AI-generated fake receipts: the answer isn't better detection, it's making fraud structurally impossible

The new face of expense fraud

Since early 2025, the barrier to entry for committing expense fraud has dropped to almost zero. Generative AI tools can produce a convincing restaurant, hotel or taxi receipt in seconds — for free, with no technical skill required.

The numbers are sobering: according to AppZen, cited by the Financial Times, 14% of fraudulent documents submitted in September 2025 were AI-generated. A figure that did not exist 12 months earlier. And a SAP Concur survey published in early 2026 found that 32% of finance professionals admit they would not be able to recognize an AI-generated receipt during a manual review.

The classic response: an arms race you can't win

Faced with this threat, many companies are tightening their controls, training their accounting teams, adding validation layers. That's understandable. But it's a short-term answer.

Generative AI models are improving faster than human detection capabilities. Every improvement in control is followed by an improvement in forgery tools. It's a race where the defender starts structurally behind.

You have to change the playing field.

The real answer: digitize the origin of the data

The question isn't "how do we better detect fake receipts?" but "how do we make sure a fake receipt is worthless?".

The answer is one simple idea: if the proof of the expense no longer passes through the employee, it can no longer be falsified by the employee.

This is the goal of fully digitizing the expense report flow — a process where 90% of expense lines come from a digital flow, not from manual entry.

Lever 1 — Rolling out the corporate card

When an expense is paid with a corporate card, the key transaction data comes directly from the bank: amount, date, location, vendor. These elements are locked at the source and cannot be modified by the employee.

The receipt is still required — it documents the detail of the expense. But any attempt at forgery immediately runs into bank data that cannot be falsified. Inflated amount? The bank says otherwise. Fictitious vendor? Geolocation contradicts it. Modified date? The statement contradicts it.

The corporate card doesn't replace control. It makes it far more solid.

Lever 2 — E-invoicing flows

For the most common travel expenses — restaurants, hotels, transport — a second lever exists: electronic invoicing directly from the supplier.

In this model, the proof of purchase no longer passes through the employee. It flows directly from the supplier to the company's management system, through a structured connection. The receipt is born where it is sold, not where it is spent. No human intermediary on the document.

This is exactly what the French e-invoicing reform enables, with mandatory rollout set for September 1, 2026 for all VAT-registered companies. And it's the direction set by the ViDA directive for the entire European Union.

What's left to control

With these two levers combined, cash payments and non-traceable flows become marginal in the overall volume of expense reports. They don't disappear, but they represent a much smaller share of the whole.

And this residue is exactly where anti-AI control makes the most sense: focused, targeted, effective. You no longer try to control 100% of receipts with the same tools. You apply surgical scrutiny precisely where the risk is real.

You no longer detect fakes. You make them structurally impossible.

What about your company?

The September 1, 2026 reform is approaching. Is your expense process ready to integrate e-invoicing flows? Do you know what share of your spend still goes through manual employee entry?

These questions deserve an answer before regulation forces one. Contact us to discuss it.