Is it safe to give an AI app access to tenant data?
It can be, if the app reads only what it needs, cannot approve money or legal steps on its own, records what it does, and can be disconnected at any time. Judge it the way the FTC tells businesses to judge any personal information: know what you hold, keep only what you need, protect it, dispose of it properly and plan for an incident.
Start with what you actually hold
Tenant data is not one thing. Some of it is harmless if seen; some can be used to steal an identity. Sort it before you connect anything.
| Data | Sensitivity | Should an AI app read it? |
|---|---|---|
| Unit, rent, lease dates, balance | Low to medium | Yes, this is what most questions need |
| Names, phone numbers, emails | Medium | Only where a task needs it |
| Payment history | Medium | Yes, for rent questions |
| Bank account numbers | High | No |
| Social Security numbers, dates of birth | High | No |
| Screening and credit reports | High, with legal rules | No |
Screening reports carry their own duty. The FTC says landlords who use consumer reports must securely dispose of the report and any information gathered from it when they are done with it. Keep those reports out of any chat.
The FTC's five principles, applied to AI
The FTC's guide Protecting Personal Information rests on five principles. They fit an AI connection well.
| Principle | What the FTC says | What it means for an AI app |
|---|---|---|
| Take stock | Know what personal information you have | List which data the app can reach |
| Scale down | Keep only what you need | Give the app the least data that answers your questions |
| Lock it | Protect the information you keep | Strong sign-in, limited access, a log of every action |
| Pitch it | Properly dispose of what you no longer need | Know how to disconnect and what happens to access afterward |
| Plan ahead | Create a plan to respond to security incidents | Know who you would call and how you would cut off access |
The same guide tells businesses to put security expectations in writing in contracts with service providers, to verify rather than take their word for it, and to require providers to tell you about security incidents. An AI vendor is a service provider.
Pasting into a chat versus connecting the app
Pasting a tenant list into a consumer chat sends that data to the chat provider under its terms. Anthropic, for example, says it uses consumer Claude chats for training only if you turn on the Model Improvement setting or a chat is flagged for safety review. Check your own provider's settings.
A connector works differently: the AI app signs in to your property software and asks for specific data when you ask a question. The software decides what the app may read and do. That is safer only if the software limits the app well, which is what the questions below test.
Questions to ask any vendor
- Exactly which data can the AI app read? Can I see the list of actions it can take?
- Can the AI app move money, sign anything or send a legal notice by itself?
- Is every action the AI takes logged, with who did it, and can the log be edited?
- How long does a connection last before it must be renewed, and how do I disconnect it?
- How are access keys stored?
- Who at the vendor can see my tenants' data?
- Will you notify me of a security incident, and is that in writing?
Example a made-up owner compares two tools. Tool A lets the AI read rent rolls and send payment reminders by itself, with no log. Tool B lets the AI read rent rolls and reject items, logs every action and expires connections automatically. Both answer the same questions about rent; Tool B gives the AI fewer ways to cause harm and gives the owner a record if something goes wrong.
How DoorKeeping handles it
You sign in to DoorKeeping with a code sent to your email or with a passkey; no password is stored. The database lets only members of your organization read your portfolio. Claude and ChatGPT connect through a secure sign-in, DoorKeeping keeps only a scrambled copy of each access key, a connection lasts at most 90 days, you choose which portfolio the app can read, and you can disconnect at any time. Every approval and every action your AI app takes is written to a log that can be added to but never changed or deleted. Payments, signatures, legal notices and applicant decisions wait in your approvals queue: your AI app can reject an item but cannot accept one. More on the security page.
Sources
- FTC, Protecting Personal Information: A Guide for Business, read Oct 29, 2026
- FTC, Using Consumer Reports: What Landlords Need to Know, read Oct 29, 2026
- Anthropic Privacy Center, Is my data used for model training?, read Oct 29, 2026
Questions
- Should I paste tenant Social Security numbers into ChatGPT or Claude?
- No. Keep Social Security numbers, bank account numbers and screening reports out of any chat. Most rental questions need only units, rents and dates.
- What makes an AI connection to property software safer than a chat?
- The software can limit what the AI reads and does, log every action and expire the connection. Ask the vendor to show you those limits.
- Does the FTC have rules on landlords and tenant data?
- The FTC gives data security guidance for all businesses, and says landlords who use consumer reports must securely dispose of them when done. Check your state's privacy laws as well.