This month, CHATGPT unveiled a new image generator as part of the 4O model, which is much better at generating text in images.
People are already using it to generate fake restaurant receipts and could add another tool to the already extended toolkit of AI deepfakes used by scammers.
Prolific social media posters and photos of the fake receipts of xx (actual) San Francisco Steakhouse, posted by VC Deedy DAS, say they were made in 4o.
Others were able to replicate similar results, including those containing food and drink stains.
The most realistic example of TechCrunch found was actually from France. Here, LinkedIn users have posted wrinkled AI-generated receipts for a local restaurant chain.
TechCrunch was also able to test the 4o and generate fake receipts for AppleBee in San Francisco.
But in our attempts there were some dead giveaways that it was forged. One uses commas in total rather than periods. Otherwise, the mathematics will not be summed. This is not particularly surprising as LLM still struggles to do basic maths.
However, it’s not difficult for a scammer to quickly fix some numbers either with photo editing software or perhaps a more accurate prompt.
Making fake receipts easily generates a great opportunity for fraud. It is not difficult to imagine this type of technology being used because it was “refunded” for a totally false cost.
Openai spokesman Taya Christianson told TechCrunch that the image contained all the metadata, which was created by ChatGpt. Christianson added that Openai “takes action” when users violate usage policies and that it “always learns” from actual usage and feedback.
TechCrunch then asked why ChatGPT could first generate fake receipts and whether this was in line with Openai’s usage policy (prohibiting fraud).
Christianson responded that Openai’s “giving users as much creative freedom as possible,” and that Fake Ai receipts such as “Teaching people about financial literacy” can be used, along with the creation of original art and product ads.
