电商平台长期依赖买家提交的照片来核验退款理由，但生成式 AI 正在瓦解这套机制。WIRED 在小红书上发现至少十几条卖家与客服的吐槽，称遭遇疑似 AI 生成的“损坏证据”：例如运单中文像乱码、陶瓷杯裂纹却呈现纸张撕裂般的层状效果。被滥用较多的品类集中在生鲜、低价美妆与易碎品等，因为卖家常在不要求退回实物的情况下直接退款。
11 月，一位在抖音卖活蟹的商家收到买家图片与视频，显示“多数螃蟹已死、两只逃走”，甚至有手指戳弄“死蟹”的画面。商家称家里养蟹 30 多年，注意到异常：死蟹腿不应朝上；两段视频里公母数量从 2 公 4 母变成 3 公 3 母；还有一只竟有 9 条腿而非 8 条。她报警后，警方认定视频系伪造并拘留买家 8 天；此案因被视为已知首例引发监管回应的 AI 退款骗局而引发关注。
这一问题并非中国独有。纽约反欺诈公司 Forter 估计，用于退款申诉的 AI 篡改图像自年初以来增长超过 15%，且仍在全球上升；该趋势始于 2024 年中，并在过去一年随图像生成工具更易获取、上手更简单而加速。其 CEO 指出，团伙也在规模化作案：曾用 AI 改图提交超过 100 万美元的退款申请，集中在极短时间窗口以“淹没”审核系统，并通过轮换 IP 隐匿身份。卖家开始用 AI 反制识别改图，但效果不稳；平台若收紧退货政策会伤及守规用户，而水印等既有护栏也往往易被移除，平台需要更强的验证与问责机制。
Ecommerce platforms have long relied on customer-submitted photos to validate refunds, but generative AI is starting to break that system. WIRED found at least a dozen RedNote posts from sellers and support staff describing allegedly AI-generated damage claims—for example, shipping labels with gibberish Chinese characters and “cracked” ceramic mugs that look like paper tears. Abuse appears concentrated in categories like fresh groceries, low-cost beauty, and fragile goods, where sellers often issue refunds without requiring returns.
In November, a Douyin merchant selling live crabs received photos and videos suggesting most crabs arrived dead and two escaped, including clips of “dead” crabs being poked. The seller, whose family has farmed crabs for over 30 years, spotted inconsistencies: dead-crab legs pointing up, sex counts shifting from 2 males/4 females to 3/3 across clips, and one crab showing nine legs instead of eight. Police later deemed the videos fabricated and detained the buyer for eight days; the case drew attention as the first known AI refund scam to prompt a regulatory response.
The pattern is global. Fraud firm Forter estimates AI-doctored images in refund claims are up more than 15 percent since the start of the year, a trend that began in mid-2024 and accelerated as image tools became widely accessible. Forter says organized groups can scale it: over $1 million in refund requests filed in a tight time window, paired with rotating IP addresses to hide identities. Some sellers now use AI chatbots to flag doctored media, but results are imperfect; tightening return policies could punish honest shoppers, and watermarks are often easy to remove—forcing platforms toward stronger verification and accountability.