Can an AI agent hedge foreign exchange exposure?

No, not by placing trades itself. A hedge is a binding contract with a bank, so it sits at depth 4 and above. An AI agent can measure foreign exchange exposure, compare it with the hedging policy, propose trades within the policy and draft the deal ticket. The treasurer approves each trade, and someone else confirms it with the bank.

Why is hedging held, even when the policy is clear?

Hedging policies look mechanical: cover a set share of forecast foreign-currency receipts over the next few quarters, using forwards with approved banks. In practice, every trade involves choices the policy leaves open. Which forecast to trust. Whether to hedge now or wait a week. Which bank to deal with. Whether a large one-off receipt should be covered at all.

Each trade also creates an obligation that cannot be undone cheaply. Closing out a forward early costs money, and a trade placed on a wrong forecast leaves the business over-hedged, which is speculation by another name. So the decision belongs to someone with a dealing mandate, and the agent's job is to make that decision faster and better informed.

What conditions does the agent work within?

• A written hedging policy. Target cover by horizon, permitted instruments, approved counterparties and limits, held in the policy library.
• Sealed exposure data. Receivables, payables and committed orders from the ledger and order system. Forecasts are declared by budget holders and labelled as such.
• Market data from an agreed feed. Indicative rates for the proposal, never used as dealing rates.
• No dealing access. The agent has no login to a bank dealing platform and no authority to instruct a bank.

What does the agent prepare, and who approves?

The separation of duties is the classic treasury split. The person who deals does not confirm or settle. The agent sits on the preparation and matching sides; it is never the dealer.

An illustrative example

Take a manufacturer that sells into two foreign markets and reports in its home currency. Its policy says to hedge a high share of committed receipts and a lower share of forecast receipts over the next four quarters.

The agent's weekly exposure report shows that a large new export order, committed last week, has pushed the next quarter's committed receipts well above the hedged amount. It proposes a forward contract for 1.2 million in that currency, maturing to match the order's payment date, and shows the indicative cost.

The treasurer reviews the proposal, asks the sales director whether the customer's payment date is firm, and approves a slightly smaller trade to allow for a partial shipment. The dealer places it. The bank's confirmation arrives and the agent matches it to the approved ticket, finding the value date one day off. The back-office analyst queries the bank, which corrects it.

What controls make it safe?

• No dealing rights for the agent. Enforced at the bank and in internal systems, not just in instructions.
• Committed and forecast kept apart. Every proposal shows how much rests on forecasts people declared.
• Proposals within policy only. A proposal that would breach a limit is refused, and the agent explains why.
• Independent confirmation. Confirmations are matched by the agent and signed by someone who did not deal.
• Everything filed. Exposure, proposal, approval, deal and confirmation are linked in one record.

How does this relate to revaluation and customer FX?

Exposure reporting and foreign currency revaluation use the same balances but answer different questions: revaluation measures the accounting effect of rates that already moved, hedging manages the effect of rates that might. The detailed exposure method is in how to prepare FX exposure reports with an AI agent. Converting currency for a customer is a separate, customer-facing activity covered in whether an AI agent can exchange currency for a customer. The wider finance controls are in AI agents for finance: books you can audit.