White paper · Guided tour

Why every brokerage should have a verified AI governance program.

Your agents already use AI. You already answer for how they use it. Here is the case for proving it, in eleven short scenes.

About 8 minutesUse the arrow keys, the buttons, or swipeFrom NIRAP, October 2026

01 · The current state

AI is already inside your brokerage, whether or not you approved it.

Delta Media surveys brokerage leaders every year. The firms in its 2026 survey handled more than two-thirds of U.S. transactions in 2025, so these numbers describe the mainstream, not the early adopters.

97%of brokerage leaders say their agents use AI

3.9%of brokerages report no AI use at all, down from 24.8% in 2024

6.38average worry score, out of 10, that AI lacks guardrails to limit risk or liability

75%of leaders at firms with 20 or fewer agents name regulatory compliance as their top AI concern

Worry fell in 2025 and then rose again in 2026, to within 0.12 points of where it started. Three years of experience has not made leaders more comfortable with the risk. It has made the risk more concrete.

Delta Media Real Estate AI & Leadership Survey, 2026, and Delta Media three-year analyses, August 2026.

02 · Why careful agents get caught

The risk is not careless agents. It is built into the tools.

Each of these flaws produces work that looks finished and confident. That is exactly why capable people miss it, and why no software update will remove it.

Hallucination

The tool invents facts and states them with confidence.

In a listingWrong square footage, or a feature the home does not have.

Sycophancy

The tool tends to agree with whoever is asking.

In a pricing talkAsk it to support your price and it usually will. That is your own view repeated back.

Bias

The tool repeats patterns in the data it learned from.

In ad copyLanguage about who a home is "perfect for," or descriptions that drift into steering.

The defense is the same for all three, and it is not a better tool. It is people who know the traps and a rule that someone checks the work, with a record that they did.

National Association of REALTORS®, "Why Every Brokerage Needs an AI Use Policy," March 24, 2026 (hallucination and bias examples).

03 · The exposure

You do not need a new AI law to be liable. The existing ones already reach you.

Broker supervision

State license law makes the supervising broker answerable for sponsored agents. Texas rule 22 TAC §535.2 is one example.

California's Department of Real Estate said in March 2026 that using AI does not change a licensee's duties and does not shield anyone from liability.

So when an agent publishes AI-written content, you are supervising AI work whether you know it or not.

Fair housing

The Fair Housing Act has no exception for advertising written by software. Intent is not the test.

An AI-written listing (illustrative)"Charming 3-bed with a newly renovated chef's kitchen, on a quiet street that's perfect for young families."

Invented feature: the kitchen was never renovated.Signals a preference based on familial status.

22 Tex. Admin. Code §535.2; California DRE Licensee Advisory, March 17, 2026. The listing is an illustration, not a real case.

04 · What one violation costs

The penalty is already set in federal law.

Maximum federal civil penalty for each Fair Housing Act violation in an administrative case. HUD made no inflation adjustment for 2026, so these amounts still apply.

First violationNo prior violations

$26,262

One priorOne earlier violation

$65,653

Two or more priorsRepeat respondent

$131,308

+ actual damages+ attorney fees+ injunctive reliefPrivate lawsuits: no statutory cap

24 CFR §180.671; HUD notice 2026-19600 (Sept. 24, 2026); 42 U.S.C. §3613.

05 · The shifting ground

The safety nets are getting thinner, so you set your own standard.

Federal guidance was withdrawn

Guidance explaining how fair housing law applies to digital housing ads was withdrawn in 2025 and 2026. The statute itself did not change.

So what

The law still applies, but the outside reference point is gone. Your own written standard, and proof you follow it, carry more weight.

Insurers are writing AI out

In January 2026 ISO introduced optional generative AI exclusions for commercial general liability policies. Some carriers have gone further in other lines.

So what

Insurers are separating AI risk so they can price it. Our projection, not a fact: documented controls will help when that reaches your policy.

NAR showed what to do

NAR's 2026 AI policy template for brokers lists what a policy should cover. It is sound, and every broker should read it.

So what

Nobody checks whether it was adopted. Two brokerages can both "use the template," and from outside you cannot tell the working one from the unread file.

HUD FHEO withdrawal notice (Federal Register, April 6, 2026); Fenwick, "The End of Silent AI" (2026); NAR AI Policy Template for Brokers.

