OFAC screening for AI agent payments
AI agents that move money must screen every counterparty before paying, because pre-payment screening is the minimum viable control.
TL;DR
TL;DR: AI agents that process payments must screen every counterparty against the OFAC SDN List before funds move. Pre-payment screening is the minimum viable compliance control.
Why agents need a built-in gate
An AI agent that pays vendors, buys services, or routes funds is acting as a payment intermediary, and OFAC strict liability means intent is not required. If an agent pays a wallet on the Specially Designated Nationals and Blocked Persons List, the violation has already happened by the time anyone reviews it. Payment rails like x402, AP2, ACP, and Coinbase AgentKit move money, but they do not screen it.
The 4-Gate Agent Payment Protocol
sanctionsai.dev applies four gates to every agent payment: SCREEN, SCORE, STOP, and STAMP. The agent first screens the counterparty, scores borderline matches for risk, stops on a flagged result, and stamps the decision into the log. A single HTTP call returns clean (ALLOW) or flagged (BLOCK) in under 100 ms, against 947 OFAC-listed crypto wallets and 19,218 SDN names synced hourly.
What to wire in
- Call sanctions_check before every payout, not after settlement.
- Use risk_score for fuzzy or partial matches.
- Log each decision so the agent's actions are auditable.
If the agent can spend, it must screen; there is no other defensible position.
Because agents scale faster than any human review team, the screening gate must be part of the agent's code path, not a monthly spreadsheet.