PDF security blog

Altered Invoice Fraud: Five Cases to Rehearse With Your AP Team

HTPBE Team··12 min read
Altered Invoice Fraud: Five Cases to Rehearse With Your AP Team

This article is a snapshot – content was accurate as of August 2025. The product evolves actively; specific counts, examples, and detection rules may have changed since publication – see the changelog for the current state.

Most invoice fraud does not start with a clever file. It starts with a normal-looking invoice arriving in a normal-looking thread, carrying one change that someone in accounts payable had no reason to question. The change is small: a bank account, a total, a vendor name. The money it moves is not.

It is written for an AP manager or a fraud-awareness trainer who wants cases to rehearse with the team. Each case is told from the inbox: what the victim organisation saw, where the process gave way, and which control would have stopped it. A catalogue of scheme types is a different job and lives in the five types of invoice fraud; a list of red flags is in how to spot fake invoices.

According to the FBI Internet Crime Complaint Center, business email compromise (BEC) losses exceeded $2.7 billion in 2022. The pattern behind the number is consistent: the fraud succeeds at the point where a payment instruction changes and nobody outside the email thread confirms it.

Why an altered invoice works

An altered invoice is persuasive because almost everything on it is true. The vendor name, the purchase order reference, the line items and the layout all match what accounts payable expects. Only the instruction that matters – where the money goes, or how much – has changed.

Three conditions make that enough:

  • A trusted channel. The invoice arrives from a known address, or from a thread that already contains genuine messages.
  • A familiar document. Reviewers compare the invoice with their memory of earlier ones, and it matches.
  • Time pressure. A due date, a month-end close or a ‘final reminder’ leaves little room for a second look.

The reviewer is not careless. The process asked them to check that the invoice looked right, and it did.

Five anatomies of altered-invoice fraud

The scenarios below are illustrative composites of patterns that defenders and published fraud guidance describe repeatedly. They are not reports of specific named incidents, and the amounts are examples.

1. The changed bank account

A manufacturer receives its regular monthly invoice from a long-standing supplier. Layout, logo and invoice number sequence match the previous eleven. The remittance details differ, and a covering line says the supplier has ‘moved banks’.

Where the process gave way: the invoice matched the purchase order, so it passed three-way matching. Nothing in that matching tests the bank account.

What would have stopped it: a rule that no payment goes to changed bank details until someone calls the vendor on a number from the vendor master file, not from the email.

2. The swapped total

A services firm is invoiced each month for a variable amount. One invoice arrives with a total about 8% above the usual range, and a line item that reads plausibly. The increase is small enough to look like normal fluctuation.

Where the process gave way: the amount sat inside the approver’s tolerance, and the approver had no source document to compare it with.

What would have stopped it: matching totals against the purchase order or contract rate rather than against last month, plus a second approver above a fixed amount.

3. The fabricated vendor

A company receives an invoice from a supplier it has never used, referencing a project that does exist. The invoice is well formed and the amount is modest. The sender explains that a colleague in the project team ‘arranged the work’.

Where the process gave way: no one owned the question ‘is this vendor real?’ The invoice went into the normal approval flow, and the approver assumed procurement had vetted the supplier.

What would have stopped it: a rule that a new vendor is created only after independent verification, never from an invoice alone, and that the requester confirms the work was done.

4. The replayed invoice

A genuine paid invoice is resent with a new due date and different remittance details, a few weeks after the original was settled. The invoice number is the same, or a near variant.

Where the process gave way: the duplicate check looked only for an exact invoice-number match, and the original was filed under a slightly different reference.

What would have stopped it: duplicate detection on vendor, amount and date as well as number, and a rule that any payment to different bank details for an already-paid invoice goes to a person.

5. The compromised thread

An attacker with access to a vendor’s mailbox joins an ongoing conversation about an overdue invoice. The replies are timely and in the vendor’s usual tone. The invoice attached is the real one with changed payment details, and the sender asks for the next payment ‘to the new account’.

Where the process gave way: the thread was real, the sender address was real, and the request came with context. Every signal the reviewer relied on was genuine except the instruction.

What would have stopped it: confirmation through a channel the attacker does not control, and a standing rule that the email thread is never the authority for a payment instruction.

What the five have in common

In every case the fraud sat in a field that finance teams treat as data rather than as a claim: bank account, amount, vendor identity, invoice reference. The file was trusted because its source was trusted. That is the gap a document-level check closes: it asks whether the PDF itself was changed after it was issued, which is a separate question from whether the sender looks right.

It is also the reason document checks and process controls work together rather than substituting for each other. A callback catches a payment instruction that was legitimately issued but is wrong. A document check can flag a file that was altered even when the thread, sender and context all look genuine. A fabricated vendor is a process problem first: whether a supplier exists is settled by vendor verification, not by inspecting the invoice it sent.

What a file check can and cannot show in each case

A file check answers one question: was this PDF changed after it was produced? That is useful in some of the five cases and irrelevant in others, and the team should know which before it leans on the result.

