Last week was Tier 2, the Guardrails Zone. Tap an answer on each one. You will see the reasoning either way, so guessing costs you nothing and teaches you something.
1Tier 2 (Yellow) data can go into AI. On what kind of platform?
enterprise AIENTERPRISE AIAn AI platform your company has licensed under a business contract, where your data is protected by that contract and is not used to train the public model. Multifamily example: a company-paid Microsoft Copilot or ChatGPT Enterprise seat approved by your AI Council, as opposed to a free account someone signed up for on their own phone.. Yellow is not a stop sign. It is a which-door sign. The data is real enough that it needs a contract standing behind it.
2Who signs off on a Tier 2 workflow before your team starts using it regularly?
The AI CouncilAI COUNCILThe group at your company that reviews and approves how AI gets used. It sets the rules, approves Tier 2 (Yellow) workflows before anyone runs them, enforces the Tier 3 (Red) line, and reviews any new AI tool before it is turned on. Some companies call it the Governance Team. Multifamily example: before your team starts using a new leasing chatbot, you submit it to the AI Council and they decide whether it is allowed and under what guardrails.. Review before use, never after. After is not review. After is cleanup.
3You have a Tier 2 document and no enterprise AI. What is the 30-second move that still lets you use AI today?
redactREDACTTo remove identifying information from a document or prompt before anyone else sees it, while keeping the parts that carry the meaning. Multifamily example: turning "Maria Delgado, Apt 4B, rent going from $1,845 to $1,982" into "6-year resident, roughly a 7 percent increase.". Strip what identifies. Keep what teaches. A fake name is still a real person wearing a disguise, and the apartment home, the dates, and the dollar figures pull the disguise right back off.
Good job. That is Tier 2 locked in. Today we close the last tier and then hand you the thing that makes all three of them fast enough to actually use.
Multifamily NEXT · AI Tuesdays · Week 06
The Foundation, Locked In
Six weeks. Every rule you need before AI stops being a risk you manage and becomes a tool you trust.
Here is the thing. You just spent six weeks on the hardest part, the part nobody looks forward to and everybody needs: how to use AI without putting your property, your residents, or yourself at risk. That groundwork is done.
Today does three things. We close the last tier you have not formally walked. We take one clean lap through the whole foundation. And then you prove to yourself that you own it. Next Tuesday the ground shifts, and we start on what AI actually is.
LAST TUESDAY · WEEK 05
Tier 2, the Guardrails Zone. Yellow data rides on enterprise AIENTERPRISE AIAn AI platform your company has licensed under a business contract, where your data is protected by that contract and is not used to train the public model. Multifamily example: a company-paid Copilot or ChatGPT Enterprise seat approved by your AI Council, not a free personal account. with AI CouncilAI COUNCILThe group at your company that reviews and approves how AI gets used. It approves Tier 2 (Yellow) workflows before anyone runs them, enforces the Tier 3 (Red) line, and reviews any new AI tool before it is turned on. Some companies call it the Governance Team. Multifamily example: before your team adopts a new leasing chatbot, you submit it to the AI Council. sign-off, or you redactREDACTTo remove identifying information from a prompt while keeping the parts that carry meaning. Multifamily example: turning a named resident and a real rent figure into "6-year resident, about a 7 percent increase." it down and it stops being Yellow.
Self-guided read-along. About 15 minutes. No timer. Pause whenever you need.
Today’s Session
About 15 Minutes
A review week, run the way adults actually retain things: close the last gap, walk the whole picture once, then recall it under a little pressure. The recall is the part that makes it stick.
Part 1 ~4 min
The last tier
Tier 1, the Green Zone. The one tier you have not formally walked, closed out with a quick sort.
Part 2 ~5 min
The whole foundation, one lap
Six weeks on a handful of slides. The two systems, the failure modes, the law, and who decides.
Part 3 ~5 min
Prove it: the review quiz
Every week, one question at a time. Answer and you see the correct answer and why, right there.
💡 How to read this
Everything you tap or type saves to your browser as you go. Star a slide and it comes with you when you print. Nothing is sent anywhere unless you press a Send button yourself.
The Real Problem
Nobody Runs a Framework at 4:47 on a Thursday.
Five weeks of governance, and here is what actually happens on a property: the rules live in a binder and the AI lives on a phone. The distance between them is where every incident is born.
It is almost never malice. It is speed. So the fix cannot be more rules. The fix has to be faster than the impulse it is competing with.
⚡ Quick check · before we go deeper
You are ten minutes out from a leadership call. You want AI to summarize your notes from three resident meetings this week so you can walk in ready. Your notes have resident names in them. What is the first thing you do?
