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

This month, ChatGPT unveiled a brand new picture generator as a part of its 4o mannequin that may be a lot higher at producing textual content inside photos. 

Individuals are already utilizing it to generate faux restaurant receipts, probably including one other instrument to the already-extensive toolkit of AI deepfakes utilized by fraudsters.

Prolific social media poster and VC Deedy Das posted on X a photograph of a faux receipt for a (actual) San Francisco steakhouse that he says was created with 4o.

Others had been in a position to replicate related outcomes, together with one with meals or drink stains to make it look much more genuine:

Probably the most real-looking instance Trendster discovered was really from France, the place a LinkedIn consumer posted a crinkled-up AI-generated receipt for an area restaurant chain:

Trendster examined 4o and was additionally in a position to generate a faux receipt for an Applebee’s in San Francisco:

However our try had a few lifeless giveaways that it was faked. For one, the whole makes use of a comma as a substitute of a interval. For an additional, the maths doesn’t add up. LLMs nonetheless battle to do primary math, so this isn’t significantly stunning.

However it wouldn’t be onerous for a fraudster to rapidly repair a couple of of the numbers with both photograph enhancing software program or, probably, extra exact prompts.

It’s clear that making it very easy to generate faux receipts presents enormous alternatives for fraud. It wouldn’t be onerous to think about this sort of tech being utilized by dangerous actors to get “reimbursed” for completely faux bills.

OpenAI spokesperson Taya Christianson advised Trendster that every one of its photos embrace metadata indicating they had been made by ChatGPT. Christianson added that OpenAI “takes motion” when customers violate its utilization insurance policies and that it’s “all the time studying” from real-world use and suggestions.

Trendster then requested why ChatGPT permits folks to generate faux receipts within the first place, and whether or not that is according to OpenAI’s utilization insurance policies (which ban fraud.)

Christianson replied that OpenAI’s “purpose is to present customers as a lot inventive freedom as attainable” and that faux AI receipts could possibly be utilized in non-fraud conditions like “instructing folks about monetary literacy” together with creating authentic artwork and product adverts.