Is AI tenant screening fair and legal?

Screening legality does not depend on whether a human or a machine reads the report. Two federal frameworks apply either way. The Fair Housing Act bars decisions based on protected classes such as race, religion, national origin, sex, familial status, and disability. The Fair Credit Reporting Act governs how you pull background and credit data and what you owe an applicant when you deny them.

AI does not exempt you from either law. If a tool helps you reject applicants in a way that disproportionately harms a protected group, that can be a fair housing problem even when no one intended it. The legal standard follows the outcome, not the tool that produced it.

The risk is not the software itself. It is how the software is built and how you use it.

• Disparate impact. A model trained on biased history can quietly penalize protected groups. Screening on factors that stand in for a protected class carries the same danger.
• Black-box scores. If a tool returns an approve or deny score you cannot explain, you cannot defend it or write an honest adverse action notice.
• Over-broad data. Pulling in social media, arrest records, or unrelated signals can violate screening rules and invite disputes. Many places limit what you may consider, and rules vary by state. See the state guides at /laws/ and confirm with your own counsel.

Treat AI as a research assistant, not a judge. A few habits keep you inside the law:

• Write your criteria down first. Set income, credit, and rental history standards before you meet applicants. A common rule of thumb is income of two to three times the rent.
• Apply them to everyone. Same questions, same thresholds, same order for every applicant. Consistency is your best defense.
• Keep a human on the decision. Read the underlying report, not just a summary score, before you approve or deny.
• Follow the adverse action process. When a report drives a denial, the law requires notice plus the applicant's right to dispute the data.
• Keep records. Save your criteria and your reasoning so you can show every applicant was measured the same way.

Before you trust a tool with an applicant decision, put it through these questions:

• Does it pull from a licensed consumer reporting source, or scrape the open web?
• Can it explain, in plain terms, why an applicant scored the way they did?
• Does it generate compliant adverse action notices when you deny someone?
• Does it let you set your own written criteria rather than hiding them inside a model?