Run the check. Five seconds tells you a real person is in it, which means Yellow. Two doors open: run it on enterprise AI, or strip the names to Resident A, B, and C and run the check again. Either way you get your summary and you make your call on time. Deleting the chat is not a control. It is a hope.
★ Make this yours · 30 seconds
Which seat in the property management company are you reading from today?
Slide 17 shows the exact moments the check fires in your seat.
The Last Tier
Tier 1. The Zone You Already Live In.
Tier 1 (Green) is the first tier of the Data ClassificationDATA CLASSIFICATIONOne of the two systems this curriculum keeps separate. It answers a single question: where can this data go? It sorts every piece of information into one of three tiers. Tier 1 (Green): safe for any approved AI, used with human review. Tier 2 (Yellow): enterprise AI only, or strip it down to Tier 1. Tier 3 (Red): never, on any platform. Multifamily example: a resident balance owed is Tier 2 data, which is a fact about the data, not about whether AI is allowed to do the task. system, the data that is safe for any approved AI. The definition is only three things: no real person in it, no real money in it, nothing a law protects. It covers more of your week than you think. It only feels small because nobody ever showed you where its edge was.
Do not read a list this time. Recall it. For each one, tap Green or Not Green before you look, then check yourself. Guessing is the point.
Your published pet policy, rewritten to an 8th-grade reading level.
Green. Category one, already public. No person, no money, no law. Use it, then read the output before it goes anywhere.
A blank renewal template full of [RESIDENT NAME] brackets, with no real values filled in.
Green. Category two, nobody in it. Brackets are not people. The moment you fill one with a real name, run the check again.
“What usually drives a summer spike in work orders at garden-style properties?”
Green. Still category two, nobody in it. General industry knowledge with nothing of yours attached. Review the answer before you rely on it, because Green protects your data, not the truth.
Your own first-draft newsletter blurb, pasted in for a grammar and length pass.
Green. Category three, only about you. Your own writing, cleaned up, before anyone else’s name is on it.
That same blank template, but you have now pasted in the resident’s actual email so AI can match her tone.
Not Green. This is paste creep. The template was Green. Her real email is a real person in the prompt, which makes it Tier 2 (Yellow). Describe her tone in a few words instead and you are back in Green.
Your aging report, pasted in with resident names and balances, to ask which residents to call first.
Not Green, and not even Yellow. A real person and their money in the same chat is a combined record, which is Red on any platform. This is the most common accidental Red in multifamily.
That is the whole Green Zone, and it falls into three buckets: things you have already published, things with no person in them, and things that are only about you. It is bigger than most people think. Everything in it can go into an approved AI today, with no approval and no waiting. What it still needs is your eyes on the outputOUTPUTWhat the AI gives back to you after you ask. The text, answer, summary, or draft it generates. Multifamily example: the three renewal responses AI hands you after you paste in the objection. before it leaves your hands.
Where Green Goes Wrong
Green Is Where People Get Careless.
Red scares people, so they respect it. Yellow slows them down, so they think. Green feels like permission, so they stop paying attention. Two quiet traps live in here.
1
TRAP ONE · PASTE CREEP
Clean generic template. Then one real name, to make it specific. Then the apartment home. Then the rent, so the tone lands right. Nobody decided to break the rule. It broke one paste at a time. The moment a real person walks into the promptPROMPTThe instructions you type telling AI what you want. Your input. Everything on the left side of the conversation. Multifamily example: the paragraph you paste into a chat window asking for a resident event brainstorm., it is not Tier 1 anymore. Run the check again.
2
TRAP TWO · GREEN IS NOT UNREVIEWED
Tier 1 says use freely with human review of output. Free to use, not free to publish. Green is a rule about where your data may GO. It says nothing about whether the answer is TRUE. A hallucinationHALLUCINATIONWhen AI confidently makes up facts that are not true. It is not lying on purpose. It is filling in what sounds right when it does not know. Multifamily example: AI invents an amenity your property does not have, or fabricates a Fair Housing citation that sounds real. does not check what tier you are in first, and neither does sycophancySYCOPHANCYWhen AI tells you what it thinks you want to hear instead of what is true. It agrees too easily, praises your idea, and softens hard news, because it is built to be agreeable. This is the twin of hallucination from Week 2: one makes things up, the other flatters. Multifamily example: you ask AI whether your renewal letter sounds too harsh and it assures you it is perfect, when a blunt colleague would have told you the second paragraph reads as a threat., its Week 2 twin, which does the opposite and simply tells you what you want to hear.
⚡ Quick check · this one separates the room
You are drafting a renewal note from a clean template full of [RESIDENT NAME] brackets. Green. Then, so AI can match her tone, you paste in the actual email she sent you. What tier are you in now?
