"Boss, My Chicken Was Undercooked": Delivery Apps and Small Business Owners Struggle With AI-Generated Fake Photos Used for Refunds

Uploading AI-Generated Photos to Get Delivery Refunds
Abusing Photo-Based Refund Policies

The inside of a delivered chicken, with a bite taken out of it, appears red and undercooked. At first glance, it looks like a photo of improperly cooked food, but in reality, it is an image generated by artificial intelligence (AI), not an actual case of poor cooking. Recently, in the food delivery app industry, cases are emerging where users create photos of food defects using generative AI to demand refunds or order cancellations, and this has become a new headache for the sector.

AI-generated image of undercooked chicken. ChatGPT

On June 17, Herald Economy reported that major delivery platforms such as Baedal Minjok and Coupang Eats are suffering from malicious manipulated reviews, where users receive refunds after altering photos with AI. Both the platforms and restaurant owners are struggling with this issue.

For major delivery platforms, when reports of food quality issues or the presence of foreign substances are received, the platform reviews evidence such as photos submitted by the user to determine whether a refund is warranted. In particular, Baedal Minjok is known to process refunds by checking evidence such as photos provided by the customer and going through a store consent procedure if the customer requests order cancellation due to food quality issues. The problem is that while both major platforms are responding to refund requests involving AI-generated fake photos, neither has a system that can perfectly detect AI-generated images at this time.

The refund process on delivery platforms heavily relies on a single photo taken by the consumer. While a prompt refund is necessary for consumer protection when the food is spoiled or contains foreign substances, on the other hand, if a malicious user manipulates a photo of normal food using AI and submits it, the store owner may suffer unjust losses. The Asia Business Daily

A New Tactic That Hurts Small Business Owners

The refund process on delivery platforms heavily relies on a single photo taken by the consumer. While a prompt refund is necessary for consumer protection when the food is spoiled or contains foreign substances, on the other hand, if a malicious user manipulates a photo of normal food using AI and submits it, the store owner may suffer unjust losses. In some cases, even if the franchise owner does not acknowledge fault, the platform may still proceed with point compensation or loss reimbursement as part of customer service. It has been pointed out that in such cases, the final burden of disputes may repeatedly fall either on the store owner or the platform.

On online communities for small business owners, similar grievances are already being voiced. Typical comments include: "These days, even photos can't be trusted," "Perfectly fine food suddenly comes back looking like it's moldy," and "Consumer protection is necessary, but malicious complaints are all shouldered by the owner." Some netizens have pointed out that "such actions make even normal consumers subject to suspicion," and "because of the abuse by a few, the entire refund process will become stricter," emphasizing the need for strong sanctions against black consumers.

Both Store Owners and Platforms Struggle to Respond

The food delivery app industry's concern lies in technical limitations. Baedal Minjok currently uses AI technology to filter out certain types of review manipulation and abuse, but still does not have a dedicated system to clearly identify food photos that have been generated or edited by AI.

An image of jjambbong created using AI, containing hair. ChatGPT

As a result, fraudulent activities such as order cancellations and refund requests using AI-generated photos must be monitored on a case-by-case basis. Major delivery platforms such as Baedal Minjok and Coupang Eats are considering usage restrictions, permanent bans, and legal action when fraudulent activity is confirmed, but as the quality of generative AI images rapidly advances, it is becoming increasingly difficult to distinguish with the naked eye whether a photo is genuine or not.

An industry official explained, "At present, technological means to perfectly detect AI-generated photos are limited. Some generative AI images can be identified through metadata or embedded information, but if that part is deleted or cropped out before submission, it is difficult to determine." He added, "Fake reviews and fraudulent refund attempts harm both honest consumers and store owners. We plan to strengthen regular monitoring and take measures such as restricting usage and taking legal action against repeat malicious users in order to maintain trust in the review ecosystem."

This content was produced with the assistance of AI translation services.