Your damage photos will be checked too: what one host’s $16,000 claim teaches a landlord

Every earlier article on this site is about photos a tenant sends you. This one turns the camera around. When you file a damage claim, whether with an insurer, against a deposit or in a deduction letter, someone now reads your photos looking for editing. Here is a case where a host’s photos did not survive that reading, and how to make sure yours do.

This is not legal or insurance advice.

The $16,000 claim that came apart over a coffee table

In the summer of 2025, a guest left a long New York Airbnb stay early. The host told Airbnb she had caused more than $16,000 of damage: a cracked coffee table, a stained mattress, a broken robot vacuum, and more. Airbnb first told her to pay more than $7,000.

She looked closely at the host’s photos. Two pictures of the same coffee table showed the crack, but not the same crack. Real photographs of one object, she said, cannot disagree like that.

After a reporter from The Guardian asked about the case, Airbnb reversed its decision. It apologised to her, refunded her stay, warned the host, and said it was “reviewing the original handling of this case” (PetaPixel and Fox Business, both citing The Guardian).

Two things about this case matter to a property manager.

• Nobody publicly established that AI made those photos. The guest suspected editing. What undid the claim was not a detector. It was two photos of the same thing that did not agree.
• The claim was checked because the other side pushed back. Most of your claims will never meet a tenant who studies your photos that closely. Some will. And the people who pay out on claims have started looking whether anyone pushes or not.

Claims reviewers now expect editing

In March 2026, the insurance data firm Verisk published two surveys, one of 300 insurance claims professionals and one of 1,000 U.S. consumers.

• 98 percent of the insurers agree AI editing tools are driving a rise in digital media fraud.
• 76 percent say AI-altered submissions have become more sophisticated.
• Only 32 percent say they are very confident they could identify a deepfake.
• On the other side, 36 percent of consumers said they would consider digitally altering a claim image or document. For Generation Z it was 55 percent.

In May 2026 the analytics company SAS put the problem in one sentence. “With just a few prompts,” said its insurance adviser Franklin Manchester, “fraudsters can use generative AI tools to create, enhance or erase visual evidence to support a false insurance claim.” Its own demonstration was a chair with a coffee stain that AI had painted on.

Those are surveys and a vendor’s warning, not a count of fraud. They do tell you the mood of the person reading your claim: they expect some photos to be edited, and most of them are not confident they can tell which.

What reviewers are being told to ask for

In January 2026 the law firm Debevoise & Plimpton wrote up how insurers can respond to AI-generated claim photos. Its list:

• “several photos of the damage from multiple angles and distances, and the inclusion of a reference object in the photos to help validate scale”
• “native files or the original documents”
• “real-time remote video inspection of the damage”
• more on-site visits
• software for detecting AI-generated images

Read that list as a preview of the questions your claim may get. Every item on it is something an honest manager can supply easily, if they planned for it at the time.

Five habits that make your photos hold up

1. Shoot sets, not single shots. Take a wide shot of the room, a medium shot and a close-up of each item, with something in frame for scale: a tape measure or a set of keys. This is the first thing on Debevoise’s list, and it is the check that caught the Airbnb host. An edit has to agree with every other photo of the same object. Genuine photos do this without anyone trying.
2. Keep every original exactly as it came off the phone. The camera writes its model, its settings and its clock into the file. Messaging apps and screenshots throw that away. Send photos as email attachments, not through a chat app. Getting the original photo covers which routes keep the data.
3. Keep the whole sequence, including the dull shots. Thirty photos taken minutes apart on one phone, with camera times that run in order, tell a reviewer they came from one visit. Three hand-picked photos tell them less. The camera clock article explains what that time can and cannot show.
4. Do not tidy a damage photo. That includes brightening it, cropping it, or using the phone’s AI eraser to remove a reflection or a tenant’s belongings. Many phones now record an AI edit inside the file. A reviewer who sees “edited with AI” on your photo will not know the edit was cosmetic. If you need a marked-up version, make a copy, say it is a copy, and keep the original.
5. Write down what you filed, and when. For each photo, record the file’s SHA-256 fingerprint and the date you sent it. If the photo is ever questioned, you can show it has not changed since. What happens at a hearing shows how to make that list yourself.

What this does not do

A camera original is not a true photo. It shows which phone wrote the file and when that phone’s clock says it did. It does not show the scene is what you say it is.

Camera data can be edited by anyone who knows how. It is strong corroboration, not a signature.

A stripped file is not a lie. If a photo has lost its camera data, that tells you it travelled through something that removes it. It does not tell you anyone cheated.

We do not know what any particular insurer or deposit programme checks. The list above is one law firm’s advice to insurers, not an industry standard.

We will not give you a confidence score. We tried twice to build a classifier that judges a picture by its pixels. The second attempt looked at eighteen ordinary photographs and called seventeen of them fake. It is switched off. A reviewer who can check your file for themselves does not need to trust our guess, or yours.

---

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.