Yellow. The template was Green. The second you pasted her real email, a real person walked into the prompt and the tier moved with her. That is paste creep, and nobody ever decides to do it. It happens one helpful addition at a time. So re-run the check after every addition, not just at the start. And note what it costs to fix: describe her tone in six words instead of pasting her email, and you are back in Green.
The Turn
That Is All Three Tiers. Here Is Everything Else.
You just closed the Green Zone, which means you have now walked all three data tiers. So the teaching is done. From here, we are not learning. We are locking in. Six weeks, one row each, the single thing to carry out of every one.
Week
Topic
The one thing to remember
Week 1
Classification, the two systems.Data ClassificationDATA CLASSIFICATIONOne of the two systems this curriculum keeps separate. It answers one question: where can this data go? It sorts information into Tier 1 (Green, safe for any approved AI), Tier 2 (Yellow, enterprise AI only or strip it down), and Tier 3 (Red, never). Multifamily example: a resident balance owed is Tier 2 data, a fact about the data, not about whether AI may do the task. asks where the data may go. Use-Case RiskUSE-CASE RISKThe other of the two systems, and the one people confuse with the first. It answers a different question: is this AI job allowed to run at all? Green (approved), Yellow (conditional, needs AI Council sign-off), Red (prohibited). Both systems reuse Green, Yellow, and Red, but Data Classification asks where the data may go and Use-Case Risk asks whether the job may happen. Multifamily example: having AI decide which applications to approve can be a Red use-case even when the data is handled perfectly. asks whether the job may run. Never the same question.
Week 2
Hallucination and sycophancySYCOPHANCYWhen AI tells you what it thinks you want to hear instead of what is true. It agrees too easily and softens hard news, because it is built to be agreeable. The twin of hallucination from Week 2: one makes things up, the other flatters. Multifamily example: you ask AI whether your renewal letter is too harsh and it assures you it is perfect, when a blunt colleague would have flagged the second paragraph.. AI will state false things with confidence, and it will tell you what you want to hear. Both are why every output gets human review.
Week 3
Tier 3, the Never Zone. Some data does not enter AI on any platform, redacted or not. The one line in this whole curriculum with no exception.
Week 4
Fair HousingFAIR HOUSINGThe federal Fair Housing Act, which prohibits discrimination in housing on the basis of race, color, religion, sex, national origin, familial status, and disability. Many states and cities protect more. It applies whether a human or AI wrote the ad, the screening rule, or the response. Multifamily example: an AI-drafted reply that treats families with children differently.. The law does not care whether a human or AI wrote it. Protected classes, screening, and advertising all still apply.
Week 5
Tier 2, the Guardrails Zone. Real but not protected data rides on enterprise AI with AI CouncilAI COUNCILThe group at your company that reviews and approves how AI gets used. It approves Tier 2 (Yellow) workflows before anyone runs them, enforces the Tier 3 (Red) line, and reviews any new AI tool before it is turned on. Some companies call it the Governance Team. Multifamily example: before your team adopts a new leasing chatbot, you submit it to the AI Council. sign-off, or you strip it down and re-run.
Week 6
Tier 1, the Green Zone. Public text, blank templates, general questions, your own writing. Use freely, with review. The zone you live in most of the week.
That is the whole foundation. Everything else in these six weeks is detail hanging off these six lines. The next few slides zoom in on the four that people most often blur together.
Review · The Distinction That Matters Most
Two Systems. Two Questions. Never One.
If you remember one thing from six weeks, make it this. The most expensive mistake in the whole foundation is answering the wrong question, because both systems use the same three colors.
1
DATA CLASSIFICATIONDATA CLASSIFICATIONOne of the two systems this curriculum keeps separate. It answers one question: where can this data go? It sorts information into Tier 1 (Green, safe for any approved AI), Tier 2 (Yellow, enterprise AI only or strip it down), and Tier 3 (Red, never). Multifamily example: a resident balance owed is Tier 2 data, a fact about the data, not about whether AI may do the task.
The question: where can this data go? Tier 1 (Green): any approved AI, with review. Tier 2 (Yellow): enterprise AI, or strip it down. Tier 3 (Red): never, on any platform.
This is about the information in front of you.
2
USE-CASE RISKUSE-CASE RISKThe other of the two systems, and the one people confuse with the first. It answers a different question: is this AI job allowed to run at all? Green (approved), Yellow (conditional, needs AI Council sign-off), Red (prohibited). Both systems reuse Green, Yellow, and Red, but Data Classification asks where the data may go and Use-Case Risk asks whether the job may happen. Multifamily example: having AI decide which applications to approve can be a Red use-case even when the data is handled perfectly.
