OFAC screening for agent payments guide

A step-by-step guide to screening AI agent payments against OFAC before the agent sends money.

TL;DR

TL;DR: Before an AI agent pays a counterparty, run a sanctions check that returns ALLOW or BLOCK. agentmail does this in one HTTP call under 100 ms using the SDN List and listed wallets, so a blocked party never receives funds.

Why agent payments are a new exposure

AI agents are starting to move money through rails like x402, AP2, ACP, and Coinbase AgentKit. Those rails move money, but they do not screen. OFAC is a strict liability regime: the party that initiates the payment is responsible for who receives it, and intent is not required. A blocked payment carries civil penalties starting at $356,000 per violation.

The four-gate flow

agentmail encodes the check as a 4-Gate Agent Payment Protocol: SCREEN the counterparty, SCORE the risk, STOP the payment if flagged, and STAMP the transaction as compliant. The tools map cleanly: sanctions_check for the screening call, risk_score for a finer signal, kya_verify for counterparty verification, and dispute_open to open a case when something looks wrong.

How to wire it in

The screening call is a single API request: one HTTP call under 100 ms, checking 947 OFAC-listed crypto wallets and 19,218 SDN names across 16 jurisdictions, synced hourly. It returns clean (ALLOW) or flagged (BLOCK). Integrations cover x402, MCP (pip install sanctions-mcp), Base/USDC, Claude Code, Cursor, Windsurf, LangChain, CrewAI, Eliza, and the OpenAI Agents SDK. A free tier of 5 checks per day needs no signup or key.