Is algorithmic rent pricing legal?
It isn't whether software may suggest a rent. It's whose data goes into the suggestion.
In short
5 things that decide this
- 01Algorithmic rent pricing is legal in most of the United States, and banned in a growing list of places.
- 02Pooling non-public rent data from rival landlords to suggest prices is what the Department of Justice sued RealPage over in August 2024.
- 03California banned algorithmic rent-setting in September 2025, New York in October 2025. So did San Francisco, Seattle, Philadelphia, Minneapolis, Portland, San Diego and Jersey City.
- 04Under the RealPage settlement filed on 25 November 2025, non-public data may train a pricing model only if it is at least 12 months old.
- 05A pricing tool holds up when it can show that every input came from that customer's own data or from public sources.
This case was about inputs, not about software
DOJ filed against RealPage in August 2024, joined by eight state attorneys general. The allegation wasn't that a computer set rents. It was that RealPage's revenue-management software took non-public data from competing landlords, then fed pricing recommendations back to those competitors.
Read that again, because it's the whole distinction. If a tool prices your buildings from your own occupancy and your own rents, that's arithmetic. If it prices your buildings using your competitor's effective rents, that's something a phone call between the two of you would also do.
No financial penalty and no admission of liability came out of the settlement. It changed the product instead.
What the settlement requires of the software
These terms come from the proposed settlement filed on 25 November 2025. They read as a product specification, which is why they matter to anyone building in this space.
| Term | What it requires |
|---|---|
| Runtime inputs | Non-public, competitively sensitive information may not be used in runtime operations of the algorithm. |
| Training data age | Non-public information may train a model only if it is at least 12 months old. |
| Excluded data | Price data and geographically localised data are largely barred from model data, whatever their age. |
| Guardrail symmetry | The Governor feature had to be made symmetrical for increases and decreases. |
| Auto-accept | Automatic acceptance cannot be on by default. The user has to set the range manually. |
| Oversight | A court-approved monitor gets broad access to the code, model training documentation and runtime logic. |
| Clock | 30 days for a compliance policy, 60 for advisors to stop sharing unaffiliated data, 180 to stop runtime use and retrain. The decree runs seven years, with early termination possible after four. |
Build the feature so it can answer one question
Whose data is in this number? If your system can't answer that for a single recommendation, the feature is a liability whatever the marketing says. Keep lineage in a column, not in a compliance document.
Aggregation isn't a safe harbour on its own. Because the decree restricts aggregated sharing too, anonymising the pool won't rescue your design. That surprises teams, because anonymising rescues almost everything else.
- 01Partition inputs by customer, and prove it from data lineage rather than a policy page.
- 02Age any pooled training data past 12 months, and drop price and neighbourhood-level features.
- 03Make guardrails work both ways. A one-way ratchet that only resists decreases is out.
- 04Default to recommend-and-confirm. A human accepts the number.
- Own dataThis customer's rents only.
- Public dataPublished sources, freely available.
- Aged poolNon-public, over 12 months.
- ModelNo price, no local geography.
- SuggestionA person accepts it.
Everything to the left of the model has to be defensible on its own. That's the design change the decree forces, and it's easier to build in than to retrofit.
Related questions
01Does this apply to us if we only price our own buildings?+
Pricing your own buildings from your own data isn't what the case was about. Your exposure starts when non-public data from a competitor reaches your recommendation, directly or through a shared vendor. Ask your vendor where their inputs come from, and get the answer in writing.
02Is RealPage software banned now?+
No. Its settlement carried no financial penalty and no admission of liability, and RealPage continues to operate under a seven-year decree with a monitor. State and city bans are separate, and they restrict algorithmic rent-setting by anyone in those places, not only RealPage.
03Where is algorithmic rent-setting actually banned?+
California and New York passed state bans in September and October 2025. San Francisco moved first among cities in August 2024, followed by Philadelphia, Minneapolis, Portland, San Diego, Seattle and Jersey City. RealPage sued New York over its ban in November 2025 on First Amendment grounds, so the map will keep changing.
04Does a landlord using the software carry risk too?+
Yes. DOJ also sued six large landlords, Greystar among them. Buying a tool didn't insulate the buyer. If you're evaluating a revenue-management product, put the data-sourcing question in your security review alongside the usual ones.
Related
- AI, automation and custom software for property managers →The hub this answer sits under.
- Custom software for property managers →Where defensible data lineage gets built into a pricing feature.
- AI phone agents for property managers →Automated decisions touch a resident's call here too.
- Fair housing rules for AI tenant screening →The other regulated model in a property stack.
- AVM →Automated valuation models and their own fair-lending exposure.