The question: may this job run at all? Green: approved use. Yellow: conditional, needs AI CouncilAI COUNCILThe group at your company that reviews and approves how AI gets used. It approves Tier 2 (Yellow) workflows before anyone runs them, enforces the Tier 3 (Red) line, and reviews any new AI tool before it is turned on. Some companies call it the Governance Team. Multifamily example: before your team adopts a new leasing chatbot, you submit it to the AI Council. sign-off. Red: prohibited use.
This is about the task, no matter how clean the data is.
⚡ Quick check · which system
“Have AI read every application and auto-decline anyone below a credit cutoff.” The data is handled on approved enterprise AI. What stops this?
Use-Case Risk. This is the trap the two-system split exists to catch. Getting the data tier right does not make the job allowed. An AI that auto-declines applicants is a high-stakes, person-impacting decision, exactly the kind of use that lands in Yellow or Red no matter how the data is stored. Classify the data AND ask whether the job may run. Two questions, every time.
Review · What Goes Wrong
The Two Ways AI Lies to You.
Week 2, and it matters more the more you trust the tool. AI does not have to be hacked to hurt you. It fails in two quiet, everyday ways, and both are handled by the same habit.
1
HALLUCINATIONHALLUCINATIONWhen AI confidently states facts that are not true. It is not lying on purpose, it is filling in what sounds right when it does not know. Multifamily example: AI invents an amenity your property does not have, or cites a Fair Housing rule that does not exist.
It makes things up, with total confidence. An amenity you do not have. A rule that does not exist. A statistic it invented. It is not malfunctioning. It is filling the gap with what sounds right.
2
SYCOPHANCYSYCOPHANCYWhen AI tells you what it thinks you want to hear instead of what is true. It agrees too easily and softens hard news, because it is built to be agreeable. The twin of hallucination from Week 2: one makes things up, the other flatters. Multifamily example: you ask AI whether your renewal letter is too harsh and it assures you it is perfect, when a blunt colleague would have flagged the second paragraph.
It agrees with you. You ask if your letter is too harsh and it reassures you. You float a shaky plan and it praises it. It is built to please, so it tells you what you want to hear instead of what is true.
One habit answers both: human review of every output. This is the second half of the Tier 1 rule, the part people drop the moment they trust the tool. Classification decides where your data may go. Review decides whether the answer that comes back is true. You owe both, in every tier, including Green.
⚡ Quick check
AI writes you a market blurb and cites “a 2024 HUD ruling” that sounds perfect for your point. You cannot find the ruling anywhere. What is the move?
Cut it until you verify. A confident, specific, hard-to-find citation is the exact shape of a hallucination. The specificity is not evidence it is real, it is what makes it dangerous. Never publish a legal citation AI handed you without confirming it in the actual source.
Review · Who Decides, and Where the Rules Come From
The Authority Behind the Rules.
Rules only hold if someone owns them and something backs them. Here is the structure underneath everything you learned, all of it grounded in United States law and our own industry.
Who decides
The AI CouncilAI COUNCILThe group at your company that reviews and approves how AI gets used. It approves Tier 2 (Yellow) workflows before anyone runs them, enforces the Tier 3 (Red) line, and reviews any new AI tool before it is turned on. Some companies call it the Governance Team. Multifamily example: before your team adopts a new leasing chatbot, you submit it to the AI Council.
Approves Tier 2 (Yellow) workflows before they run.
Enforces the Tier 3 (Red) line.
Reviews every new AI tool before anyone turns it on.
Where the rules come from
NIST AI RMFNIST AI RMFThe National Institute of Standards and Technology Artificial Intelligence Risk Management Framework, the voluntary United States federal framework for managing AI risk. Four functions: Govern, Map, Measure, Manage. Web: nist.gov and RETTCRETTCThe Real Estate Technology and Transformation Center. Its AI Governance framework is written specifically for rental housing, eight points covering fairness, privacy, human oversight, and vendor due diligence. Web: rettc.org
NIST: the United States federal framework. Classifying before you act is its Map function.
RETTC: the framework written for rental housing. Its point 2 names the laws behind the rules.
Bookmarked in full on the NIST Standards Library slide.
Who enforces
Fair HousingFAIR HOUSINGThe federal Fair Housing Act, which prohibits discrimination in housing on the basis of race, color, religion, sex, national origin, familial status, and disability. Many states and cities protect more. It applies whether a human or AI wrote the ad, the screening rule, or the response. Multifamily example: an AI-drafted reply that treats families with children differently. and the FTCFTCThe Federal Trade Commission, the United States consumer-protection agency that polices unfair or deceptive business practices, including biased or misleading use of AI and consumer data. Web: ftc.gov
Fair Housing (HUD): screening, advertising, and responses, human or AI written.
