When Your Broken Mug Is a Deepfake and Your Crabs Have Nine Legs

Online refunds used to be a little morality play.

You order something. It arrives looking like it got suplexed by a delivery truck. You take a couple photos like a dutiful citizen of the Consumer Republic, fire off an email, and some exhausted customer service worker hits the “refund” button to make you go away. Everybody keeps their dignity, more or less.

Now generative AI is waddling into the scene like a raccoon that learned to pick locks.

The story rolling out of China is the kind of modern farce that would be hilarious if it weren’t so perfectly inevitable: scammers are using AI-generated “damage photos” to get refunds. Not subtle stuff, either. We’re talking shipping labels with Chinese characters that look like a keyboard sneezed. Ceramic mugs “cracked” in ways that resemble paper being peeled into layers, like the mug is secretly made of lasagna. And the crown jewel: a “dead crab” video where the crab anatomy can’t keep its own lies straight—gender ratios changing between clips and one crab rocking nine legs like it’s auditioning for a Lovecraft reboot.

This is where we are: the refund system is being taken down not by criminal masterminds, but by people with a free image generator and the ethics of a wet paper bag.

The Refund Economy Was Always a Handshake Deal

Let’s not pretend ecommerce refunds were ever built on some granite foundation of truth. It’s been a handshake deal, just with more tabs open.

For a lot of products, it costs more to process a return than it does to just eat the loss. Groceries, cheap cosmetics, fragile odds-and-ends—if you’re selling $8 face cream or shipping a ceramic mug halfway across the country, you’re doing a constant math problem: “Is it cheaper to argue with this person, or cheaper to hit ‘refund’ and pray they don’t come back?”

Most merchants choose “refund and pray.”

And customers learned the rules of the game. If you’re honest, the system is a mercy. If you’re not, it’s a vending machine. All you need is a photo that says, “Look what happened,” and a bored human on the other end who has a queue of 700 tickets and a supervisor breathing down their neck about average handle time.

That last part matters. Fraud doesn’t require perfection. It requires fatigue.

Forter, a fraud detection company, says AI-doctored refund images are up more than 15 percent since the start of the year and rising. That number doesn’t surprise me. Frankly, I’m shocked it’s not higher. Give people a tool that prints plausible evidence on demand and tell them the worst consequence is “account warning,” and you’ve basically invented a new slot machine.

Pull lever, receive money.

AI Didn’t Invent Lying. It Just Put Lying on Autopilot.

Before AI, if you wanted to fake damage, you had to do arts-and-crafts fraud. You had to crack the mug yourself. Smear something on the package. Photoshop a little if you were fancy. The barrier wasn’t morality—it was effort.

Generative AI changes the labor economics of dishonesty.

Now you can type: “photo of torn bedsheet in shipping package, label in Chinese, indoor lighting,” and the machine spits out a convincing-enough lie in seconds. You don’t need to own the product. You don’t need to damage it. You barely need to think. The scam becomes less “criminal act” and more “creative writing prompt.”

And because customer service is trained to smooth things over—not conduct forensic investigations—these fakes only have to clear a low bar. The image doesn’t need to fool a judge. It needs to fool Tina in refunds, who is running on instant noodles and company policy.

That’s the ugly secret: most systems aren’t secured by truth. They’re secured by inconvenience.

AI removes the inconvenience.

The Nine-Legged Crab Is the Most Honest Character Here

The crab case is my favorite, because nature itself tries to intervene.

A merchant selling live crabs on Douyin gets sent videos: most crabs “arrived dead,” two “escaped,” and there’s even a human finger poking the corpses like it’s a crime scene. It’s got drama. It’s got tactile realism. It’s got the kind of “proof” that makes a refund agent sigh and reach for the approve button.

But the seller knows crabs. Her family’s farmed them for decades. She notices the dead crab legs pointing up—apparently not a thing. Then the sexes change between clips. Then a crab sprouts a ninth leg like the AI got bored and started improvising.

That ninth leg is basically the universe whispering: “This is fake, you idiots.”

And it worked, sort of. Police got involved. The buyer was detained for eight days. The whole thing became a spectacle because it was one of the first known AI refund scams to trigger a regulatory response.

Which is a polite way of saying: this isn’t just annoying anymore; it’s becoming a civic problem. When fraud gets easy enough, it stops being a fringe behavior and turns into a hobby.

Why China, Why Now? Because Scale Turns Glitches into Disasters.

This isn’t “unique to China,” and the article admits it. But China is a perfect pressure cooker for it: massive ecommerce volume, hyper-competitive marketplaces, thin margins, and a consumer culture trained to use chat-based platforms for everything from dinner reservations to financial services.

When you operate at that scale, anything that works 1 percent of the time works thousands of times a day.

Refund fraud is like termites. One termite is nothing. A million termites is structural engineering.

And the platforms have created the conditions for it by design: frictionless shopping, frictionless refunds, frictionless everything. “Customer obsession” sounds noble until you realize it also means “we’ll pay people to stop complaining.” The scammer is just a customer who studied the incentives better than you did.

The Real Crime Scene Is “Proof”

Here’s the uncomfortable part: the entire refund process rests on a concept that’s now getting melted down in real time—the idea that a photo is proof.

We’ve all been living in this lazy assumption that images are little windows into reality. You snap a picture, it captures what happened, end of story. That was never fully true—photos can be staged, cropped, timed, lit, manipulated—but AI takes the last scraps of confidence and grinds them into dust.

