Fake Receipts Taken over by Generative AI

According to data from fintech AppZen, in March of 2025, no fake receipts were detected that were generated through the use of AI. While fake receipts did exist, they were template-based frauds. By mid-May of 2026, however, the ‘market share’ had shifted, with 70.8% of fake receipts now generated by AI tools.

Verma Kunal, co-founder and CTO of AppZen, indicated that ‘fake receipts still represent an extremely small percentage of all receipts submitted by employees,’ but added that ‘the speed and magnitude of this shift is dramatic.’

The use of fake receipts is highly concentrated to a small percentage of employees. Kunal comments, ‘AI-generated expense fraud is a high-volume, low-dollar attack pattern, deliberately sized to slip under most companies’ auto-approval thresholds.’

He cited an example where one employee submitted 45 fake receipts with an average fake receipt amount of just under $50 USD. The largest fake receipt total for a single individual totaled $12,900 USD equivalent. He cited one Fortune 10 company with a repeat offender rate of 41%.

Another reason for concern is that those generating fake receipts use multiple methods to attempt to elude detection. An example of this is that some who are submitting expense reports are adding forged scanning watermarks in an effort to make it appear to be a legitimate invoice that is scanned to a PDF or other format.

Fake receipt submissions vary by country/region and total value. AppZen has provided the following top 5 rankings:

Fake Invoice Lines (Total)

• 1            India
• 2            United States
• 3            United Kingdom
• 4            Canada
• 5            France

Value of Fabricated Invoices

• 1            United States
• 2            Australia
• 3            India
• 4            Canada
• 5            Brazil

AI is being used to replace other tools used in business workflows, and this includes inappropriate activities such as fake-invoice generation. The speed of the transition to AI tools is surprising, in this case from 0% to 70% in about 14 months. Given the risk and complexity of these cases, it is increasingly apparent why companies deploy purpose-built tools that leverage AI to catch AI-enabled fraud and falsified documents.