Expense Fraud in the Age of AI: What We Learned from Testing Deepfake Receipts
We generated deepfake receipts and built a forensic pipeline to catch them — what AI still gets wrong, and why Photoshop remains the bigger threat.
The rise of generative AI means receipts can be faked in seconds. However, the flaws are still visible to those who know what to look for.
Receipts aren't just scraps of paper; they're essential for forensic audits and fraud investigations. In conversations with clients and integrity colleagues, one growing concern keeps surfacing: AI “deep-fake” receipts. With free tools, anyone can now create photorealistic restaurant stubs, taxi slips, or airfare confirmations — raising real risks for audit trails and expense reviews.
Over the summer, STB Integrity set out to test this threat directly. Consultant Lannie Su, CFE, led a project comparing AI-generated and real receipts, combining human observation with a Python-built forensic pipeline. The findings were both sobering and practical.
📥 Want to see what the differences look like?
We've created a visual guide breaking down 5 forensic tests used to detect AI-generated receipts.
Continue reading
This article runs to about 1,167 words. The rest is part of membership, along with the full searchable library and the Job Corner.
Topics: AI & Tech · Audit & Forensic Accounting · Fraud & Corruption
