TL;DR: A fake invoice is either a genuine invoice edited after the seller's system produced it, or a document generated to describe a sale that never happened. It shows up as collateral in factoring and supply-chain finance and as a payment instruction in accounts payable. The FBI's 2025 Internet Crime Report recorded 24,768 business email compromise complaints and $3.046 billion in losses, and the 2026 AFP Payments Fraud and Control Survey found BEC hit 74 percent of organizations, mostly through impersonated vendors changing payment instructions. Visual review fails because a competent fake carries the right logo, the right layout, and the wrong history.
An Invoice Is a Claim, Not a Receipt
A fake invoice asserts a receivable that does not exist, exists in a different amount, is owed to someone else, or is payable to an account the seller does not control. No institution stands behind an invoice. Everyone downstream — factor, supply-chain finance platform, AP clerk, wire desk — takes the seller's word that goods moved and money is due. That makes it useful as collateral, turning a fictitious or already-financed receivable into cash, and as a payment instruction, turning a real debt into a payment to the wrong account.
The FBI's 2025 Internet Crime Report put business email compromise at 24,768 complaints and $3.046 billion in losses. The 2026 AFP Payments Fraud and Control Survey found BEC affected 74 percent of organizations in 2025, with impersonated vendors requesting payment-instruction changes the leading pattern.
Four production paths cover most files. Edited-after-creation starts with a real invoice and changes one field: amount, due date, debtor name, or remittance account. Template-generated fakes build a fresh invoice for a receivable never earned, to a debtor that is invented or real but uninvolved. Double-pledged invoices are genuine receivables sold to two lenders at once; they pass every document test because the document is authentic. AI-generated invoices use a language or image model to produce a plausible bill in a target company's style, the newest reason a PDF that looks issued may never have touched an accounting system.
Tells That Still Work, and the Ones That Don't
Logos and layout tell a reviewer nothing, because every real invoice template becomes public the moment it is emailed to a customer. The tells that survive connect the document to something outside itself.
Real invoicing systems number in sequence, so a high number on an early date, or a number already used on a different invoice for a different debtor, is a production tell. A VAT number, EIN, GST or ABN can be checked against the issuing register. Line items should total and tax should equal the jurisdiction's rate, because edits to one field rarely rebuild every downstream number. Debtor plausibility catches template fakes: does the buyer exist, does it plausibly buy this product in this quantity, and has the seller ever dealt with it before? A remittance account that differs from every prior invoice from the same vendor is the strongest tell in AP, which is why wire fraud controls treat a beneficiary change as a verification event rather than a data update. Duplicate detection catches double pledging and near-duplicate resubmission, and only works across the portfolio.
The row that costs lenders the most is duplicate detection. The Secured Finance Network's 2026 Blue Ribbon Fraud Task Force report, which analyzed 26 asset-based lending and factoring fraud cases, found accounts receivable and billing schemes in roughly 58 percent of them, including the same invoices submitted to a bank lender and a factor at once. Its account of the First Brands collapse — fabricated invoices and purchase orders, assets double-pledged through special purpose vehicles — names direct verification with account debtors, bypassing the borrower's contacts, as the control that would have caught it.
When the Invoice Is the Laundering
Invoice manipulation is also a money laundering method. FinCEN's advisory on trade-based money laundering (FIN-2010-A001) lists over-invoicing, under-invoicing, and invoicing the same goods more than once as the basic techniques for moving value under cover of trade. In trade-based money laundering the invoice was genuinely issued; it misstates price, quantity, or existence so that criminal proceeds look like settlement of a commercial debt. Document authenticity is therefore necessary and not sufficient. The review has to reach why this buyer, why this price, what shipped, and whether both sides recorded the same trade.
How Detection Actually Works
Detection that holds up reads the file's story and then reconciles it to the trade. Six classes of signal cover edited-after-creation, generated-from-scratch, and duplicate presentation. They are questions a reviewer should answer about any invoice behind a funding or payment decision, not a recipe for defeating a check.
Production method asks how the bytes were made. Invoices from an ERP carry that platform's fingerprint; invoices that left a desktop PDF editor or a generative pipeline carry a different one. Edited-after-creation is a distinct finding from generated-from-scratch, because a genuine issuer file may sit under a later save and the case note should say which fields moved. Timestamp trail asks whether creation, modification, and submission dates cohere with the invoice date and payment terms. An invoice dated in March, generated in September, and uploaded the day a borrowing base was due needs a benign explanation. Re-saves create noise; the trail is evidence to weigh, not an auto-reject.
