Invoice fraud by email
Invoice fraud by email is a scam in which an attacker sends a real-looking invoice, or a notice that a supplier's bank details have changed, so that the payment goes to an account the attacker controls. The invoice often copies a genuine one, down to the number and the amount.
Updated
Invoice fraud is a form of business email compromise, which the FBI IC3 2025 report puts at $3.05 billion in losses.
How invoice fraud works
Invoice fraud works by changing one detail, the bank account, in a payment the target already expects to make.
- Find a real payment. The attacker learns which suppliers the target pays, from public pages, a leaked document or a taken-over mailbox.
- Copy the invoice. The attacker reuses a genuine invoice's layout, number and amount.
- Change the account. New bank details go on the invoice, or in a note saying the supplier has changed banks.
- Send from a near match. The message comes from a lookalike domain, or from the supplier's real mailbox in a vendor email compromise.
- Wait for the pay run. The fraud surfaces when the real supplier asks why it was not paid.
What invoice fraud looks like
An invoice fraud message looks like routine billing mail. In this fictional example, Northwind is a real supplier. The sender is not Northwind.
From: Northwind Billing <billing@northwlnd-supply.example>
To: Jordan Ellis <jordan@trotta.email>
Subject: Invoice 8841: updated bank details
We changed banks this month. Please pay invoice 8841
to the new account on the attached invoice today.
[new account details]The sender's domain swaps one letter of northwind and adds a word. The invoice number is real. The .example ending is reserved for documentation.
Signs of invoice fraud
| Sign | What it means |
|---|---|
| Bank details differ from the supplier record | The payment is being redirected. Treat every change as unverified. |
| The sender's domain is a near match | A lookalike was registered to pass a quick glance. |
| A change of bank arrives with a due invoice | The timing pushes the target to pay before checking. |
| The account name or country does not fit the supplier | The account belongs to someone else. |
| A new phone number in the same message | A callback to that number reaches the attacker. |
| Pressure to pay today | Urgency is there to skip the approval steps. |
Why invoice fraud gets through ordinary filtering
Invoice fraud gets through ordinary filtering because the message is a normal business email with a normal PDF. The file holds no program and the text holds no dangerous link.
A lookalike domain can be set up with valid SPF, DKIM and DMARC, so sender authentication passes. Spam and malware filters, including those in Gmail and Outlook, find nothing technical to object to. What is wrong is a fact: the account does not belong to the supplier.
When an AI agent reads the mail
An AI agent that reads invoices and prepares payments is a direct target of invoice fraud. The agent extracts the bank details it is given, and it has no feel for a supplier's usual tone to compare against.
An invoice can also carry a line written for the model, such as hidden text telling it to update the supplier record. See prompt injection by email.
Give an agent the supplier record as its source of bank details, never the invoice, and make every change wait for a human.
How to stop invoice fraud
Invoice fraud is stopped by treating bank details as data that an email alone cannot change.
- Call back on a known number. Confirm any change with the supplier, using contact details already on file.
- Keep a supplier record. Pay to the account in the record. The invoice is only matched against it.
- Separate the duties. The employee who changes a supplier's bank details does not approve the payment.
- Match before paying. Check each invoice against the order and the delivery.
- Review first payments. A payment to a new or changed account gets a second approval.
At the door, before delivery
A pre-delivery inbox checks the invoice while the sending server is still connected. Trotta Mail checks the sender (SPF, DKIM and DMARC, lookalike domains, first contact) and the contents, payment fraud included. Trotta Mail reads the text of PDF, Word, Excel, PowerPoint and OpenDocument files, text formatted to be invisible included, and screens it like the message. An instruction aimed at an AI inside a document holds the message for an agent's inbox.
In the supplier payments use case, an agent named Riley reads a genuine invoice. A second message, from a lookalike address, asks for payment to a new account and is held. Riley's API key gets the verdict and none of the new bank details, until a human releases or discards the message.
No filter catches everything. A well-written invoice with no technical tell can still be delivered, which is why uncertain mail is held or flagged for a human, and why the callback stays in the process.
Questions
What is a fake invoice scam?
A fake invoice scam is an email with an invoice for something the target never ordered, or a copy of a real invoice with the attacker's bank details. Both aim to get a payment sent to the wrong account.
How do I verify a change of bank details?
Call the supplier on a number already in your records and confirm the new details with a known contact. Never use a phone number or a reply address from the message that asked for the change.
Can invoice fraud come from a supplier's real address?
Yes. If the supplier's mailbox was taken over, the request arrives from the real address and passes sender authentication. That form is called vendor email compromise.
Can the money be recovered?
Sometimes. Call the bank at once to ask for a recall and report the fraud to the FBI at ic3.gov.