Latest / AI Ethics with Fexingo: Bias, Safety, and Responsible Artificial Intelligence / When Your AI Landlord Decides Who Gets an Apartment
Transcript
- Lucas: So you submit a rental application — credit check, income verification, references — and then a landlord runs it through an AI screening tool that spits out a score. If the score is too low, you're denied, and you may never know why. Luna: And this is already happening. Not just in some futuristic scenario — today, in major cities, these systems are deciding who gets an apartment. Lucas: Right. A 2025 investigation by the California Civil Rights Department looked at one case in Los Angeles. A tenant with a stable job, good credit, and clean rental history was denied by an AI tool called LeaseScore. The reason? A sealed eviction record from five years prior that the tenant had legally removed from public databases. Luna: But if it was sealed, how did the AI find it? That's the whole point of sealing — it's supposed to be invisible to landlords. Lucas: Exactly. And that's the core problem. These AI screening tools often scrape third-party data brokers that don't always respect legal sealing orders. So the system flagged a record that shouldn't have existed in the tenant's profile. She was denied, and the landlord couldn't explain why — because the AI just gave a score, no reasoning. Luna: That's a transparency nightmare. And it's not just sealed records — what about plain old biased data? Lucas: Right. Let's talk about the data. These models are trained on historical rental outcomes — who paid rent on time, who didn't, who got evicted. But historical data reflects systemic bias. For example, Black and Hispanic renters have been evicted at higher rates due to discriminatory policing and housing policies. So the model learns that certain zip codes or demographic signals correlate with higher risk — and it perpetuates that bias. Luna: So it's not that the AI is intentionally racist — it's that the data it's fed is biased, and it just amplifies those patterns. Lucas: Precisely. And that's the classic 'garbage in, garbage out' problem. A 2024 study by the Urban Institute found that algorithmic rental screening tools disproportionately flagged applicants from majority-Black neighborhoods — even when controlling for income and credit score. So you could have two identical applicants, but the one from a historically redlined area gets a lower score. Luna: That sounds like a fair housing violation waiting to happen. Is anyone regulating this? Lucas: There's movement. In February 2026, California State Assemblymember Tina McKinnor introduced AB 2890, which would require landlords to disclose when they use an AI screening tool, explain the factors that went into the score, and offer a human review process. It also prohibits using arrest records or evictions older than seven years. Luna: So the tenant I mentioned earlier — with the sealed eviction — would have had a chance to appeal. That's huge. Lucas: Right. And the bill also requires that the AI be tested for disparate impact — meaning the developer has to show that the tool doesn't disproportionately harm protected classes. That's a big step, but it's still just a bill. It's not law yet. Luna: And even if it passes, it only applies in California. What about the rest of the country? Lucas: That's the challenge. The federal Fair Housing Act hasn't been updated to explicitly cover algorithmic screening. The Department of Housing and Urban Development, HUD, has issued guidance saying that AI tools can't discriminate, but enforcement is spotty. So companies like LeaseScore and SafeRent operate in a regulatory gray area. Luna: Let's talk about SafeRent — there was a big lawsuit, right? Lucas: Yeah, in 2024, a class-action lawsuit was filed against SafeRent Solutions, alleging that their scoring system discriminated against Black and Hispanic renters in Massachusetts. The plaintiffs argued that the model used housing voucher status as a negative factor — which disproportionately affects people of color. The case settled in early 2025, with SafeRent agreeing to remove voucher status from the model and pay restitution. But it's a reminder that these systems are far from neutral. Luna: And if you're a tenant who gets a bad score, what recourse do you have right now? Lucas: Very little, honestly. Most landlords don't give you the score — they just say you were denied. Some states, like Washington and New York, have laws requiring that applicants be told if an AI was used. But even then, you might not get the actual data behind the decision. And if you want to challenge it, you'd have to hire a lawyer and possibly sue — which most renters can't afford. Luna: So the burden is on the tenant to prove discrimination, but they don't have the data. That's a power imbalance. Lucas: Exactly. That's why transparency requirements like AB 2890 are so important. They shift the burden to the landlord and the developer to show that the system is fair. Luna: Switching gears a bit — we talk a lot on this show about how AI affects people's lives in real ways. And one thing we're proud of is that we keep these conversations ad-free, because we think that kind of independence matters. If you find value in episodes like this, you can support the show at buy me a coffee dot com slash fexingo. It's a simple way to keep us going, and it makes a difference. Lucas: Yeah, completely agree. That support is what lets us dig into these topics without worrying about sponsors. So thank you to anyone who's contributed. Now back to the rental screen — I want to talk about one more angle: the idea of a 'tenant score' that follows you around. Luna: Oh, like a credit score but for renting? That sounds dystopian. Lucas: It's already being built. Companies like Experian and TransUnion already offer rental payment reporting, but now some startups are creating portable tenant scores that aggregate your rental history across properties. The idea is that you can take your good payment history with you. But the flip side is that a single bad score — maybe from an unfair AI screening — could lock you out of housing in multiple cities. Luna: So one bad algorithm decision could have a cascading effect. You get denied once, the score drops, then you're denied again, and it spirals. Lucas: Exactly. And there's no federal law like the Fair Credit Reporting Act that gives you the right to dispute your tenant score. Some states have rental scoring laws, but they vary wildly. So in practice, your housing access could be determined by an opaque model that you can't see or correct. Luna: What would a fair system look like? If you were designing it from scratch. Lucas: I'd start with transparency. Applicants should get a clear explanation of the factors in their score, and the ability to correct errors. Second, human review — a landlord should have to look at the full context, not just a number. Third, regular auditing for bias, with public results. And fourth, a ban on using certain data like arrest records, sealed evictions, or housing voucher status. Luna: And maybe a cap on how many years of history can be considered? Like a seven-year lookback, similar to credit reports. Lucas: Exactly. AB 2890 has that. And some advocates are pushing for a 'right to explanation' — meaning that if an AI denies you, you get a detailed reason. Not just 'low score' but 'you were flagged for an eviction record from 2019 that was later dismissed.' That kind of specificity is crucial. Luna: It's wild that we're in 2026 and this still isn't standard. AI is making life or death decisions about where people can live, and there's barely any oversight. Lucas: Well, there's momentum. The California bill is one of several — similar bills have been introduced in Illinois, New York, and Washington state. And the FTC has signaled that it's watching this space. In 2024, they reached a settlement with one of the big tenant screening firms, RealPage, over allegations that their revenue management software facilitated price-fixing. That wasn't about tenancy scoring, but it showed that regulators are paying attention. Luna: So there's hope. But for now, if you're a renter, what can you do? Lucas: First, ask your landlord if they use an AI screening tool. They might not tell you, but it's worth asking. Second, check your credit report and any rental history reports from companies like Experian RentBureau or TransUnion. You can dispute errors. Third, if you're denied, ask for the specific reasons in writing. In some states, they have to provide it. And fourth, if you suspect discrimination, file a complaint with HUD or your state's civil rights agency. Luna: That's good practical advice. I think the key takeaway here is: just because an AI says you're a bad risk doesn't mean it's right. And the system needs more checks. Lucas: Absolutely. These tools can be useful if they're transparent, fair, and auditable. But right now, too many operate as black boxes. And when the stakes are a roof over your head, that's not acceptable. Luna: Well said. Until next time, keep asking questions.