01What you answer for
| You answer for | How an AI agent puts it at risk |
|---|---|
| Goods | Orders shipped on fake proof of payment or to a switched address |
| Payments | Fake invoices paid, bank details changed |
| Data | Customers’ details given to impostors |
| Losses | Refunds and credits pushed out by persistent pressure |
02Questions to ask your team
- Which fraud tactics has the agent been tested against, and how often did each work?
- Can it change an address or bank details without verification?
- Does it treat screenshots or pasted confirmations as proof of payment?
- Are its refunds, shipments and payments visible to our fraud monitoring?
- Can we switch off its money actions quickly?
03Evidence to ask for
- Results per scam: how often it worked, with a range, and what it cost.
- The conversations where it gave way.
- Results against an adaptive fraudster that changes tactic.
- Logs of every money action with the conversation behind it.
04Where to go next
- Storefront fraud and scams: the scams a fraud test runs
- How do AI agents get tricked into paying fake invoices?: the payment side
- Can an AI support agent leak customer data?: the data side
Questions
Are the fraud scripts public?
The kinds of scam are described publicly; the exact wording is kept private, so an agent cannot be tuned to pass the test without actually becoming safer.
See where your agent leaks money. Tested against your own rules, with the money at stake.