Yes, it can spot the red flags; it cannot conclude there is laundering. An AI agent can read invoices, shipping documents and payment instructions together, compare prices with reference ranges, check routes and parties, and flag mismatches. Deciding the mismatches amount to suspicion, and reporting, stay with a compliance officer and the money laundering reporting officer.
What makes trade-based money laundering hard to see?
Anti-money laundering controls are built mostly around payments. Trade-based money laundering hides value in the trade itself: goods invoiced at more or less than they are worth, shipped in different quantities from those invoiced, invoiced more than once, or not shipped at all. The payment can look perfectly ordinary. The laundering is in the gap between the documents and reality.
Finding that gap means reading documents that rarely sit together: the invoice, the bill of lading, the packing list, the letter of credit, the customs declaration and the payment. It also needs knowledge of what goods are worth and how they normally move. That is slow, specialised work, which is why trade finance teams often check documents for compliance with the credit's terms but not for laundering risk. An agent can do both readings at once.
Which red flags can the agent check?
- Pricing: unit prices well above or below the reference range for the goods.
- Quantity: invoiced quantity inconsistent with the shipping weight or container count.
- Description: vague goods descriptions ("spare parts", "general merchandise") on high-value shipments.
- Routing: shipments routed through jurisdictions with no commercial reason, or transhipped unusually.
- Parties: third parties paying or receiving funds with no visible role in the trade; consignee different from the buyer without explanation.
- Repetition: the same goods invoiced in several transactions, or a pattern of amended invoices.
- Fit: goods outside the customer's known line of business.
The agent flags each with the documents and figures behind it. It does not treat any single flag as conclusive; a high price may reflect quality, and an odd route may reflect a shipping line's schedule.
What may the agent prepare, and what stays with the officer?
| Work | The agent | The officer |
|---|---|---|
| Document review | Reads and cross-checks every document in the transaction | Decides which discrepancies matter |
| Price check | Compares with reference ranges and the customer's history | Judges whether a price is explained |
| Party check | Screens all parties, including vessels and ports | Decides on matches |
| Customer fit | Compares goods and routes with the customer's profile | Decides whether to ask the customer |
| Case drafting | Drafts the red-flag summary with evidence | Decides whether there is suspicion |
| Reporting | Drafts the report for the reporting officer | Files, or decides not to; see can an AI agent act as a money laundering reporting officer? |
What does a flagged trade look like?
Take a trade finance bank. A long-standing customer, a textile importer, presents documents under a letter of credit for a shipment of cotton fabric.
The agent reads the invoice, bill of lading and packing list. The invoiced price per metre is about four times the range for that grade of fabric and three times what the same customer paid six months ago. The shipping weight is consistent with the invoiced quantity, so the goods probably exist. The seller is a new supplier registered in a free trade zone. Payment is to be made to an account in a third country, held by a company with no role in the documents.
The agent drafts a red-flag summary: over-invoicing likely, third-party payment, new counterparty in a high-risk zone. It also notes what could explain the price, such as a premium specification not shown on the invoice. The trade finance officer asks the customer, through the relationship manager, for the specification and the reason for payment to a third party. The answers are unconvincing. The case goes to the money laundering reporting officer, who files.
Which controls make this reliable?
- Reference prices are sourced and dated. The agent cites where each range comes from.
- Risk-based depth. The agent applies deeper checks to higher-risk goods, routes and customers, in line with the FATF risk-based approach applied to AI agents.
- No contact with the customer about suspicion. Questions go through normal relationship channels with neutral wording.
- Flags are not findings. The record separates what the documents show from what the officer concluded.
- Not market trade surveillance. Securities trading surveillance is a different discipline; see how to review trade surveillance alerts with an AI agent.
The general principle is in our guide to AI compliance agents.
Questions people ask
Can the agent check open account trade, not just letters of credit?
Yes, if the bank receives the documents. In open account trade the bank often sees only the payment, which limits what any reviewer can check.
Where do reference prices come from?
Customs and trade statistics, industry price sources and the customer's own history. The firm decides which sources to trust.
Does every price discrepancy need escalation?
No. Discrepancies combined with other flags, such as third-party payment or high-risk routing, are what justify escalation.