Generative AI seems to be solving one of the most obvious issues with AI-generated images: the text. While this is great news for those who are tired of seeing images with gibberish in them, it could also have unintended effects. It turns out that ChatGPT can now create fake receipts.
Fake receipts aren’t always convincing, but they can be spotted if you know where to look. They’re too clean, too CGI and sometimes have the wrong sums (like our example above). They’re easy to miss, but with a little fiddling, you can make them more realistic by adding creases and food staining.
It seems that for now, the best way to do this is to have ChatGPT recreate an existing receipt. This can then be modified to show any prices you want. OpenAI isn’t worried about the new tool, but it appears to be a powerful one for digital fraudsters.
Why OpenAI is okay for now with fakes
There are too many real-world verification flows that rely solely on “real images” for proof. This era is gone. pic.twitter.com/9FORS1PWsb March 29, 2025
When we contacted OpenAI, a spokesperson informed us that it is keeping an eye out for trends such as AI-generated receipts. They told us “We monitor image generations on and off our platform, use internal tools to verify they were created by our products, and take action when we identify violations of our usage policies,” . The spokesperson for OpenAI added
“We’re always learning from real-world use and feedback, and we’ll continue refining our policies to balance creative freedom with preventing misuse. All images include industry-standard C2PA metadata indicating they were AI-generated by OpenAI,” .
OpenAI spokesperson Taya Christianon went even further, saying TechCrunch said that fake receipts could be used in situations other than fraud: “teaching people about financial literacy”
It would also be great for creating props or illustrations for articles or real or virtual staging. Fake receipts can be used to commit fraud. The more convincing the images are, the more likely they are to facilitate fraud.
The’made by ChatGPT metadata safeguards’ are fine, but exporting images without metadata is as easy as a single click. I’m not convinced that these safeguards will stop people from using fake images for fraud in the same way as social media “be nice” policy has stopped people being horrible on the internet.
I must admit, though, that it would be hilarious if OpenAI employees used the technology to inflate their expenses claims.
- Can AI help you be less gullible, or is this a conspiracy?
- I haven’t noticed any difference in ChatGPT 4o, but it just got better.
- Tasks within ChatGPT can take over your calendar to remind you to complete your to-do lists
Carrie Marshall is a writer, broadcaster, musician, and kitchen gadget enthusiast who has been writing about technology since 1998. She has contributed sage advice, odd opinions, and written more than a dozen different books to magazines and websites. Her memoir, Carrie Kills a Manis now on sale and her next book about pop music is due out in 2025. She is the lead singer of Glaswegian rock group Unquiet mind
More on artificial intelligence