When a mug can look “cracked” in physically impossible ways, the problem isn’t that scammers exist. The problem is that the evidentiary layer of the internet is rotting.

Customer service teams are about to become accidental experts in artifact detection: weird text, impossible shadows, inconsistent reflections, hands that look like melted wax. And even then, they’ll be wrong half the time, because modern AI can get 95 percent right and still win. Fraud doesn’t need excellence. It needs plausible deniability.

Meanwhile legitimate customers are about to get punished for living in the same world as scammers. More hoops. More delays. More “please provide additional documentation” emails that make you feel like you’re applying for a mortgage instead of trying to replace a $12 bedsheet.

The innocent always pay in paperwork.

The Platforms Will Respond the Way They Always Do: With More Surveillance

You can already see the solution shape forming in the corporate mind palace: not “let’s improve trust,” but “let’s collect more data.”

They’ll want videos instead of photos. They’ll want “live” verification, like you holding the mug, rotating it, saying today’s code word like you’re proving you’re not a hostage. They’ll ask for metadata. They’ll ask you to scan a QR code and film the unboxing like you’re producing low-budget content for the Unboxing Cinematic Universe.

Some marketplaces will push “returnless refunds” down and make returns mandatory, even when it’s absurd. Congratulations: now we’re shipping garbage across the country to “prove” it’s garbage, so we can throw it away with extra steps. The planet really needed more cardboard and diesel in the name of “integrity.”

Others will go full algorithm: risk scores on customers, sellers, addresses, devices. If you’ve ever felt the cold hand of automated suspicion, you know how that goes. One weird pattern and suddenly your account is treated like a raccoon in a jewelry store.

And the slickest fix will be “provenance,” meaning cryptographic signing of images at capture time, watermarks, secure camera pipelines—basically a fancy way of saying: “We’re going to rebuild the camera so it can testify in court.”

That might help, but it’s also a reminder that we’ve reached the stage where we need notarized pixels.

The Scammers Aren’t Geniuses. They’re Just Early.

This is the part where everyone wants a villain with a cape. But most of these scams are probably being run by regular people who realized the system is soft in the middle.

Fresh groceries are a prime target because sellers often won’t demand returns. Low-cost beauty products too—nobody wants to re-stock something that’s been opened, or argue about whether a $6 serum “smells weird.” Fragile items are perfect because damage is believable and shipping companies have the reputation of stampeding buffalo.

So the scammer’s calculus is simple:

1. Pick an item that’s cheap to buy but expensive to investigate.
2. Generate convincing damage evidence.
3. Get refund.
4. Keep item.
5. Repeat until banned.
6. Start new account.

It’s not criminal genius. It’s spreadsheet thinking with a dash of shamelessness.

And as the tools get better, even the telltale nonsense—gibberish characters, paper-tear cracks—will fade. Today’s nine-legged crab is tomorrow’s anatomically perfect crustacean with Oscar-worthy lighting.

When the fakes get good enough, the real shift won’t be technological. It’ll be psychological: merchants will stop believing customers by default, and customers will stop expecting to be believed. That’s the kind of societal erosion that doesn’t make headlines, but it changes how everything feels.

How Do You Defend Against a World Where Pictures Lie?

If you’re a seller, you’re going to need layers, and none of them are fun:

• Process design: Make return policies smarter by category. Returnless refunds are a gift to scammers in certain segments. You might keep them for low-risk customers only.
• Friction in the right places: Not every refund needs a forensic lab, but some need more than “send pic.” Randomized checks, staged requirements, or requiring short videos for certain claims can reduce abuse.
• Better tooling: Basic AI detection for obvious artifacts will help with today’s sloppy fakes. It won’t solve tomorrow’s.
• Human escalation: The crab lady won because she knew crabs. Domain experts catch weirdness that generic reviewers miss.
• Customer reputation: Yes, it can be unfair, but reputation systems exist because life is unfair. The trick is transparency, appeals, and not letting a model quietly ruin someone’s ability to buy socks.

If you’re a buyer who isn’t running scams, you should prepare for the refund experience to get worse before it gets better. Keep packaging. Take unboxing videos if you’re buying fragile stuff. Use payment methods with dispute options. And don’t be shocked when the merchant asks for more proof than feels reasonable. They’re not calling you a liar. They’re reacting to a world that’s getting flooded with synthetic “truth.”

And if you’re one of the people generating fake damage photos to steal refunds, congratulations: you’re not Robin Hood. You’re just another opportunist raising costs for everyone, like the guy who steals all the pennies from the take-a-penny tray and then looks offended when the cashier glares at him.

The Absurd Future: Customer Service as Digital Forensics

What I can’t stop thinking about is the slow mutation of everyday work into something paranoid and surreal.

Refund agents will become amateur detectives, squinting at shadows and counting crab legs. Sellers will become skeptical archivists. Customers will start preemptively filming their lives to create “evidence” for the platforms. “Here’s me opening the box, here’s the shipping label, here’s a shot of the mug next to today’s newspaper, here’s my cat as a witness.”

We’re turning commerce into court.

And all because we built a system optimized for speed and scale, then acted shocked when someone used the same speed and scale to lie.

The saddest punchline is that AI is sold as this grand leap forward—efficiency, productivity, the future. And a chunk of its immediate real-world impact is: making it easier to fake a cracked mug and scam a crab farmer.

Progress, baby.

Now if you’ll excuse me, I’m going to pour something brown and stare into it like it can still tell the truth.

Source: Scammers in China Are Using AI-Generated Images to Get Refunds