The world of online shopping is never peaceful. As long as there is profit, there will be people ready to take action. The "refund only" policy that everyone is clamoring for has made a comeback again. The difference is that today's Woolies no longer rely on acting skills, but rely on AI to mass-produce "evidence." Ms. Gao, who runs a hairy crab online store in Jiangsu, sent 8 crabs to customers as usual according to the order. Unexpectedly, she received pictures and videos of "6 dead crabs" from the customer the next day, and successfully received a refund of 195 yuan.
After investigation, the customer committed fraud by making false videos and maliciously applying for refunds. The police punished him with administrative detention for 8 days in accordance with the law. This is also the first case in the country that was administratively punished for using AI forged videos to defraud online shopping refunds.
Since this year, many merchants have suffered from fraudsters using AI fake pictures to apply for "refund only". Now, the industry has finally fired its first shot, finally letting merchants vent their long-simmering grievances.
However, even if there is a law to follow, when the cost of counterfeiting approaches zero, there will still be wool gangs wandering on the edge of illegality. At this time, how can merchants protect their rights and interests?
Businesses defeated by AI
With the increasing popularity of AI image editing technology, the "refund only" problem that has already caused merchants a headache has become more and more serious. The "AI fraud" experienced by Ms. Gao is not an isolated case.
As early as when AI technology was not so perfect, many users would create their own photos and falsely claim that the products had quality problems, thereby applying for "refund only." However, because most people lack the ability to professionally retouch images, the forged images are often full of loopholes that merchants can see through at a glance.
But even so, there are still many merchants complaining that the platform still accepts refund applications when faced with photos that can be easily distinguished from fake ones at a glance, and even photos with large model watermarks.
Xiaoqian (pseudonym), who works in customer service, said that some buyers will submit a "refund only" application late at night, accompanied by a photo that can be seen as genuine or fake at a glance. It is not even generated using AI, but is just modified casually using photo editing software. Unexpectedly, the platform quickly displayed "Refund Successful". In less than 3 minutes, the customer had already blocked the store.
If most customers' photo retouching skills were a little clumsy in the past, now with the support of AI, many AI photos can be fake and real. Even the "verification video" requested by customer service to be reissued can be generated by AI without any flaws.
For example, the crab merchant Ms. Gao mentioned above, the photos she received can almost be said to be fake. Later, she carefully identified the photos and videos sent by the customer and found that the number of male crabs and female crabs did not match, and she realized that it was AI fraud.
We put the photos of crabs she received into large models such as Qianwen and Doubao to identify the true and false. The answers given by the large models were either true or false. We put the picture into a software that specializes in detecting AI concentration, and came to the conclusion that "it is difficult to determine whether it is generated by AI."
Some merchants have reported on social platforms that today’s AI image-generating technology is difficult to be 100% recognized, and it is easy to be tricked. Moreover, even if merchants are cautious, it is still difficult to prevent the massive attacks from wool gangs.
On the one hand, today's AI tools have extremely low operating thresholds. In just a few seconds, they can generate pictures of defective products "with stains, damage, and extrusion marks." The user's cost of counterfeiting is almost "zero."
We asked Qianwen and Doubao to help transform a normal apple photo into a "mouldy" one. The whole process only took less than 20 seconds, but the effect was slightly worse. Some netizens said that fake effects such as blueberries and bananas will be more realistic, which is also related to the shooting angle, light, etc.
Moreover, even if netizens do not know how to use AI, there is still a gray industrial chain of AI P diagrams. According to the "IT Times" report, someone claimed to be selling "paid knowledge" courses, but in fact they were selling AI P diagram tutorials. The teacher even said, "The tuition is 788 yuan, which includes teaching and training. It also provides easy-refund store links. You can earn up to 500 yuan per order."
Some cheaper P-picture teaching documents are also quietly circulating in the gray industry chain. One called "AI Pseudo-picture Teaching Document" sells for only 9.9 yuan. Some people even provide a one-stop service of "representing P-pictures and making complaints on behalf of others."
On the other hand, most e-commerce platforms have a relatively relaxed review of "refund only" and basically "refund if there is a picture." Even if the merchant finds that the customer submitted a fake AI picture, they can only appeal afterwards.
But for merchants that process orders in large quantities, this kind of screening will undoubtedly add huge costs. Especially for small transactions, many merchants can only choose to "accept failure."
Consumers are also "badly played"
While merchants are worried about the endless AI fraud refunds, consumers on the other end also find themselves trapped in the dilemma of "seeing is not believing".
Ms. Yang recently purchased a thin spring sweater in a live broadcast room. The fabric was delicate and soft when the model showed it, but after receiving the actual product, she found that the style and color of the dress were very different from the picture. Ms. Yang said, "When the model wore the clothes, they looked high-quality, but when they were received, they looked like street stalls, and the workmanship was average."
