Pixels Can Lie, Devices Don't: 3 Strategies To Combat The Rise Of AI-Generated Return Fraud

FORBES ·

Dan Pinto is CEO and co-founder of Fingerprint . With over a decade in tech, he is an entrepreneur behind many startups. Return fraud has always been a cost of doing business in e-commerce. A customer claims a package never arrived. Another says the product was defective. A third insists they received the wrong item entirely. For years, platforms handled these disputes by asking for photographic evidence. Generative AI has made it easier than ever to fabricate “pixel-perfect” images of damaged, defective or missing goods. Scammers in China have already operationalized this at scale, submitting AI-altered images to extract refunds on products that arrived without a single defect. Even images that were clearly AI-generated still resulted in successful refund payouts because the content review process is too slow and too broken to catch everything. Catching and stopping this type of fraud is a bit like a game of cat and mouse. E-commerce platforms can build a great image detector, and fraudsters will build a better image generator or overwhelm review systems with hundreds of photos. AI-generated images used in refund claims have increased by more than 15% since the start of 2025, and continue to rise globally. To combat this trend, e-commerce platforms need to change what they verify. Instead of using a “smarter” photo scanner, companies need to focus on the device that’s submitting the image. To understand why content verification is failing, it helps to understand why platforms adopted it in the first place. Companies like Amazon, DoorDash and Lime built image-based verification into their refund and claims workflows as a friction mechanism. Requiring a photo meant effort, with the assumption that only those with a defective product would go through with submitting a claim. What once required fraudsters hours to edit images can now happen almost instantly with a prompt. What platforms actually need to do now is verify the legitimacy of the person submitting it. A customer with a substantial purchase history, completed orders and no prior disputes has a different risk profile than newly created accounts from the same IP address in a short period of time. The practical implementation of context verification starts with device intelligence. Device intelligence evaluates a device's characteristics and behavioral patterns in real time before a refund request is approved or even reviewed. The moment a refund is requested, device intelligence allows platforms to establish a trust context and ask the questions a photo can’t answer: • Has this device been associated with confirmed fraud elsewhere on the platform? • Do the device's characteristics match the signatures of automation tools or bot networks? • Is this device connected to multiple accounts? • Is the device presenting itself as a mobile app user while running on a desktop emulator? Technical indicators associated with tools commonly used to industrialize refund fraud leave their own traces that device intelligence can surface to quickly answer these questions before any human reviewer ever looks at a submitted image. As AI agents’ popularity increases in e-commerce, transactions become harder to verify and content verification alone becomes even less useful as a fraud signal. The solution is straightforward: Use device context to aid in the refund process. Serial return fraudsters operate on a simple assumption: Resetting their digital identity—whether by creating new accounts, cloning mobile applications or activating a VPN—means starting with a clean slate. From the platform's perspective, each attempt to reset a digital identity looks like a brand-new customer, but from a device-intelligence perspective, these attempts can be linked. Traditional fraud prevention measures typically introduce unintended friction for good customers. With device intelligence, merchants can still provide frictionless experiences for devices with a strong, clean history of consistent purchase patterns, few prior disputes and clean device characteristics. For devices that lack history, exhibit tampering or automation signals or have other suspicious attributes, the technology can trigger step-up friction, such as an additional verification step or a hold on their account. This mitigates risk and creates an economic deterrent for bad actors. Fraudsters can operate at scale because the per-attempt cost is low and the yield is high. When an attempted return encounters friction, such as a hold period or a verification call, the time and cost of the attempt rise. When these deterrents are applied, fraudsters who run these schemes at scale cannot afford to spend 20 minutes on a manual verification call for a $20 refund. The rise of AI-generated fraud is a direct threat to the digital trust that makes modern e-commerce possible. Platforms that continue to rely on easily spoofed visual “proof” risk creating a landscape in which every legitimate customer is treated with suspicion, even as they continue to lose revenue to scammers. The future of e-commerce belongs to the platforms that can efficiently and accurately distinguish between a return requested by a fraudster using a falsified image and one requested by a loyal customer. By expanding our focus beyond what is shown to include identifying the device that shows it, we can leave behind the cat-and-mouse game of content verification. Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?

AI 시장 분석

The advancement of AI technology is leading to a surge in return fraud using sophisticated fake images and videos, causing mounting losses for the retail industry. Companies are now strengthening defense strategies by adopting device identification and digital fingerprinting technologies to block fraudulent returns. Investors should focus on how the implementation of these security solutions will impact retail cost reduction and profitability improvement.

상승 영향

하락 영향

DYAX 전담 분석

AI-driven fraud is rapidly evolving, making it difficult for traditional systems to distinguish between legitimate and forged returns. Retailers are shifting toward advanced authentication tools to mitigate these operational risks.

The widespread adoption of device fingerprinting and real-time verification allows companies to proactively identify suspicious patterns. While these investments entail upfront costs, they are essential for long-term margin protection against the rising tide of retail shrinkage.

AI가 생성한 분석으로 투자 자문이 아닙니다.

DYAX Investor Sentiment

Bullish (Long) 70% · Bearish (Short) 30%

364 participants

Related News

원문 보기 — FORBES