The Federal Trade Commission: unfair, deceptive, or biased AI and data use.
Neither cares whether AI made the mistake for you.
This is your answer when an owner, a regulator, or an attorney asks where your AI rules came from. Not “our trainer said so.” A named federal framework, a framework your own industry wrote, and the two agencies that enforce the law.
Review · The Whole Thing in One Move
Classify First. Then Prompt.
Everything from six weeks collapses into one habit, and it is the habit that makes AI safe to use at speed: you decide what you are holding before you decide what to do with it.
THE RISK-FIRST SPINE
Before you type anything into AI, ask what is in front of you, and whether the job is even allowed. Only then do you prompt.
Classification protects the data. Use-case risk protects the decision. Review protects the truth. Get those three reflexes down and you can move fast, because the thinking already happened.
★ Reflect · 60 seconds · saves as you type
Of the six weeks, which one changed how you already work? Name the habit you have actually started.
Not the one that was most interesting. The one that changed a behavior. That is the one that stuck, and naming it makes it stick harder.
✓ Saved to your browser. Print this slide and your answer comes with it.
Questions
The Questions People Ask at the Six-Week Mark
Tap any question to open the answer.
Q: Six weeks of rules. When do we actually get to use AI?
Starting now, and that was always the point. The rules were never meant to slow you down, they were meant to tell you the difference between the ninety percent of your week where AI is safe and the ten percent where it is not. Most of what you do lands in Green. The foundation is what lets you move fast there without guessing.
Q: If I only remember one thing, what should it be?
Classify before you prompt. Ask what data is in front of you, and whether the job is even allowed, before you type. Every other rule in six weeks hangs off that one reflex. If you do only that, you will avoid nearly every serious mistake.
Q: I keep mixing up the tiers and the Green-Yellow-Red use levels.
Everyone does, because they share the colors. Hold onto the two questions instead of the colors. Data Classification asks where can this data go. Use-Case Risk asks may this job run at all. Different questions, different answers, and you owe both before you act.
Q: What happens if I get a classification wrong?
If you are unsure, the answer is the more cautious tier, and you ask your AI Council. A cautious wrong guess costs you one conversation. A careless wrong guess can put a resident’s data somewhere it cannot be pulled back. Only one of those has a do-over, which is why “not sure” always rounds up.
Q: We are moving to a new topic next week. Is the foundation done?
The risk foundation is done, yes. Next week we start on what AI actually is, how it works and why it behaves the way it does. That is not a replacement for the foundation, it is the thing that makes the foundation make sense. You cannot fully govern a tool you do not understand, so now we go understand it.
★ Your turn · before we move on
What from the first six weeks is still fuzzy? This is the moment to ask.
Anonymous. No login. No reply expected.✓ Thank you. Your question is in the queue.
Applied Review · Put It All Together
One Tuesday Morning
Four ordinary moments before lunch. Every rule you built over six weeks, in the order you would actually meet them.
No new material here. This is the foundation running live. Walk each moment, make the call, then check yourself. If these four feel automatic, you are ready for what comes next.
THE MORNING AHEAD
8:40 · a follow-up to write 9:15 · a draft to check 10:30 · an ad to place 11:00 · a shortcut a colleague suggests
Two on this run, two on the next. Tap your call on each before you look.
Applied Review · 8:40 and 9:15
The Follow-Up and the Draft
⚡ 8:40 · the follow-up
A prospect’s guest card is open: name, phone, the two-bedroom she toured, her move date. You want AI to draft a warm follow-up. Where does this data land?
Tier 2 (Yellow). A name and a phone number are a real person, which is one tell. Nothing here is protected by law, so it is not Red. Two doors: enterprise AI, or describe the tour without her identifying details and draft a template you personalize by hand. You do not need her phone number in the prompt to get a warm note.
⚡ 9:15 · the draft
AI’s follow-up draft calls your property “the highest-rated property in the district.” You have no idea if that is true. What do you do?
Cut it. AI reached for a claim that sounds good and invented a fact to back it. Green data going in never guaranteed a true answer coming out. Publishing “highest-rated” with nothing behind it is both a hallucination you failed to catch and, potentially, a deceptive-advertising problem. Review the output, every time.
Applied Review · 10:30 and 11:00
The Ad and the Shortcut
⚡ 10:30 · the ad
AI offers to write a listing aimed at “young professionals with no kids, perfect for our quiet building.” It reads clean. What is the problem?
