A tenant’s photo of a leak: what retailers learned about AI damage photos, and why you have it easier

A tenant opens a maintenance request through the portal. Water under the kitchen sink, a stained ceiling, a patch of mould behind the bed. There is a photo. On the strength of that photo a plumber gets dispatched, a restoration company gets called, or the tenant asks for a rent credit while the unit is “not liveable”.

Most of these photos are exactly what they look like. This piece is about the few that might not be, and about why the answer is cheaper for a property manager than for almost anyone else being sent AI-made damage photos right now.

This is not legal advice.

Retail got there first

Online retailers are the first business to deal with this at volume, because a refund claim is a photo and a sentence. Modern Retail reported on 2 March 2026 that brands are “battling a surge of AI-driven return fraud”. Its clearest example came from Scott Tannen, chief executive of the bedding company Boll & Branch: a customer sent photos of sheets that had supposedly arrived torn. The rip did not look like anything cotton does when it frays, and one of the images carried an AI watermark. “They were just so blatantly AI that I couldn’t believe it,” he said.

What the company did next is the useful part. It asked the customer to show the damage live on FaceTime. The customer never responded. Another brand in the same report, Bogg, started requiring the product back before refunding suspicious defects.

The insurance side has reached the same place on paper. A January 2026 note from the law firm Debevoise & Plimpton on AI images in insurance claims lists the controls insurers are considering: several photos from different angles with a reference object for scale, detection software, the original files rather than copies, live video inspections where an adjuster tells the claimant where to point the camera, and more site visits.

Notice what the strong controls have in common. None of them is “look harder at the photo”. Each one goes around the photo, to the thing itself.

You already have the thing itself

A retailer has to ask for the sheets back. An insurer has to send an adjuster. You manage the building. In most leases and most states you can enter a unit to make repairs after giving the notice your lease and your state require, and for a real leak you would send someone anyway.

So for a maintenance request, the photo’s job is small: it helps you decide how fast to go and whom to send. It should never be the thing a payment rests on. Three places where it quietly becomes that:

• A rent credit or concession granted on the photo alone. “The bathroom has been unusable for two weeks, here is the picture.” If nobody from your side saw it, the picture is the whole case.
• A vendor bill backed by the tenant’s photo. The restoration company invoices for remediation, and the only image of the damage in the file came from the tenant, not from the vendor who was there.
• Automated triage. Maintenance software increasingly asks tenants for photos at intake and sorts requests by urgency. That is sensible. It also means a picture can move a request up the queue, or trigger an emergency call-out, before a person has looked at anything.

None of this assumes the tenant is lying. A genuine tenant is served just as well by a manager who goes and looks, because the repair gets done.

What the file can tell you, and what it cannot

We looked for a published case of a tenant using an AI-made photo in a maintenance request and did not find one. We are not claiming it is common. What we can say is what the photo file itself records, and that part applies to real and fake alike.

A photo taken on a phone usually carries the phone’s make and model, the exposure, and the time the phone’s clock recorded when the picture was taken. Our checker reads all of that in your browser and uploads nothing. If the image was made by an AI tool that labels its output, such as ChatGPT with its Content Credentials, and the file still carries that label, the checker names the tool.

Two cautions, both as important as the reading itself:

• Message apps rebuild the image, and so may your maintenance portal (we have not tested which portals do). Rebuilding can throw the camera data away. A photo with nothing in it is not a fake. It tells you nothing either way. If the file matters, ask for the original, the way this piece describes.
• A photo of a screen, or of a printed picture, is a real phone photo with a real camera and a real time, whatever it shows.

The capture time is often more useful than the question of AI. It tells you when the water was first photographed, which matters when you are working out how long a leak ran, or whether the stain in the move-out photos was already there at move-in.

Keep the tenant’s photo, exactly as it came

The photo in a maintenance request is evidence for both sides. If a tenant later says they reported a leak in March and nobody came, their March photo and your record of receiving it are what settle it. If a tenant later says the damage was there before they moved in, a dated photo of it appearing is part of the answer.

So keep the file as it arrived, not a screenshot of the portal. Our paid report lists each file’s SHA-256, a fingerprint that changes if a single byte of the file changes, and files that list on a date at an address anyone can open. Months later, anyone can confirm that the photo in front of them is the one you received.

A policy you can adopt this week

1. Photos help you triage, never approve. No rent credit, concession or remediation bill is approved on a tenant photo alone. Someone from your side, or a vendor you sent, sees the damage first.
2. Vendors photograph what they find. Ask for the vendor’s own photos on arrival and on completion, as original files. Those, not the tenant’s, are what the bill should rest on.
3. When you cannot go at once, ask for a live look. A two-minute video call where you ask the tenant to point the camera at the ceiling, then the floor, then out of the window. It is the same step Boll & Branch took, and it is quicker than an argument.
4. Keep the original file of every photo that ends up mattering, and write down when you received it.

The limits, stated plainly

• We found no published case of an AI-made photo in a rental maintenance request. The retail and insurance sources above describe other industries.
• The Modern Retail example is one company’s account, reported by the press. The Debevoise note lists possible controls; it does not report incidents.
• Your right to enter a unit, how much notice you owe, and whether a tenant can withhold rent or repair and deduct all depend on your state and your lease. Check both before you write a policy.
• Our checker reads evidence that is present in a file. It does not judge whether a picture of a leak looks real, and the report we file gives no confidence score. A photo with nothing in it is not cleared, and it is not condemned.

---

ImposterShield reads the evidence inside image files. It runs in your browser and uploads nothing. If a file has been stripped, it says so. That is the product working, not failing.