Issuer matching asks whether the claimed producer matches the seller named; a vendor that has billed from the same system for three years does not suddenly issue from consumer software. Consistency is the underwriting check: line items to totals, tax to rate, PO reference to a real purchase order, remittance account to vendor history. Model artifacts asks whether generative tooling left traces a standard viewer does not show, including the structural absences of screenshots and image-only PDFs. Recycled patterns asks whether this template, this debtor-and-amount pairing, or this exact receivable has appeared before; a reviewer looking at one file will not see the fifth copy. Document fraud controls across a bank follow the same order: establish what the file is before deciding what it proves.
What to Do With a Suspect Invoice
A flagged invoice is not yet a rejected receivable or a stopped payment. Separate the findings. Arithmetic breaks, invalid tax identifiers, and issuer mismatches are usually enough to hold the file and request the native export from the seller's accounting system. A changed remittance account is a verification event on its own.
The strongest break-test is to confirm the receivable with the debtor, using contact details sourced independently of the seller, and to match the invoice to the purchase order and proof of delivery a real sale generates. A confirmation that arrives through an address the borrower supplied is the borrower confirming itself. Real receivables also produce deposits from the named debtor within terms; a payoff from another finance company instead is the double-pledging signature, and a UCC or equivalent lien search belongs in the same review.
Escalate when the document and the trade disagree and the seller cannot restore agreement. Route to fraud and write the inconsistency the way a SAR narrative needs it: which invoice, which field, which independent record it failed to match. Reject the document, not automatically the customer, when no authentic replacement arrives within policy time. "The invoice looked wrong" is not a basis. "The invoice was modified after the issuer generated it and the debtor does not recognize the PO" is.
Break-test the control itself. Check a document in the Watchdoc playground with a native invoice PDF — clean or disputed — and compare the x-ray to what the reviewer recorded. The first file is free, no email required.
Where Sphinx Fits
Sphinx scores the invoice the way a forensic reviewer would, then shows the evidence. Watchdoc runs six checks — production method, timestamp trail, issuer matching, consistency, model artifacts, and recycled patterns — and returns a verdict with the manipulation highlighted. Published figures: 94.3% correct verdict, 2.8x more forgeries caught, clean files clearing in under 28 seconds, 1 million documents processed, $0.45 per document with no seats and no platform fee. The Watchdoc playground is the same x-ray, free to try. Sphinx Doc Fraud is built for edited-after-creation as well as generated fakes, which is the split invoice files need. The companion checklist is how to spot a fake bank statement.
Frequently Asked Questions
What is a fake invoice?
A fake invoice asserts a receivable that does not exist, exists in a different amount, is owed to a different party, or is payable to an account the seller does not control. It can be a genuine invoice edited after creation, a template-generated invoice for a fictitious sale, a real receivable pledged to more than one lender, or an AI-generated invoice.
How do factoring companies verify invoices?
Factoring companies verify invoices by confirming the receivable with the account debtor through independently sourced contacts, matching the invoice to purchase orders and proof of delivery, checking the seller's numbering and tax identifiers, and running lien searches for prior assignments. File forensics on the PDF come first.
What is a double-pledged invoice?
A double-pledged invoice is a genuine receivable pledged to two or more lenders at the same time. The document is authentic, so detection depends on comparing invoices across the portfolio, lien searches, and noticing that collections arrive from another finance company rather than the named debtor.
Can AI generate a convincing fake invoice?
Yes. Language and image models can produce a plausible invoice in a specific company's layout that passes a visual check. Detection relies on signals the page does not show: production method, timestamp trail, model artifacts in the file structure, and reconciliation to the seller's numbering, the tax register, and the debtor's records.
Should a bank file a SAR for a fraudulent invoice?
A bank should file a SAR when it knows, suspects, or has reason to suspect that a transaction involves fraud, including payments directed by altered invoices or financing obtained on fabricated receivables, whether or not a loss occurred. FinCEN asks filers to use the key term "BEC FRAUD" for email-compromise schemes and "TBML" where an invoice misstates a trade.
.png)
