built by the housing ministry out of tax records on actual habitual-residence rentals; it runs a public web application returning a reference range for any home, and the whole underlying database for 2011 through 2024 can simply be downloaded. Hold that beside the American object. One rent index is proprietary, trained on landlords' confidential lease data, and sold back to them. The other is public and sets a maximum.
It is not a small experiment. As of August 3, 2026, 317 localities across five autonomous communities are declared stressed zones, covering 9.3 million people — practically 19% of Spain's population; the government says public data from Catalonia, Euskadi and Navarra demonstrate the measure works, while a Basque regional minister cautioned it would be premature to call one quarter a trend. Catalonia went first, in March 2024, requesting that 140 municipalities be declared — a policy won by a tenants' union, the Sindicat de Llogateres, rather than by an antitrust division — and it has spread since over strong opposition from Partido Popular regional governments.
Does it work? Genuinely contested, and I am not selling you the happy half. For it: average new-contract rents fell 1.28% in the capped municipalities since March 2024 while rising 9.53% in the rest of Catalonia — though the owners' chamber insists the effect is transitory. Against it: supply is set to remain at historic lows, on a February 2026 forecast with Catalan rents still rising around 6.8% this year. A June 2026 working paper by José García-Montalvo finds the policy reduced prices as intended but with exit at the top of the distribution not compensated by entry at the bottom; a February 2026 preprint by Luis Pérez García finds that under alternative identification strategies the effect on rent growth becomes statistically insignificant, while a decline of around 13% in tenancy agreements signed stays robust.
And regulation does not escape the loophole problem either. Landlords in Catalonia began converting long leases into "seasonal" ones to duck the cap, and the Catalan Parliament had to pass a law regulating seasonal rentals to stop the evasion on December 18, 2025. Cap the price, someone reclassifies the contract. Age the data, someone finds another signal.
Spain is now asking the same question, in Spanish: reporting there suggests the big property portals' algorithms could favor corporate networks and large owners in ways that push prices up, and a junior coalition partner has moved to put an AI-housing monitoring body inside the national competition commission. A motion — not a law. The live instrument is still the ceiling.
Now run the clock forward
Suppose the court enters the judgment roughly as filed. The monitor's term ends around year three; the judgment becomes terminable at four and lapses at seven.
Now picture the product in year five. It has never touched an active lease. It trains on twelve-month-old leases, which is permitted; plus public asking rents scraped from listing sites, which nobody restricted; plus its own clients' occupancy and traffic, which is not a competitor's data at all; plus a demand model very good at inferring the present from the recent past. The output lands within a few dollars of where the old system would have landed — not because anyone cheated, but because a year-old lease plus a live listings feed plus a good forecast is a serviceable substitute for what was taken away. Nobody breaks the decree in that scenario. That is what makes it plausible.
Barcelona got seasonal leases. American renters may get something quieter: a scrupulously compliant engine that arrives at the same number anyway — at which point the question stops being what did you train on and becomes could two competitors have reached this number independently?
Who is saying what, and from where
This fight is not left against right. It is concentration against efficiency, and the sides don't line up the way you'd guess.
Maurice Stucke told the Senate Judiciary Committee in October 2023 roughly what the settlement would concede two years later, citing a figure of RealPage's software being responsible for the pricing of 8% of all rental units nationwide. The economics predate the lawsuits: a 2020 American Economic Review paper found that pricing algorithms consistently learn to charge supracompetitive prices, without communicating with one another.
Handle the magnitude numbers carefully. A White House Council of Economic Advisers analysis publicized in December 2024 estimated that pricing algorithms added about $70 per month to rent and cost renters more than $3.8 billion in 2023 — framed as a likely lower bound. Work presented at the FTC in February 2026 by Sophie Calder-Wang and Gi Heung Kim lands at about $53 per month per unit of markup attributable to algorithmic coordination — then declines to oversell it, noting this is not a full counterfactual and that responsive pricing can be welfare-enhancing.
The anti-monopoly case rests on concentration: RealPage's clients allegedly comprise about 90 percent of the U.S. market for investment-grade multifamily units — a share that turns a pricing tool into infrastructure. From the law-and-economics side, Alden Abbott argued in December 2025 that antitrust challenges to algorithmic pricing may discourage the use of algorithms, and thereby reduce market efficiency, calling this decree highly regulatory. Not a claim that RealPage was innocent — a claim that the remedy is regulation wearing an antitrust costume, and it deserves an answer rather than an eye-roll.
What does this mean for you?
In order of how soon it matters:
If you rented in the United States between October 2018 and November 2025, check whether you are in the class. The court-authorized settlement website lists covered properties and the deadlines that actually bind you: opt out or object by September 1, 2026 — just over two weeks from now — a fairness hearing on October 15, and claims through January 29, 2027. Most eligible people will do nothing, which is what the arithmetic of a $359,925,000 fund assumes.
Ask your building one boring question at renewal. Not "why is it going up." Ask whether the property uses pricing-recommendation software, and who supplies it. An agent who has to say it out loud is one who has to think about it.
Know which rule applies where you live. New Jersey's FAIR Act, signed July 20, 2026 and effective July 1, 2027, doesn't age the data — it prohibits algorithmic systems performing a "coordinating function" on rents, lease terms or occupancy outright. A widening list of cities has done the same, Jersey City the latest after San Francisco, Philadelphia, Minneapolis and San Diego.
Don't wait for Congress. The federal bill that would ban using competitors' competitively sensitive information to train a pricing algorithm was introduced in February 2024. It is still a bill.
When you read that a case "ended," check whether the judgment was entered. Proposed and entered sit in different columns of a public docket, and the gap between them is the whole distance from a plan to a rule. I learned that the expensive way, in front of people.
What I would tell my March self
The absence of a fine was never the story.
The story is that the United States government looked at a rent-pricing algorithm, decided the problem was not the math but the freshness of what it ate, and negotiated a diet. Twelve months, because that is how long a lease runs. Sixteen to be sure. A monitor for three years, a judgment for seven, no dollars, no apology. The outlet that broke the story in 2022 recorded the ending in one sentence: the settlement did not include admissions of wrongdoing and does not involve financial penalties.
I still think that clause concedes more than the press release does, and I am less certain than I was in March that it fixes anything — the same Antitrust Division told us in 2023 that machine learning makes old data valuable again. What I am sure of is that the terms are where the policy actually lives, and that "no fine, no admission" is the phrase that stops people from reading them. It stopped me.
So read the terms. They are public, they are searchable, and every one of them was written by someone who assumed you wouldn't.
If a friend of yours pays rent — which is to say, if you have friends — forward this before September 1; the deadline is real and the fund is finite. More close reading of the fine print at the HAIA Foundation, and in your inbox if you subscribe.