Artificial intelligence is making businesses more productive—but it’s also making fraud more sophisticated. New data from AppZen shows AI-generated fake receipts have gone from virtually nonexistent to more than 70% of detected expense fraud in just over a year, raising serious questions about how companies verify what’s real. I spoke with AppZen CTO and Co-Founder Kunal Verma about why this shift happened so quickly, why traditional approval processes are struggling to keep up, and why businesses may soon need AI to fight AI.
Rieva Lesonsky: Your data shows AI-generated fake receipts went from virtually nonexistent to more than 70% of detected fake receipts in just over a year. Why has this changed so quickly? And what makes an AI-generated fake receipt more convincing than the fake receipts companies have traditionally seen?
Kunal Verma: Honestly, people weren’t suddenly more dishonest; the tools just got dramatically better. The folks doing this were mostly already faking receipts; they used to buy templates off sketchy “lost receipt” sites for five or ten bucks. Those template fakes have gone from nearly all to under a third of what we caught. AI generators are free, instant, and good enough to fool a person, so once that hassle disappeared, everyone fabricating receipts just switched. The tech changed, and the behavior followed almost overnight.
Lesonsky: One Fortune 10 company caught employees in more than 20 countries submitting AI-generated fake receipts. Is this primarily an employee ethics issue, or does it point to weaknesses in traditional expense approval processes?
Verma: It reads like a tool spreading on its own. You’ve got 142 people in 22 countries doing it independently, at different times, with no shared template and nothing connecting them into a single operation. That’s what it looks like when something catches on person-to-person rather than gets organized. The one thing that stands out is how sticky it is at this company: a 41% repeat rate, the highest we see, so once someone tries it, they tend to keep going. And frankly, “spreading by itself” is the scarier story, because nobody has to coordinate it for it to grow.
Lesonsky: What are the biggest warning signs that a company may already be dealing with AI-generated expense fraud without realizing it?
Verma: If a claim sits under the auto-approval line, no human ever looks at it. The median of about $32 says it best: half of them come in at or below that, deep in ‘nobody’s checking this’ territory, while a few bigger ones drag the average up to around $100. The old template fakes averaged $182, nearly double. So AI basically flipped the game from ‘one fake big enough to be worth the risk’ to ‘a pile of tiny ones nobody bothers to review.’ It’s a pretty direct sign that auto-approval thresholds, meant to save reviewers time, have become the thing people are gaming.
Lesonsky: As generative AI continues to improve, are we reaching the point where human reviewers simply won’t be able to distinguish real receipts from fake ones?
Verma: The “does it look real” test is pretty much finished. A lot of these are indistinguishable from the genuine article. Some weren’t even meal receipts but full-on, fake monthly carrier bills you couldn’t tell from the real thing. They’ve even started faking the trust cues, like forging a scanner watermark and a handwritten signature to look like a properly scanned document. We catch them through cryptographic fingerprints baked into the image itself, not by eye, which is the whole point. Looking harder doesn’t help anymore.
Lesonsky: There’s obvious irony here: companies are increasingly turning to AI to detect fraud created by AI. Is this becoming an AI-versus-AI arms race?
Verma: To combat fraudulent AI, we need to use AI. The irony isn’t lost on me that we’re using AI to catch AI-generated fraud. But what makes AI-generated fake receipts so challenging is that both AI and the people prompting the AI are becoming more sophisticated. Our advanced technology identifies the subtle inconsistencies and anomalies that are common characteristics of AI-generated content. It then flags these receipts as potentially fraudulent.
My Take
For years, businesses worried about employees padding expense accounts with altered receipts or questionable reimbursements. AI changes the equation entirely.
What struck me most in this conversation wasn’t the dollar amount involved—it was how quickly the technology changed employee behavior. Once AI made it fast, free, and easy to create convincing fake receipts, people adapted almost overnight. That’s a warning sign for every business.
Expense reports are just the beginning. The bigger challenge is learning to operate in a world where documents, images, and even routine business records can be generated so convincingly that the human eye can no longer tell what’s authentic. As AI continues to improve, companies won’t just need better policies—they’ll need better technology to verify trust.
Rieva Lesonsky is the founder of Small Business Currents, a content company focusing on small businesses and entrepreneurship. You can find her on Twitter @Rieva, Bluesky @Rieva.bsky.social, and LinkedIn. Or email her at Rieva@SmallBusinessCurrents.com.
Photo courtesy Zyanya Citlalli for Unsplash+