Ms. Yang originally thought it was just a difference between "real pictures" and "refined pictures", but later she discovered that the live broadcast room used AI digital humans. She believed that digital humans could use rendering technology to eliminate wrinkles in clothes, enhance the luster, and make cheap fabrics look expensive.
In addition to AI digital humans, product detail pages are also the hardest hit area for abuse of AI images. Some consumers complained on social platforms, "Some merchants sell clothes, and they don't even have real photos. They are all virtual models. I can't see the material at all, and I don't dare to buy them."
If consumers are faced with AI-refined product pictures, they can still choose to buy or not; but in the after-sales process, they may encounter even more absurd "AI traps."
Recently, some consumers said that they found hair when buying chicken feet online, and then applied for a refund from the merchant, but were asked to send a "picture of the God of Wealth." The reason given by the merchant was that it was required by the platform for review, and they also stated that no refund would be possible without sending the picture.
Subsequently, many consumers said they encountered similar "routines." Some netizens speculated that merchants inducing customers to send AI pictures can allow the platform to identify consumers as malicious fraud, thereby exempting them from negative reviews. User @omnivorex posted on Xiaohongshu that he had complained to the relevant merchants, but the platform has not yet responded.
Some netizens joked that consumers use AI fake photos to harvest wool, and merchants also use AI photos to fool consumers. This wave is really a "two-way rush."
For those merchants who originally planned to make a "one-and-done deal" and consumers who just wanted to collect the wool and leave, the two sides used AI to set up situations and counterattack each other, which may have become some kind of tacit "offensive and defensive game."
But for those merchants who operate with integrity and consumers who want to shop well, this kind of abuse of AI to create chaos is constantly eroding the trust relationship between buyers and sellers. In the end, the entire e-commerce ecosystem may pay the price for this.
It’s urgent to take action on the platform
AI technology itself has no clear direction, but the user’s intentions will determine its nature. It is urgent to establish targeted regulations and supervision.
Recently, the Beijing Consumers Association teamed up with eight e-commerce platforms including JD.com, Meituan, Pinduoduo, Vipshop, Douyin, Kuaishou, Xiaohongshu, and WeChat Live to sign the country's first "Commitment Letter to Promote the Application of AI Technical Standards", which covers technical identification, content review, responsibility traceability and other aspects, and delineates "compliance red lines" for AI technology applications.
However, Sheng Yuhua, associate professor at the School of Civil and Commercial Economics at Gansu University of Political Science and Law, pointed out that the biggest difficulty with the application of AI technology in business activities is not that there is no law to follow, but that it is difficult to follow the law.
Although there are relevant laws and regulations to comply with, many businesses still have to take the road of rights protection.
Merchant Awen (pseudonym) said, "We are actually fighting against the algorithm." Awen believes that the problem of some consumers taking advantage of loopholes in rules to maliciously "snatch" has always existed, but the emergence of AI technology has made this kind of behavior invisible, and merchants are "fighting wits and courage" with customers. The essence is that the platform cannot defend the first line of defense.
To this end, merchants have also spontaneously established mutual aid organizations. Some people have shared the latest "anti-fraud experience." For example, identifying fake pictures can start with the details, focusing on whether the edges are blurred and whether the lighting and shadows are reasonable. Some people have also shared AI counterfeiting tools to try to find "loopholes in the algorithm." Some people have also made suggestions to identify whether users have "bad behavior" records through data.
However, this kind of risk identification and management work should be the responsibility of e-commerce platforms. A reporter from the "Beijing News" asked e-commerce platforms such as JD.com, Pinduoduo, Douyin, and Taotian about this:
JD.com stated that it has comprehensively built multi-modal AI fake image recognition capabilities for fresh products, which is in the trial operation stage and is expected to be officially launched by the end of the year;
Douyin also announced that it has established relevant AIGC recognition capabilities and is in testing operation; Taotian said that the platform has a strict risk control system to verify the authenticity of the certificate; Pinduoduo responded that it will continue to strengthen after-sales technical support in accordance with relevant regulations.
It can be seen that although major platforms have been aware of the impact and risks brought by AI technology and have begun to take countermeasures, they are still at the stage of "crossing the river by feeling for the stones" and the relevant mechanisms need to be further improved.
In the past few months, e-commerce platforms have successively stopped their “refund only” policies, but the governance of the e-commerce ecosystem obviously cannot be achieved overnight.
Today, AI technology has become a new variable. While it helps businesses improve operational efficiency and optimize user experience, it is also inevitably used by some people to exploit loopholes in the rules. This is also a necessary stage for a new technology to grow from barbaric growth to standardized maturity.
Next, e-commerce platforms need to upgrade their ability to identify AIGC technology as soon as possible and improve supporting regulatory policies. At the same time, we must actively listen to and extensively collect the real voices of merchants and consumers, and work together from multiple dimensions, including rule design, technical prevention and control, and complaint mechanisms, to jointly build a fair, transparent, and sustainable e-commerce ecosystem.