- False receipts are now easy to generate with chatgpt
- They are convincing, and further fine tuning makes them look very real
- Openai says that metadata and use policies will protect us
Generative AI seems to solve one of the most obvious problems with AI-generated images: the text. And while it is welcome news for anyone who is tired of seeing pictures that contain absolute gibberish, it can also have unintended consequences – it turns out that Chatgpt is now pretty good at creating false receipts.
Fans -Forgsing is not entirely compelling if you know what to look for: They are a little too clean, a little too CGI and sometimes get the sum wrong (as in our example above). But they are also easy to miss – and with a bit of fiddling you can make them look more realistic with folds and food spots.
For the moment, it seems that the most effective way to do this is to get chatgpt to recreate an existing receipt, which can then be changed to show the prices you want.
Openai says it’s not too concerned, but it seems to be an effective new tool for digital scammers and unrest.
Why Openai is okay with the forgeries, so far
You can use 4o to generate fake receipts. There are too many verification flows in the real world that depend on “real images” as evidence. That era is over. pic.twitter.com/9Fors1pwsbMarch 29, 2025
We reached Openai, and a spokesman told us it is watching trends such as AI-generated receipts. “We monitor image generations on and from our platform, use internal tools to verify that they were created by our products and intervene when we identify violations of our use policy,” they told us.
“We always learn through use and feedback in the real world, and we continue to refine our policies to balance creative freedom of preventing abuse. All images include industrial standard C2PA metadata indicating that they were AI-generated by Openai,” the Openai spokesman added.
However, Openai spokesman Taya Christianson went a little longer and told Techcrunch that there was a positive potential use for false receipts: “To teach people about financial literacy” in non-fraud-like situations.
I can also imagine using it to illustrate articles or to create props for real or virtual staging. But false receipts can also be used for fraud, and the more compelling the images, the more fraud they will probably ease.
These ‘made of chatgpt’ metadata protection measures are fine, but exporting an image without metadata is a job with a click. I’m not sure these protective measures will do the smallest thing to stop fake images used for fraud more than social media “be lovely” policies have prevented people from being terrible on the Internet.
But I have to admit it would be a lot of fun if Openai’s own employees started using the technology to inflate their spending requirements.