It steers by protected characteristics. “Young,” “no kids,” and “perfect for quiet” signal preferences around age and familial status, both protected. Fair HousingFAIR HOUSINGThe federal Fair Housing Act, which prohibits discrimination in housing on the basis of race, color, religion, sex, national origin, familial status, and disability. Many states and cities protect more. It applies whether a human or AI wrote the ad. Multifamily example: an AI-drafted listing that signals a preference for or against families. does not care that AI wrote it. Describe the apartment home and the amenities, never the resident you are picturing.
⚡ 11:00 · the shortcut
A colleague says: “Let’s have AI score every application and auto-decline the bottom ones. It is all on our approved enterprise tool.” Your call?
Stop, and take it to the AI CouncilAI COUNCILThe group at your company that reviews and approves how AI gets used. It approves Tier 2 (Yellow) workflows, enforces the Tier 3 (Red) line, and reviews new AI tools before anyone turns them on. Multifamily example: before adopting a new leasing chatbot, you submit it to the AI Council.. This is the two-system trap in the wild. The data being on the right platform answers Data Classification. It says nothing about whether an AI that auto-declines real applicants is an allowed use, and a person-impacting automated decision is exactly the kind that needs sign-off or is prohibited outright. Clean data never makes a risky job allowed.
Four moments, four systems, one morning. Classification, the truth check, the law, and use-case risk, and not one of them slowed you down more than a few seconds. That is what the foundation buys you: speed without the guessing.
Your Homework (the fun kind)
Try This Today
Five moves to turn six weeks of learning into six weeks of habit. Under twenty minutes.
1
Name your one habit out loud. The reflex from these six weeks you have actually started using. Saying it is what moves it from a slide to a practice.
2
Open the last thing you put into AI and classify it after the fact. Which tier, and was the job even a Green use? No judgment, just a rep.
3
Catch one hallucination on purpose. Ask AI something you already know cold and watch for where it bluffs. It builds the instinct to never trust an output unchecked.
4
Teach one coworker one rule. The two systems, or the Never Zone, or strip-and-re-run. Teaching a thing once is how you find out you actually own it.
5
Download the cheat sheet and put it where you paste. On the monitor, not in a folder. A rule you can see is a rule you use.
Six weeks in, you hold the entire risk layer. From here the curriculum stops asking whether you may use AI and starts building what you do with it. Good job. I am proud of you.
Reference · Bookmark This Slide
The Foundation on One Card.
Star this slide and it prints with you. Three reflexes, in order. Run them and you have run six weeks.
BEFORE YOU PROMPT, RUN THESE THREE
1
Classify the data. Where can it go? Tier 1 Green: any approved AI, with review. Tier 2 Yellow: enterprise AI, or strip it down. Tier 3 Red: never.
2
Check the use. May this job run? Green approved, Yellow needs AI Council sign-off, Red prohibited. Clean data never makes a risky job allowed.
3
Review the output. Is it true? Every tier, including Green. Watch for hallucination and sycophancy. Never publish a fact or citation you have not confirmed.
💡 The tiebreaker
Not sure is a yes. Round up to the more cautious tier and ask your AI Council. One conversation is cheaper than one resident’s data in a place you cannot pull it back from.
Reference · Bookmark This Slide
Your Role’s Biggest Foundation Risk.
Same foundation for everyone. But each seat has one trap it walks into more than the others. Find yours.
Leasing professional
Paste creep on guest cards. A prospect’s name and number turn a Green template into Yellow the moment they go in. Personalize by hand, not by paste.
Maintenance
Photos. A picture of a document is still that document. Classify what is in the image, not the fact that it is an image.
Property manager
Meeting notes with names. The fastest way to leak resident detail is a summary request ten minutes before a call. Strip first.
Marketing
Fair Housing in ad copy. AI will happily write wording that steers by a protected class. Describe the apartment home, never the resident you are picturing.
Accounting / AR
Combined recordsCOMBINED RECORDA real person and their real money in the same AI chat. On its own a name is Tier 2 and a balance is Tier 2, but together in one conversation they cross into Tier 3, Red, because the AI holds the whole conversation. Multifamily example: pasting the aging report, names beside balances, to ask which residents to call first.. A name and a balance in the same chat is the most common accidental Red in the building. Never paste the aging report.
Regional
Named properties plus numbers. Aggregate is usually Green. The moment a property is named next to its performance, you are back in Yellow.
Asset management
Deal terms and unpublished financials. Higher up the org chart the money gets bigger, so the classification reflex matters more, not less.
Executive / C-suite
Use-case decisions. You approve the jobs, not just the data. An AI that makes person-impacting decisions is a use-case call, and it is yours and the AI Council’s.
