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

This month, ChatzPT has unveiled a new figure generator Its 4O model is as part of the model that is much better to generate the text inside the images.

People are already using it to make fake restau receipts, potentially adding one more tool Already broad toolkit AI dipfkes used by fraudsters.

Rich social media posters and VC DD slave X posts A (real) San Francisco is a picture of a fake for stakehouse that he says with 4o was made with.

You can use 4o to generate fake receipts.

There are many real verification flows that depend on “real image” as proof. That era is over. pic.twitter.com/9Fors1pWSB

– DD (@DDDS) March 29, 2025

Others were able to replicate similar results, With A food or drink stained that shows it more authentic:

I think the letters in the original figure are very perfect and they do not bend with the paper. They look like wandering on paper. Here is my attempt to make it more realistic. Tell me what you think. pic.twitter.com/eixrshubey

– Michael Gofman (@Mitchelgofman) March 29, 2025

The most real -looking example found in TechCrunch is actually France, where the LinkedIn user Post A CrINCED UP AI-IF-SUPPED Rashid for local restaurant chains:

TechCrunch 4O has been tested and was also able to create a fake receipt for Appleby in San Francisco:

However, in our efforts a number of dead discounts were given that it was fake. For one, the total uses comma instead of a period of total. For the other, mathematics is not associated. Llms Strive to make basic maths stillSo this is not especially surprising.

However, it will not be difficult for any fraud to fix a few numbers quickly with photo editing software or probably more specific prompts.

It is clear that producing fake receipts makes it really easy to present huge opportunities for fraud. It will not be difficult to imagine that this kind of technology is being used by bad actors to get “reward” for full fake expenditure.

OpenAI spokesperson Taya Christianson told TechCrunch that all its images include metadata, indicating that they were created by ChatzPT. Christianson has added that Open is “steps” when users violate its use principles and “always learn” from real-world use and response.

TechCrunch then asked why chatzipti first allowed people to make fake receipt and whether it was compatible with OpenAI Use policy (No nets restricted.)

Christanson replied that Openai’s “goal is to give users the same creative freedom” and fake AI receipts can be used in non-fascinating situations such as “teaching people on financial literacy” and can be used in addition to creating original art and product advertisements.