Using AI-Generated Images to Get Refunds
Scammers are generating images of broken merchandise in order to apply for refunds.
Scammers are generating images of broken merchandise in order to apply for refunds.
Jon • December 30, 2025 8:18 AM
Strange that you would tag China in this post without any supporting context. I assume this is happening in the western world or is it ooh bad China or nasty China Ai. Nobody would dream of doing such a horrid thing in England or the goodie two shoes USA with western Ai tools, would they?
Your blog is terrific by the way.
egan • December 30, 2025 8:30 AM
Strange that you would tag China in this post without any supporting context.
The title of the linked article is litteraly : “Scammers in China Are Using AI-Generated Images to Get Refunds”
Clive Robonson • December 30, 2025 9:33 AM
@ ALL,
The wired article is reproduced from an earlier article about the Chinese online ecommerce scams.
Toward’s the end it notes that the scams started with retailers and then the purchasers.
Which is actually not that surprising as “honesty in marketing” legislation is generally quite low in most countries as online retailing is only a recent phenomenon from this century. Where as purchasers are legislated against more thoroughly by legislation on “theft and fraud” that often pre-dates not just ecommerce but goes back two or three centuries.
At the bottom of the article we find,
”But really, these trends are two sides of the same problem: Ecommerce relies heavily on trust, and widespread availability of AI is making it increasingly difficult to operate under the assumption that the majority of people are honest actors.”
AI easily breaks “trust and honesty” but all to often does not suffer reputational harm. And even where AI has broken the rules on “honesty and trust” the nature of the ecommerce markets, are that they are so separated out. That reputational harm is unlikely to be known to either party in a transaction.
This was an issue Prof Ross j Anderson had considered but came up with little to resolve it.
In fact the notion of ensuring trust in ecommerce is almost ridiculous because of the distance between retailers and purchases.
The normal suggestion is to insert a third party into the system to act as an observer. However it does not take long to realise that that will not resolve the issue.
Back last century was the Vesa-Mastercard system called “Secure Electronic Transactions”(SET) which was a horrific set of ASN.1 rules, but it failed. Mostly because perchasers did not want the level of security given.
KC • December 30, 2025 11:40 AM
Of all the telltale signs of an AI-generated crab: “One of them also had nine instead of eight legs.” 🙂
In a video a Yunnan lawyer reportedly outlines the penalties.
Fraudulent refunds over $424 are a criminal offense. Amounts over $4,240 could result in 3 to 10 years imprisonment. If under $424, the AI dilettante can be detained for 5 to 10 days and fined $283. Note: I don’t know if these are yuan conversions?
lurker • December 30, 2025 2:19 PM
Shoot the Messenger:
wired(dot)com is broken for me. Oh, I can connect and get the page and wired keeps on downloading garbage at 10Mbits/sec for a further 10 seconds. Using curl to dump the page to a text file I find the page itself comes to 1.3MBytes of scripts and trackers concealing 8kB (including markup) of useful text.
@Bruce’s headline says “Using AI-Generated Images to Get Refunds”
but wired emphasises the evidence showing Chinese engaging in criminal activity. Will this crime cause as much damage to the US economy as misquided tariffs?
@KC: yes, the values have been translated for a US audience, US$424 = CNY3000. And I have my doubts about some similar videos on YT where the “other language” voice-overs are AI. On the one you quoted it is almost impossible to hear what the Chinese lawyer is saying.
So, somebody’s smart idea of saving the vendor and customer the hassle of shipping back physical goods, looks to be not so smart. I had the experience this year of a small electronic device that failed after 2 days use. The vendor requested photos, which I thought was daft. Apart from the pre-AI possibility of Photoshop, how could a photo show a calculating device not calculating?
Carl Fink • December 30, 2025 5:55 PM
“Of all the telltale signs of an AI-generated crab: “One of them also had nine instead of eight legs.” 🙂
True crabs (and many other crustaceans called “crab”) are decapod crustaceans. “Decapod” means “ten legs”.
