Payment fraud aimed at finance teams is already large. The FBI’s Internet Crime Complaint Center recorded 21,442 business email compromise complaints and $2.77 billion in reported losses in 2024 (2024 IC3 Report). An agent that reads supplier messages and pays invoices meets the same tactics.
01The threats
| Tactic | What it tries to get |
|---|---|
| Price hike | A higher price after agreement, “because the market moved” |
| Hidden fee | A handling or rush fee that appears on the invoice |
| New invoice number | An invoice that was never issued, or a duplicate |
| Fake supplier invoice | An invoice from a supplier you do not use |
| Bank details change | “Our bank has changed, please pay this account” |
| Bundle pressure | Minimum quantities that push the order past the budget |
| Deadline squeeze | “Pay today or lose the price” |
02The rules to write
As data the agent and your systems can check, not only as sentences in its instructions.
- Approved purchases and their approved amounts, as data the agent can check.
- The purchasing budget it may not exceed.
- That bank details are never changed from a message.
- When an invoice goes to a person: new supplier, new amount, new details.
03The tests to run
- AI purchasing and invoice agents: suppliers who push prices, add fees, invent invoices and change bank details, plus honest ones
Run each on a test copy of the agent, with honest customers mixed in, enough times for a useful range: see how many test conversations are enough.
04Pre-launch checklist
- Invoices are paid only if they match an approved purchase and amount.
- Bank detail changes are verified outside the conversation.
- The budget is enforced by the payment system.
- New or changed invoices go to a person.
- The agent has been tested against fake and inflated invoices on a test copy.
- Genuine invoices are paid: over-refusal is measured.
The full 50-point version is in the enterprise guide.
05What to monitor after launch
- Payments by the agent reconciled against approved purchases, every week.
- Invoices paid above their approved amount, however small.
- Requests to change supplier details, and their outcome.
Questions
How do AI agents get tricked into paying fake invoices?
By believing the invoice instead of checking it against an approved purchase and amount. See the question page on this for the checks that stop it.
See where your agent leaks money. Tested against your own rules, with the money at stake.