CaseWhat the attacker handed youWhat a file check can reportWhat only a process control settles
1. Changed bank accountUsually the vendor’s own invoice with the remittance block editedModification of the file, when the invoice was edited. Nothing when the attacker built a fresh invoice from scratchWhether the new account belongs to the vendor: callback to a known number
2. Swapped totalAn edited copy of a real invoiceModification of the file. A vendor’s own legitimate re-issue also shows as modified, so the vendor is askedWhether the amount matches the PO or contract rate
3. Fabricated vendorA new document that never had an originalOften only an inconclusive result with a stated reason. No edit exists to findWhether the vendor exists and was onboarded: vendor verification and requester confirmation
4. Replayed invoiceA genuine, already-paid file, resentintact when it is resent unchanged, modified when the payment details were editedDuplicate detection on vendor, amount and date; a person reviews payment to new details
5. Compromised threadThe vendor’s real file, or an edited one, plus a request in the message bodyModification only when the file was edited. Nothing when the real file is attached and the instruction sits in the emailConfirmation through a channel the attacker does not control

The pattern: a file check is strongest where the attacker touched the PDF, and silent where the attacker only changed the instruction around it. That is a statement about where the check helps, not a reason to skip either layer. Reading the file by hand is covered in what PDF metadata can and cannot prove; the cost side of the same decision is in the business case for a pre-payment check.

A 30-minute tabletop

You can rehearse this with the team without any software. Pick two of the five cases and give people only what the victim had: an invoice and the email around it.

  1. Ten minutes, individually. Each person writes down what they would check, in order, and what would make them hold the payment.
  2. Ten minutes, together. Walk the timeline below for the chosen case and mark where each control fires.
  3. Ten minutes, gaps. For every point where nobody would have stopped it, name the owner of the missing control.

An illustrative timeline for case 1, with the controls in place:

WhenWhat happensControl that fires
Monday 09:10Invoice arrives from the usual contact with a new account–
Monday 09:40Matched to the PO; amount fineThree-way match passes. It never tests the account
Monday 10:00Bank details differ from the vendor masterRule: any difference is a hold
Monday 10:15The PDF is checkedFile check: modified, with markers
Monday 11:30Callback on the number held in the vendor masterVendor says no change was requested
Monday 11:45Security and the vendor are informedEscalation path
ThursdayPayment run goes out without this invoice–

Without the hold rule and the callback, the same invoice would have been paid on Thursday and discovered when the real vendor chased it, days later, when a recall is harder. The point of the drill is that people see how few controls actually stand between the email and the bank, and who owns each one.

A defender checklist

Use this at invoice intake, before an invoice is approved for payment.

Payment instruction

  • Compare the bank details with the vendor master record. Treat any difference as a hold, not a note.
  • Confirm changes by calling a number you already held, never one from the email or the invoice.
  • Apply a cooling-off period before the first payment to changed details.

Amounts and references

  • Match the total against the purchase order or contract rate, not against the previous invoice.
  • Check for an existing invoice from the same vendor with a similar amount or date.
  • Require a second approver above a set amount and for any payment to new details.

Vendor identity

  • Create vendors from verified onboarding records, not from invoices.
  • Have the requester or budget owner confirm that goods or services were received.

The file itself

  • Run each PDF through a document check before approval and route anything flagged to a person. HTPBE? returns a verdict and named findings for a PDF invoice and does not need the original file for comparison.
  • A result with no findings does not replace the callback, and an inconclusive result goes to a person rather than counting as clean.
  • Treat a flagged result as a reason to verify with the vendor, not as proof of fraud. A document check reports whether a file was changed, not why or by whom.

After the fact

  • Keep the original email, the PDF as received and the approval trail together.
  • Know the first calls to make: your bank, your internal fraud contact, and the vendor.

If a payment has already gone

Speed matters more than anything else. Contact your bank immediately to ask for a recall, then notify law enforcement – in the United States, a complaint to the FBI IC3 – and tell the genuine vendor so they can secure their own mailbox. Preserve the original email and the PDF exactly as received, since a changed copy loses evidentiary value. Recovery is often partial or impossible, which is why the controls above sit before payment, not after it.

Checking the file before you pay

A document check does not replace a callback, but it gives accounts payable a signal that does not depend on how convincing the email is. HTPBE? analyses a PDF to report whether it was modified, with five free checks on signup and pay-per-check credit packs afterwards. Teams that want it at intake can use the tamper detection API directly in their AP workflow, and the invoice fraud detection use case describes the setup for finance teams.

For the same pattern applied to agreements rather than invoices, see contract fraud; for a closer look at supplier bank-detail requests, see vendor bank change fraud.

Conclusion

Altered-invoice fraud depends on one thing: a payment instruction that changes inside a channel everyone trusts. The five cases above differ in the field that was changed and in how the attacker got into the conversation, but the defence is the same each time – verify the instruction outside the thread, match amounts against source documents, create vendors from verified records, and check the file as well as the sender.

Share this article

Found this article helpful? Share it with others to spread knowledge about PDF security and fraud detection.

https://htpbe.tech/blog/invoice-fraud-how-criminals-modify-pdfs

Secure your workflow

Create your account – check PDFs on the web or with an API key, both ready on signup.
No sales call. Cancel any time.