Numbers
Fake and edited documents are rising. What the numbers say, and how to check yours
Landlords get bank statements, small businesses get invoices, employers get payslips. Fraud specialists say faked versions of exactly those documents grew fast in 2025, helped by AI. Here’s what was measured, by whom, and what you can do with an ordinary PDF on your own.
Inscribe, a company that checks documents for banks and lenders, reported in January 2026 that the monthly volume of AI-generated document fraud it detected grew nearly fivefold between May and December 2025, and that about 6% of all documents it processed in 2025, roughly 1 in 16, were flagged as fraudulent. Bank statements worried risk leaders most. You can’t run a bank’s detection system at home, but you can check a PDF’s history, signatures, fonts and covered text in minutes.
The numbers
| Finding | Source |
|---|---|
| Monthly volume of AI-generated document fraud up nearly fivefold, May to December 2025 | Inscribe, 2026 Document Fraud Report, January 20, 2026 |
| About 6% of all documents processed in 2025 flagged as fraudulent (1 in 16) | Inscribe, same report; data from millions of documents across hundreds of banks, credit unions, fintechs and lenders |
| 85.6% of fraud and risk leaders named bank statements as the document they worry about most; pay stubs and business financials next | Inscribe survey of 90 fraud and risk leaders, November–December 2025 |
| Among AI-generated fakes flagged in mid-2026: bank statements 23.7%, invoices 13.9%, payslips 13.9% | Inscribe, July 2026 update to the report |
| “AI-assisted fraudulent paystubs soar 500%”, citing Inscribe’s April–December 2025 data | Frank on Fraud, January 30, 2026 |
| A banking consultant calls AI-made bank statements and utility bills “nearly impossible to spot”, especially when submitted electronically | American Banker, “This is an arms race”, February 5, 2026 |
Keep in mind: Inscribe sells fraud detection, and its numbers come from documents its customers chose to check, which skews toward risky applications. The trend is consistent across the sources, but “1 in 16” describes documents submitted to lenders, not every PDF in the world.
Who receives these documents
Not only banks. Anyone who decides something based on a PDF someone else sent:
- landlords checking a tenant’s bank statements and payslips;
- small businesses paying supplier invoices, where a changed bank account number is a classic scam;
- employers and schools receiving certificates and letters;
- anyone getting back a contract “signed and unchanged”.
What you can check yourself
An AI-generated document made from scratch can be very clean. But many fakes are edits of real documents, and edits leave traces in the file. Our guide on how to check if a PDF was edited walks through each one; in short:
- Signatures: bank and payroll PDFs are often digitally signed. A signature that reports changes after signing is the strongest signal you can get.
- Saved versions: a statement a bank produced once shouldn’t have been saved on top of itself three times.
- Fonts: a balance in a slightly different font from the rest of its column.
- Covered text: copy all the text on the page; the old figure is often still under a white box.
- Metadata: a “bank statement” produced by a consumer PDF editor.
- The source: when it matters, ask for the document through the bank’s or employer’s own channel.
None of these proves fraud alone, and a clean result doesn’t prove a document is genuine. They tell you when to ask more questions.
Keepsign’s What changed? runs these checks on your iPhone, iPad or Mac, without uploading the document. It lists every save in the file with its program and date, shows text changed after a signature word by word (“30” → “14”), and flags figures in a different font from their column, text covered by boxes or by other text, hidden layers and dates that don’t add up, each with a button to the spot on the page. It never says “fake”: it shows the evidence and its limits. More about What changed?
Sources
- Inscribe, “2026 Document Fraud Report”, published January 20, 2026, updated July 2026: inscribe.ai/reports/2026-document-fraud-report
- Frank on Fraud, “Report: AI-assisted fraudulent paystubs soar 500%”, January 30, 2026: frankonfraud.com/report-ai-assisted-fraudulent-paystubs-soar-500/
- Penny Crosman, “‘This is an arms race’: Banks cope with AI-faked documents”, American Banker, February 5, 2026: americanbanker.com/news/this-is-an-arms-race-banks-cope-with-ai-faked-documents