The pattern: on-site, the risk is a resident’s data. Up the org chart, the risk is a decision that touches many residents at once. The foundation covers both. Which is why everyone took the same six weeks.
Reference · Bookmark This Slide
Where This Foundation Comes From.
We did not invent this. This whole foundation is a compression of work that United States regulators and our own industry have already done, translated out of legal language and into something you can run at a keyboard. Here is the lineage, all of it grounded in United States law and rental-housing practice.
NIST AI RMFNIST AI RMFThe National Institute of Standards and Technology Artificial Intelligence Risk Management Framework. The voluntary United States framework for managing AI risk, organized around four functions: Govern, Map, Measure, Manage. Multifamily example: when you identify what data is in front of you before you act on it, you are performing the Map function. Web: nist.gov
National Institute of Standards and Technology · Artificial Intelligence Risk Management Framework · the United States federal framework, voluntary
Four functions: Govern, Map, Measure, Manage. Map is the one that asks what you are actually holding and what context it lives in, before anything happens to it. Classifying before you act is that Map step, run in seconds instead of five meetings. Every time you look at a paste and ask what is in this, you are performing a named function of the United States federal AI framework.
RETTCRETTCThe Real Estate Technology and Transformation Center. An industry body whose AI Governance framework is written specifically for rental housing, with eight points covering fairness, privacy, human oversight, and vendor due diligence. Multifamily example: its second point names the exact laws behind Question 1: the Fair Housing Act, the Fair Credit Reporting Act, and the Americans with Disabilities Act. Web: rettc.org AI Governance Framework
Real Estate Technology and Transformation Center · responsible AI for rental housing · the framework written for our industry
This is the one built for us. Eight points, in plain rental-housing language, and several of them are exactly what this workshop teaches. Its second point, Fairness and Consumer Protection, names the laws behind Question 1 by hand: the Fair Housing Act, the Fair Credit Reporting Act, and the Americans with Disabilities Act, plus state and local law. Its fourth point, Protect Privacy, is data minimization and informed consent, which is Question 2 and the whole reason you strip before you paste. Its fifth point keeps a human in significant decisions, which is why Green is never unreviewed. Its eighth point holds third-party vendors to the same standard, which is new-tool review by your AI Council.
Fair Housing (HUD) · the FTCFTCThe Federal Trade Commission. The United States consumer-protection agency that polices unfair or deceptive business practices, including how companies use AI and handle consumer data. Multifamily example: an AI tool that quietly makes biased or misleading decisions about applicants can draw FTC as well as Fair Housing exposure. Web: ftc.gov · the agencies that actually act
The frameworks tell you what good looks like. These are who acts when it goes wrong. Fair Housing, enforced by the Department of Housing and Urban Development, governs how you screen, advertise, and respond, whether a human or AI wrote it. The Federal Trade Commission polices unfair, deceptive, or biased use of AI and consumer data. Neither one cares whether AI made the mistake for you. That is the entire reason Question 1 exists.
Why this slide runs every week. When someone asks where your AI rules came from, “our trainer said so” will not survive a hard question from an owner, a regulator, or a plaintiff’s attorney. The United States federal framework, the framework our own industry wrote, and the two agencies that enforce the law, all pointing the same direction. That is a strong answer, and now it is yours.
Practice Exercise
Spot the Tier: Run the Check on All Eight.
Eight scenarios. Do not think about them. Run the two questions and answer in five seconds. Speed is the skill being tested, not knowledge. You already have the knowledge.
Every answer unlocks the reasoning, right or wrong. Guess honestly. A wrong one here is worth more than a right one.
1
You paste your property’s published pet policy, the exact text already live on your website, and ask AI to rewrite it at an 8th-grade reading level. No names. No dollar figures. Nothing that is not already public.
2
You want AI to summarize your handwritten notes from three resident meetings this week so you can prep for a leadership call in ten minutes. Your notes have resident names in them.
3
A resident emails a note from her doctor requesting a reasonable accommodation for a service animal. You want AI to help you word the response so you get it right.
4
A leasing professional asks AI: “What are the federal Fair Housing protected classPROTECTED CLASSA personal characteristic that anti-discrimination law specifically shields. Under the federal Fair Housing Act these are race, color, religion, sex, national origin, familial status, and disability, and many states and cities add more. Multifamily example: you may not screen, advertise, or word an AI-drafted response in a way that treats families with children differently from anyone else.es, and what kinds of phrases in a listing can get a property in trouble?” No resident names. No property data. Just the question.
5
An accounting associate pastes the full aging report into AI and asks which residents to call first. The report has resident names, apartment homes, and outstanding balances, all on the same row.
6
Marketing pastes next quarter’s lease rates by floor plan into AI and asks for three positioning angles for the fall campaign. No resident names anywhere in the prompt.
