AI-Enabled Chargeback FraudHow Generative AI Could be Used to Weaponize Chargebacks & What the Industry Must Do About It

Fake, AI-Generated Evidence & Chargebacks

I was reading a report published by UK‑based insurance provider Admiral the other day, which warned of a precipitous increase in cases of fraudulent, AI-assisted insurance claims.

Picture this: you’re an adjuster, and you’re assigned to review a new claim made by a customer. It looks like an entirely routine claim. The buyer submitted a precise description of damage to their Land Rover, including photos of the vehicle with clear signs of impact. You have no obvious reason to second‑guess the evidence. When the fraud team takes a closer look, however, a different story emerges.

The same image of the same vehicle appeared in another claim. The license plate was different, and the damage was altered, but it was clearly the same photo. So, could it be that a customer was reusing an image from a genuine accident to try and claim multiple payouts? Not quite.

Both photos, as it turned out, were AI‑edited renditions of a single original shot, changed just enough to support two separate claims. It was a perfect example of the kind of manipulated evidence that Admiral blamed for a 71% rise in auto insurance claims fraud in 2025.

Insurers aren’t the only ones that could see growing numbers of bogus claims backed by AI‑generated images. The same playbook is also now showing up in payments fraud, too. In fact, one recent report on AI‑enabled financial scams found cases jumped 456% over a twelve‑month span. Deception is cheap, easy, and disturbingly convincing; it’s now a DIY skill that nearly anyone can master.

That got me thinking: could fake documents and doctored photos be used as evidence to help cardholders engage in chargeback fraud? 

The New Playbook: From Perfect Product to Perfect Lie in Seconds

Here’s how an AI-enabled chargeback scam could work, based on patterns we and other fraud teams are seeing:

• A customer orders something online. The item arrives on time and in perfect condition.
• The buyer snaps a photo of the item, uploads it to a generative AI tool, and types a simple prompt: “make this look damaged,” or “add stains to this shirt,” or “add a small burn mark to this garment.” In seconds, they get an image plausible enough to pass as real.
• The cardholder files a refund request and attaches the AI-generated photo as “proof.” If the merchant pushes back, they escalate to a chargeback and submit the same manufactured evidence to their bank.

This is not some futuristic “what if” scenario. In February 2026, fraud‑prevention firm Ravelin documented multiple real cases in which AI‑manipulated photos were used to inflate refund and dispute claims. Their research found that one in three refund abusers agree that AI and technology make it easier to get refunds for online purchases.

The barrier to entry has collapsed. It used to take Photoshop skills and a practiced eye to fake an image. Now it’s point, click, type a few words, and — boom! — you have imagery that could fool many veteran fraud investigators. Organized fraud groups are even offering “fraud‑as‑a‑service” tools on the dark web, with custom AI models trained specifically to generate plausible refund evidence.

Current Tools Are Already Failing Merchants

If cardholders are countering merchants’ evidence with equally convincing, AI‑generated “proof,” then that raises the question: why is this trick working so well?

Part of the problem is speed and workload. Bank employees have just a few minutes in which to judge each claim. In addition, spotting AI‑generated content by eye is exceptionally difficult. Even detection tools struggle; multiple evaluations show that state‑of‑the‑art deepfake detectors can drop 45–50% in accuracy when moved from controlled lab data to real‑world fake data. Meanwhile, real-world tests show that AI‑generated content can bypass many detection tools in as much as 90% of cases.

The traditional representment playbook assumed both sides were using real evidence. That assumption has now gone out the window. The gap between what fraudsters can create and what reviewers can catch is growing; research firm Gartner reports that 30% of enterprises are expected to move away from relying on standalone verification tools by the end of this year. This is now an arms race where AI generation outpaces AI detection. For merchants, the odds aren’t just worse; the system itself has tilted the other way.

