Refund Fraud Has a New Co-Pilot

It was expected that professional fraudsters would jump on GenAI and experiment with the ways it can help them commit more fraud, more successfully, and more quickly. What is more surprising is how fast regular shoppers have started using the technology to commit refund fraud. 

Researching and interviewing for the book The Fraud Fighter’s AI Playbook (early release here) with my co-authors Gilit Saporta and Chen Zamir has highlighted how widespread serious concern about refund fraud has become. In the 2025 holidays, fraud prevention firm Forter found that AI-altered damage images increased refund abuse use cases by over 15%. 

So what’s behind this shift, and what can fraud fighters do about it?

Fertile Ground for Refund Fraud 

Refund fraud has been growing for years, and notably since the Covid-19 pandemic. The 2024 report from Ravelin, The Rise of Friendly Fraud, found that 17% of those surveyed were willing to admit that they were cheating on refunds. There are probably more who do it, but don’t admit it – especially since the same report found that 72% said they felt guilty about committing fraud.

So why do it, if they don’t feel good about it? Two likely reasons, in combination:

• When money feels tight, many people are willing to push boundaries they otherwise might not.
• It’s easy. 

For many folks, economic uncertainty has been a fact of life since the pandemic, more than was previously the case. At the same time, many retailers made a point of easing their returns policies and processes during the pandemic to help minimize both contact and stress, leaving customers with high expectations about how easy refunds should be. 

Professional refund fraudsters grew up to provide refund-as-a-service for people looking to cheat, making it even easier for consumers to get something for nothing, or at least almost nothing once they’d paid the 5%-25% of the item’s value to the fraudster.

GenAI is the Ideal Amplifier

That’s the landscape into which GenAI entered. There was already a sizable group of people willing to attempt refund fraud. Policies made it relatively simple to get away with, so the retailer side was fairly streamlined from the perspective of the would-be cheater. GenAI eases the other piece, making it simple to create the materials for successful refund fraud without advanced technological ability. 

GenAI’s image generation is so good now that anyone can take a picture of a newly arrived item and instantly create a version that looks broken, dirty or otherwise damaged. The image is brand new, so an image search won’t find it. It perfectly fits the real item, because that’s what it’s based on. All someone has to do is type a request into their preferred GenAI platform. 

The increase in refund fraud, powered by GenAI, is attacking every physical goods vertical. Torn clothes, dirty sneakers, rusted jewelry, used beauty items, broken electronics, mouldy food, you name it, a retailer has seen it faked. 

Identifying Faked Refund Images

Identifying faked refund images is like other kinds of fraud prevention efforts; it’s a game of statistics. You won’t catch every single one. But there are procedures you can incorporate into your protections to help flag which ones are almost certainly GenAI, likely to be GenAI and possibly GenAI. 

Just like other elements of fraud prevention, this is best approached through the lens of risk assessment levels and dynamic friction. 

Here are some techniques to consider:

• It just looks wrong. The surface might be too smooth, the lighting might be off somehow, the damage might not be the kind that you see in real life, or maybe the logo or writing isn’t quite right somehow. GenAI is very good, but sometimes all you need is a human to look at something to know it’s a fake.
• Make sure you’re looking at metadata. Missing or inconsistent data like the camera model, timestamp, GPS etc, is common with GenAI images. 
• File creation date. Does it match the story of the rest of the purchase?
• Hash value. Does the “digital fingerprint” of this image match others? Because it shouldn’t.
• Synthetic image classifiers – more than one, for a good layered effect – can be helpful in making your system’s judgements more accurate to flag the ones that need further review.
• Smart linking for refunds, the way you do for transactions. Is the rush of refunds connected somehow? Make sure your system is looking to find out. 
• Take the account, into account. Has this person got a history of dodgy refunds? Is it a new account with way too many refunds? This is relevant information for risk scoring.
• If this is a high value claim, consider live proof procedures rather than static images.

Ultimately, as with all kinds of fraud, it’s about the whole picture of the individual, the account, their behavior, and this specific purchase. 

New Challenge Can Create Impetus for a New Approach

Fraud fighting teams at many merchants have been tearing their hair out for years, trying to get growth-focused teams and companies to listen to the seriousness and scale of the problem. GenAI makes the issue worse, but it also presents an opportunity to engage other departments in a productive discussion. 

Working to identify the fake images is important, but the wider problem is stopping the fraud at scale. For that, policy changes may well be necessary. It doesn’t matter how good your detection is, if the company insists on refunding anyway. 

If this is something your team has struggled with, a few steps to take this forward come to mind:

• Map the scale of the problem in your own organization, and make sure you’ve got images of the fakes to illustrate it
• Analyze the financial loss this is causing your company
• Determine how many users (both numerically and as a percentage) are involved, to what extent, and how their abuse compares to their spend
• Reach out to key stakeholders who may have been skeptical of changing refund policies in the past, framing it as an issue created by GenAI, and arrange an educational session – with your illustrative images
• Come prepared with suggestions about how to personalize refund flows and policies to individuals, given that you have historical data about who cheats, how much, and what they’re worth to your business
• Aim for concrete agreements about directions for change, and follow up to cement this as necessary

GenAI is letting refund fraud run rampant. That doesn’t mean you can’t stop it.