It’s arachnids that more often have eight legs.
Red • December 30, 2025 8:39 PM
Scamming is everywhere these days.
Saw on Reddit earlier today where this guy got rocks instead of his Geforce RTX 5080.
…evidence I managed to compile like the weight change where the package gained nearly 3lbs in transit with FedEx….
It’s reached the point where I won’t do cash-back when paying with a credit card. Lets merchants unload counterfit bills. Even at banks, I double-check with a UV light & pen. I just know, sooner or later, it’s going to be a problem…
I wonder what next year will bring…
KC • December 30, 2025 10:07 PM
@ Carl Fink
From a little bit of research, I leaning into the crabs from the Wired article being Chinese mitten crabs (Eriocheir sinensis).
Here are two more videos on this story.
Do you believe one could say this crab has eight walking legs plus two claws, totaling ten appendages?
There were six crabs in this particular order. And, honestly, there’s a lot of legs going on. At this point I can’t claim for certain to have spotted the extra leg Wired references.
There were also other issues that made the merchants suspicious, such as the positions of the claws and that two separate videos had a different number of male and female crabs.
It’s reported that this 195 yuan ($27) refund fraud resulted in an eight day administrative detention. I’m still reading, but it appears that in more than a few countries, AI-enabled fraud could move the rating of an offense into a higher tier.
K.S • December 31, 2025 9:25 AM
Chinese low-trust high-enforcement society is better positioned to deal with this problem than high-trust society in the West. For example, I don’t expect a PD in USA to do anything about similar issue (AI fake refund scam) unless it reaches organized crime magnitude. In turn, US-based e-retailers will be forced to deal with the problem, likely by minimizing effort and offloading costs of dealing with it. This means that online shopping is about to get much, much worse for everyone involved.
K.S • December 31, 2025 9:35 AM
@lurker
There is no point complaining, browsing in 2025 requires DNS filter and/or good ad blocker. For quick and easy setup, I recommend Firefox with uBlock Origin. If you have time, then setting up PiHole DNS redirect and a browser with per-domain JS blocking capability (native or via extension) is the way to go, but the latter requires some basic understanding of modern web infrastructure (e.g., you need to make educated guesses what each domain is likely used for and allow only ones necessary for site functioning).
Thankfully, this site still renders as basic HTML with nothing else enabled.
369 • December 31, 2025 6:43 PM
Study Results Identify Perils of AI
https://www.psychologytoday.com/us/blog/the-first-impression/202512/study-results-identify-perils-of-ai
‘>Studies like the ones recently reported in both Nature and Science demonstrate that users who regularly engage with chatbots are vulnerable to being manipulated (Lin et al., 2025; Hackenburg et al., 2025). As Hackenburg et al. (2025) identify, users oftentimes are more vulnerable when engaging with AI than with other technological tools because they are less attuned to the persuasive potential of these interactions. Engagement via chatbots often feels more personal or intimate for users, lowering their defenses.
Chatbots serve as an additional pipeline for mis- and disinformation, both of which are already running rampant in digital spaces. The scope and scale with which users have adopted chatbots suggest significant perils to the embrace of these tools, particularly the way they are currently being rolled out in our culture, with virtually no guardrails in place to protect users.
The best way consumers can combat the penetration of even greater technologies into their daily lives is to be as well-informed as possible on their likely effects. This way, consumers can be more vigilant when deciding how—or whether or not—to engage with them. In the case of chatbots, evidence is building to suggest the costs may be greater than the potential rewards.’
John White • January 5, 2026 3:29 PM
@KC: Yes, 424 USD at this point = 3000 Chinese Yuan/renminbi. It’s a reasonably significant threshold, as the actual purchasing power for a consumer is much higher in China.
Subscribe to comments on this entry
Sidebar photo of Bruce Schneier by Joe MacInnis.
Paul Sagi • December 30, 2025 7:45 AM
There’s an Airbnb scam doing it too.
Fake damage photos after a booking, so the guest gets charged for damages.
Yet another reason to avoid Airbnb.