32% admit they cannot recognize AI generated fake receipts

A recent survey has found that more than a third of accountants conceded that they would not be able to recognize an AI-generated receipt if it came across their desk, a number that will likely grow higher as image generation models become more powerful. 

AP solutions provider Medius surveyed 1,000 finance professionals in the US and UK and found 30% are already reporting a rise in fake receipts since the beginning of last year, and 32% say they wouldn’t be able to recognize a fake receipt from an AI image generator if they came across one. Presumably the 30% and the 32% are distinct from each other, as one would probably need to be able to recognize such fake receipts in order to report that their number is rising.

Meanwhile, the poll also found that 42% of finance professionals have suspected a colleague of submitting a fake or altered receipt, while 34% say they’ve been pressured to approve an expense that didn’t seem legitimate. When asked about what some of these very questionable expenses are, respondents named things such as a diamond ring, a luxury car, fees for a Japanese school, and even expenses for a strip club.

AI-generated forgeries of financial documentation, such as receipts, is a growing issue that has come with the sophistication of image generation models, which are now capable of  producing fake documents realistic enough to fool certain automated systems (see previous story). While forged documentation has always been an issue, for a long time it was difficult to do at scale as it often involved things like literally using whiteout on the receipt. However, AI image generation has essentially poured gasoline onto the fire, particularly with the launch of OpenAI’s image generation update in March, when people found that, with just a simple prompt, they could produce realistic-looking receipts, invoices and other records, sometimes even adding wrinkles or smudges for extra verisimilitude. Very quickly people realized the fraud potential of such tools, as these images were found to fool even certain automated software systems. 

With the release of ChatGPT 5.0 yesterday, Gary Hall, chief product officer at Medius, felt this issue would become even more dire. 

“We’re at a tipping point. GPT-5.0 promises even more realism, more precision and more ease for the user. That’s great for innovation, but it’s also a gift to fraudsters. When AI-generated documents are indistinguishable from the real thing, legacy finance systems simply can’t cope. This is no longer a niche IT issue – it’s a frontline finance challenge,” he said.