From Photos to Video & Synthetic Identities: The Threat Continues to Grow

If photos were the first wave, video is the next looming threat. Package delivery clips “proving” non-delivery or wrong items can now be fabricated. Unboxing videos can be edited (or fully generated) to show damage that never occurred. Even “proof of condition” videos, once a fallback for merchants, have lost their reliability. Any of it can now be fully synthesized with little to no tech savvy. 

Synthetic‑identity fraud is feeding this ecosystem, too. Using AI, fraudsters can create highly believable fake personas, complete with social media profiles, purchase histories, and review activity. A “customer” can have a years‑long digital footprint, even when they don’t exist. Even more unsettling, some of these synthetic identities are built from real people’s data (taken without consent), raising privacy and security concerns alongside the financial ones.

“I think AI-generated fake or modified evidence is happening much more frequently than is reported publicly.”

- Judge Erica Yew, member of California’s Santa Clara County Superior Court

This is nothing less than an evidence crisis. Tools like photos, documents, emails, etc., were meant to resolve disputes. Now they’ve become weapons of deception, and the whole system is starting to crack. And, if cardholders learn that fake evidence wins, dispute abuse ramps up.

On the other hand, if merchants think they can’t win even with real proof, they stop fighting. How can you make a fair judgement if you can’t trust evidence from either side? And remember what I said earlier: this isn't a potential issue. It’s already happening.

What Can Actually Be Done

If this is going to be solved, it’s going to take coordination across the whole ecosystem. No single player can fix it alone.

The biggest gap right now is the lack of consistent standards. Tools like Google’s SynthID watermarking are a step forward, but not widely used yet. There really are no shared standards for verifying where digital content comes from. We need clearer rules for how AI-generated evidence should be treated in disputes. 

Issuing banks need a system for triaging AI‑generated evidence, instead of just forwarding whatever a cardholder uploads. Card networks have a role here, too, as dispute rules must be updated to account for synthetic evidence as the norm, not the exception.

There’s also a data problem. Eighty‑five percent of retailers say they’re already using AI against return fraud, but those efforts are fragmented. The needle will only move when merchants are able and willing to share intelligence: patterns, repeat offenders, emerging tactics, and so on.

At Chargebacks911, we’re working to address this problem by building AI-detection signals into representment workflows and cross-checking evidence for consistency across metadata, timestamps, and sources. Still, I’ll be the first to admit that one provider’s solutions aren't enough.

For merchants, there are still practical steps: multi-angle photos, tighter evidence collection, and monitoring repeat dispute behavior. But these, too, are only temporary defenses against a system that is evolving quickly – and not in merchants’ favor.

The Window Is Closing

AI-generated content is improving faster than detection. Every month without real coordination makes fraud more familiar, more sophisticated, and more costly. 

At the same time, consumer awareness of these tools is growing fast. Forgery is advertised openly, discussed on forums, and increasingly treated more as a “life hack” than a crime.

Again, the response has to be shared. Each player in the payment ecosystem has a part to play here:

• Card networks need to update dispute rules and strengthen authentication requirements.
• Payment processors need to start building AI detection into core payment acceptance systems.
• Merchants need layered verification, plus better intelligence sharing while remaining compliant with data security requirements.
• Technology providers need to align on detection standards that actually scale and adapt to new developments as they arise. 
• Regulators need to establish clearer rules to define and prosecute AI-driven evidence fraud.

Without alignment, dispute resolution risks sliding further into a “he said/she said” situation. And, if the problem persists, it’s likely that some merchants are going to start fighting fire with fire, using AI to generate fraudulent evidence to counteract the fraudulent evidence presented by cardholders. It’ll be a AI-enabled fraud arms race; the question will no longer be who’s right, but how anything could be verified at all.

The bigger issue is structural. The chargeback system was built on the idea that evidence reflects real events. But, we simply can’t afford to assume that anymore. The system will have to adapt, and no doubt it will; the question is whether it can act quickly enough to not kill customers’ faith in the system. I believe it can, but only if we act now, together, and treat this threat with the urgency it deserves before the threat spirals out of control.