How AI-Generated Receipt Fraud Can Affect Private and Public Institutions
And how TruthScan is trying to stop it.
If you ask an artificial intelligence (AI)-powered chatbot to generate a fake receipt, it likely won’t refuse. Four in 10 U.S. employees admit to using AI to generate fake expense receipts, a survey by Emburse says.
“AI-generated receipts are one of the most common and easiest items of fraud to create,” says Christian Perry, CEO of TruthScan, a company focused on detecting AI images and documents.
According to Perry, one company had more than $2 million worth of fake AI receipts submitted in just one month. Fraudsters can generate fake receipts to commit expense fraud. Perry believes both public and private institutions could be harmed.
“We have seen verified instances of receipt fraud impacting private companies,” Perry says. “But we suspect government institutions could also stand to be impacted by this type of fraud.”
Some receipt upload and rebate programs compensate people for uploading their receipts. Real money is a potential earning.
How can fake receipts be spotted?
To the naked eye, fake receipts can look real. They might contain coffee stains, wrinkles, or even the shine of a camera flash. This is where TruthScan comes in. The software performs a metadata analysis and forensic visual inspection; it looks at how things are arranged and finds visual patterns consistent with AI generation.
Perry says TruthScan is trained on more than 250 image generation sources, which he says means the detector can spot some of the most nuanced patterns. AI-made fake receipts are created using AI image-generation software.
“That’s the thing, fake receipts are just AI images. But that’s one of the ways we can detect them,” Perry explains. “Even down to a few pixel patterns, because TruthScan is trained on millions of data points, it can identify even the smallest discrepancy that human eyes can’t.”
Humans are not replaced
Even though some humans may need detection software to catch the most convincing fake receipts, they still play a vital role. Perry believes companies very much still need humans in the loop. “TruthScan is by no means a replacement for the people responsible for investigating fraud. It’s absolutely supplementary.”
There are two problems. AI receipts can flood databases in bulk because they’re easy to create and upload. They can also look so real that without detection software they might be treated as legitimate. If a receipt gets flagged, it’s just a starting point for the investigator.
Per Perry, “100%, companies need people to investigate any claims that are flagged as AI-generated. The detection layer is just the first step, it makes investigating easier and more workable. Our detection software isn’t replacing people; I think it’s absolutely doing something that, in many cases, humans can’t really do on their own anymore.”
TruthScan’s philosophy is that as new AI capabilities emerge, defense mechanisms must emerge. “We’re living through an unprecedented time in history. We’ve never seen anything like this before, and we need to fight the bad parts as fast as they come along,” Perry says.
Making AI-generated receipts obsolete
Broadly, TruthScan’s mission is to thwart generative AI fraud entirely. But soon, Perry says, TruthScan’s technology could make AI-generated receipts obsolete.
“We really want to make all AI fraud obsolete. With AI images and fake receipts, we think we’re on the cusp of making it so inconvenient for criminals that AI fraud losses drop significantly, and it becomes more effort than it’s worth for the criminals.”
Recently, TruthScan released a fake PDF detector, which uses computer vision, visual, text, and metadata analysis to identify fraudulent PDF documents created by AI.
One of the only challenges now is getting companies to adopt the latest technology as fast as the criminals do, Perry says.
Entrepreneur Media was not involved in the creation of this content.