7
A leasing professional strips a real renewal objection down to “6-year resident, roughly a 7 percent increase, said she feels taken for granted,” then asks AI for three ways to respond out loud.
8
An applicant’s background check comes back with a criminal record. The leasing professional wants AI to help draft the denial letter so the wording is careful and legally safe.
Take Action
The Review Quiz
Eight questions, one from each stop across the six weeks. Tap an answer and the correct choice and the reason appear right away. Guess freely: the wrong answers are where the learning is, so there is no reason to play it safe.
Running score: 0 correct / 0 of 8 answered
1
WEEK 1Data Classification and Use-Case Risk are two different systems. Which question does Data Classification answer?
💡 The learning moment
2
WEEK 1On your approved enterprise AI, a workflow auto-declines any applicant below a credit cutoff. The data is handled correctly. What still stops this?
💡 The learning moment
3
WEEK 2An AI draft cites a specific 2024 HUD ruling that would be perfect for your point, but you cannot find it anywhere. Most likely, what is it, and what do you do?
💡 The learning moment
4
WEEK 2You ask AI whether your firm renewal letter sounds too harsh, and it assures you it is perfectly worded. What is the risk?
💡 The learning moment
5
WEEK 3Which statement is true about Tier 3, the Never Zone?
💡 The learning moment
6
WEEK 4AI offers to write a listing aimed at “young professionals, no kids, perfect for our quiet building.” What is the problem?
💡 The learning moment
7
WEEK 5You want AI to draft a payment reminder, and the resident’s name and exact balance both need to go in. What is the cleanest safe path?
💡 The learning moment
8
THE FOUNDATIONYou genuinely cannot tell whether a piece of data is Tier 1 or Tier 2. What is the rule?
💡 The learning moment
0 / 80%
✓ Submitted anonymously. Thank you, this shapes what comes next.
Take Action
Your Foundation Cheat Sheet
One page. Three reflexes. It goes on the monitor, not in a binder, because a binder is where governance goes to die.
The Foundation Cheat Sheet
The three reflexes from Slide 16, formatted to print and post. Download it, put it where you actually work, and take one honest question to the person you report to.
1
Download the cheat sheet and print it. Put it where you paste, not where you file things.
2
Pick the one rule your team breaks most. Be honest about it. Every property has one, and naming it is the first move toward fixing it.
3
Send the cheat sheet to your supervisor with that one rule and one question: how do we make this the default here? The button opens a blank email with the message written. You fill in the address.
A cheat sheet on your monitor changes your habits. A cheat sheet your supervisor knows about changes your team’s habits. That is the difference between a workshop and a policy.
Session Complete
Phase One Is Done. You Built It.
Six weeks of rules, the hard and unglamorous part, and it is behind you. Here is what you carry out of the foundation.
✓ Phase 1 Complete · What You Are Taking With You
✓The two systems. Data Classification asks where the data may go. Use-Case Risk asks whether the job may run. Never one question.
✓All three data tiers. Green use freely with review, Yellow guardrails, Red never. And the bright line that has no exception.
✓The failure modes. Hallucination and sycophancy, both handled by reviewing every output, in every tier.
✓The law and the authority. Fair Housing and the FTC enforce it, your AI Council decides, and NIST and RETTC back it.
✓One reflex. Classify first, then prompt. The habit that makes all of it fast enough to actually use.
Here is the thing. The foundation is the part that protects everyone. It is also the part that is now finished. Next Tuesday we start something new: what AI actually is. How it works, why it guesses, why it flatters, and what it genuinely cannot do no matter how confident it sounds. You cannot fully govern a machine you do not understand. So now we go understand it.
You showed up for six weeks. You built the whole risk foundation. See you Tuesday.
★ Before you close this · 90 seconds
Three things you are taking into the rest of your week.
Learning becomes action only when you name it. Anything you type saves to your browser. Print this slide and your answers come with it.
1
2
3
✓ Saved. Use Print Starred or Email My Notes in the toolbar to take this with you.
★ Bring someone with you · the Tuesday handoff
Pick one coworker. Share one rule. Five minutes over coffee.
The fastest way to lock in six weeks is to teach one piece of it. Copy the block below and send it to one teammate today.
I just finished the six-week AI foundation in AI Tuesdays and it is the clearest picture of using AI safely I have seen. The whole thing comes down to one habit: before you put anything into AI, ask where the data can go and whether the job is even allowed, then check whatever comes back. That is it. Want to grab 5 minutes this week so I can walk you through the one part that matters most? Workshop: https://multifamilynext.com/52-ai-tuesdays/week-06-of-52/ (password: 0652)