06 · What a program contains

A handful of documents, and the records that show they are used.

What NAR says a policy should cover

The document that puts it in writing

The record that proves it is followed

Approved and prohibited tools

Approved AI tool register

Dated approvals, with who approved each tool

Scope, data protection, use limits

Firm AI policy

Signed acknowledgments from every agent and staff member

Human and fair housing review

Work review standard

A review log for listings, marketing and client messages

Incident reporting

Retention schedule and incident procedure

An incident log with the correction and the originals kept

Training and supervision

Rollout plan

Training completion records and check-ins

Most brokerages picture the middle column when they think of an AI policy. The gold column is what makes it defensible, and its absence is invisible unless someone looks.

The left column follows NAR's guidance. The mapping to documents and records is NIRAP's.

07 · Why verification

A policy shows what you intended. Verification shows what you did.

After an incident, these are the questions a regulator or opposing counsel would reasonably ask. Switch between the two answers.

Question 1Did you have a rule for AI-written content?

Answer

A policy exists. Nobody can say who received it, or when.

The policy, plus signed acknowledgments that a reviewer has already checked.

Question 2Did a person review it before it was published?

Answer

The policy says review is required. There is no record that it happened.

A review standard and a review log, and a reviewer has confirmed the log is kept.

Question 3What did you do when you found out?

Answer

The response is rebuilt from memory after the complaint arrives.

An incident procedure that existed beforehand, a dated log, and kept originals.

These three questions are NIRAP's characterization of foreseeable questions, not a quotation from any regulator.

08 · What you gain

Evidence you build before you need it.

A documented track recordBuilt before an incident, not assembled under pressure after one.

Gaps found by a friendly reviewerWhile fixing them is cheap, instead of by a regulator.

A signal outsiders can checkClients and recruits cannot read your policy. They can check a credential someone else issued.

A clearer answer for insurersDocumented controls instead of good intentions. (A projection.)

One standard across officesEvery team works from the same rules, which matters most for multi-state firms.

Four tests for any verification program

Not every badge carries the same weight. A program that fails any of these is closer to self-certification.

Does a person review the evidence, or does software score your own answers?
Is the reviewer independent of the AI tools being used?
Is the standard published, so an outsider can see what was checked?
Does the credential expire and require renewal?

09 · The honest part

What verification does not do, and the best arguments against it.

The limits

Point in timeIt confirms practices were in place at the review, not every action afterward.

Not a guaranteeNo private credential can promise compliance with every law in every state.

Not a safe harborNo statute gives a verified brokerage reduced liability. It produces evidence, nothing more.

Not insurance or adviceStill review coverage with your carrier and obligations with counsel.

Three objections, answered

NAR's template is free.Use it. It says what the rules should be. Verification answers whether they are in use, which is the question asked after an incident.

We're too small for this.Small firms have the least margin for one penalty, and their program is small too: a handful of documents, not a binder.

We'll wait until it's required.The liability already exists. Waiting only postpones the record you will want when someone asks.

10 · Getting started

Six steps any brokerage can take, with or without NIRAP.

Find out what is in use. Ask agents and staff which AI tools they use, and for what.
Name an owner. One person approves tools and keeps the program current.
Approve tools and write the policy. Include what client data may never be entered.
Require review, and log it. A person checks AI-drafted content before it goes out.
Train everyone who uses AI. Assistants and marketing staff too. Collect acknowledgments.
Have it checked. By someone who did not write it, on a fixed cycle.

Want the whole argument?

The full paper has every source, all five objections and the fine print. NIRAP's brokerage program covers all six steps: a guided toolkit, staff training, and RAICB™ verification by a human assessor.

Download the full white paper PDF Book a call to get started → See the brokerage program →

This tour summarizes a NIRAP white paper and is not legal advice. NIRAP offers the verification program it recommends. RAICB™ certification is a point-in-time attestation, not a guarantee of compliance.

How this tour was made. A person at NIRAP initiated this tour, and AI tools helped produce it. After it was first published, a person at NIRAP reviewed it, including its facts against their sources. It was reviewed and approved on October 28, 2026. The label does not guarantee the tour is free of mistakes or errors. It means a person applied due diligence to verify the output. Errors: contact@nirap.org. How NIRAP uses AI