Fake receipts were just the start: AI is now faking damaged parcels for refunds


This is a sponsored article brought to you by Refundid.


Expense teams got the warning first. On AppZen’s platform, AI-generated receipts went from 0% of flagged fraudulent expense claims in March 2025 to 71% by mid-May 2026, a rise set out in TechFinitive’s July feature on real and fake receipts.

SAP Concur’s Chris Juneau told customers “do not trust your eyes”, and the feature argued that the answer had to be structural (matching receipts against card transaction data, for instance) rather than relying on reviewers to look harder at each image.

Retail returns teams are now learning the same lesson. The forgery has moved from the expense receipt to the photo of a damaged item, and the payout is a refund.

Damage that only exists in the photo

In March 2026, the OECD’s AI Incidents Monitor logged a case in which fraudsters used generative AI to create fake images of damaged goods to claim refunds on platforms such as Vinted, Amazon and Fnac.

The entry records financial harm to sellers, and notes that the fake images undermine automated moderation systems and make fraudulent claims harder to detect. French outlets covered the case widely that week.

UK retailers shouldn’t file it as a French story, given that Vinted is big with UK sellers and Amazon was on the list too.

Why photo-only damage claims are the weak point

Plenty of returns processes still run on some version of “send us a photo and we’ll refund you”. Some never ask for the item back at all.

The claim then lands with an automated check or a lightly staffed review queue, and either way the decision rests on the picture. Both made sense at a time when faking a convincing photo of damage took effort and some skill with editing software.

Generative AI has removed both barriers. The economics now look like the fake receipts did: a convincing image costs next to nothing to make, it rides on a claim small enough to wave through, and the same trick works on order after order.

Say a customer reports that a £60 jacket arrived with a torn seam and attaches a sharp, well-lit photo of the tear. Under a photo-only policy, that refund goes out without anyone at the retailer handling the jacket, which is the same blind spot Juneau warned SAP Concur’s customers about.

What to check before a refund goes out

For retailers, the structural fix is to take the evidence off the screen and put it back in the parcel, because a damaged item is far harder to fake than a picture of one. Returns management software such as Refundid builds that step into the process: the retailer’s team scans each parcel when it arrives and checks the items before closing the return, while return rules based on product, order value or country decide which claims qualify in the first place.

Asking for every item back would irritate honest customers and add handling costs to orders that were never at risk. The job is to aim the checks, and a sensible first set of rules looks like this:

  • Ask for the item back on any damage claim above a value threshold you set, and on the categories where your own data shows claims clustering.
  • Scan each returned parcel on arrival and check the contents against the reason the customer gave before anyone approves the refund.
  • Let low-value orders and customers with a clean history keep the photo route, so the extra friction lands where the risk is.
  • Record what each inspection found against the original claim, so you learn which kinds of “arrived damaged” claims hold up.

Detection software for AI-generated images is the obvious shortcut, and it’s the same look-harder approach the receipts feature argued against. Once the item is on a bench, the question stops being whether a photo looks real and becomes whether the damage is there, which anyone holding the item can answer at a glance.

Your returns data sees what a photo can’t

A single photo can’t tell you that it’s the third damage claim from the same account this year, or that claims on one product have suddenly jumped. Your returns data can, provided someone reads it.

Say damage claims on one style of trainer triple in a month while the rest of your footwear holds steady. That could be a faulty batch or a run of fake claims, and you only find out which by getting a few pairs back and looking at them.

The scale justifies the effort. In the US, the National Retail Federation’s 2025 returns research estimates that shoppers will send back 19.3% of online sales in 2025 and finds that 9% of all returns are fraudulent.

The same report found that 45% of shoppers say it’s acceptable to “bend the rules” when returning items, which matters more now that bending them with a fake photo takes no skill at all. They’re US figures, so treat them as a prompt to pull your own numbers rather than a forecast for a UK business.

The receipts lesson, applied to returns

The receipts feature’s answer was to check each claim against the card transaction behind it, which a fraudster can’t generate. For returns, the equivalent is the item itself and the account’s history.

Retail isn’t the only sector facing this. When TechFinitive covered AI-driven fraud in financial institutions in January 2025, identity firm Signicat’s data showed deepfake attacks growing from 0.1% of the fraud attempts it detected three years earlier to around 6.5%.

Of the fraud decision-makers Signicat surveyed, 76% recognised AI as a growing threat, yet only 22% had started putting protective measures in place. Retailers have less excuse for that gap than most.

None of the changes above depends on spotting a fake. If a retailer makes only one of them, it should be to stop refunding damage claims on a photo alone above a value it has decided it can’t afford to lose.

About The Author

image
Gabriel Jones

This author has published on TechFinitive as part of a sponsored article. Sponsored articles are not endorsed by TechFinitive's Editorial team. Gabriel Jones is a versatile content specialist with a passion for writing about technology, education, and digital solutions. With a keen eye for detail and a commitment to delivering engaging, insightful content, Gabriel helps readers navigate complex topics with ease.

Read more from this author.

We take journalism seriously. To learn more on why you should trust us, head to our editorial guidelines page or meet our team.

NEXT UP