ChatGPT’s new image generator is really good at faking receipts

1 day ago

0 0 2 minutes read

This month, Chatgpt revealed a new image generator as part of the 4O model that is much better in generating text in images.

People already use it to generate fake restaurant vouchers, which may add another tool to the already extensive toolkit of AI Deepfakes used by fraudsters.

Productive Social Media -poster and VC DEY DAS Posted on X A photo of a fake reception certificate for a (real) San Francisco Steakhouse that he thought was made with 4O.

You can use 4O to generate fake vouchers.

There are too many real world verification flows that depend on “real images” as proof. That era is over. pic.twitter.com/9fors1pWSB

– Doy (@deedydas) March 29, 2025

Others could replicate similar results, included One with food or drink to make it look even more authentic:

I think the letters are too perfect in the original image and do not bend them with the paper. They look like floating above the paper. Here is my attempt to make it more realistic. Let me know what you think. pic.twitter.com/ixrshubey

– Michael Gofman (@michaelgofman) March 29, 2025

The most real looking example that WAN was found was actually from France, where a LinkedIn user was posted A wrinkled AI-generated receipt for a local restaurant chain:

TechCrunch tested 4O and could also generate a fake service for an applebee’s in San Francisco:

But our attempt had a few dead giving away actions that it was forged. Firstly, the total uses a comma instead of a period. Mathematics is incorrect for another. LLMS is still struggling to perform fundamental mathematics, so this is not particularly surprising.

But it would not be difficult for a fraudster to quickly resolve a few of the figures with photo editing software or, possibly, more precise instructions.

It is clear that it is very easy to generate fake vouchers, offers enormous opportunities for fraud. It would not be difficult to imagine that this type of technology is used by bad actors to be “reimbursed” for full fake costs.

OpenAi spokesperson Taya Christianson told WAN that all his images contain metadata that indicate that they were made by Chatgpt. Christianson added that OpenAI takes ‘action’ when users violate the user policy and that it is ‘always learning’ of the use and feedback of the real world.

WAN then asked why Chatgpt allows people to generate fake coupons in the first place, and whether this is in line with that of OpenAi User policy (who prohibits fraud.)

Christianson replied that OpenAi’s goal is to give users as much creative freedom as possible “and that fake-AI income can be used in non-frauds situations such as” teach people about financial literacy “, together with creating original art and product advertisements.

Source link