OFAC Screening for AI

Applying sanctions screening to AI agent payments, a compliance frontier with unique velocity and opacity risks.

TL;DR

TL;DR: AI agents can pay thousands of counterparties a second with no human review, which makes automated OFAC screening the only way to keep those flows compliant.

Why AI payments are different

AI agents introduce two risks traditional compliance never faced. Velocity: an agent can initiate far more transactions than any human team, so a single unscreened flow becomes many violations quickly. Opacity: the agent may choose counterparties at runtime without a human in the loop, so there is no manual review to catch a bad match. Together these make pre-payment automation essential. Because an agent can complete a transaction in milliseconds, the screen must return in the same time budget, which is why sub-100 ms latency matters here.

Money rails that do not screen

Many rails that enable agent payments move money but do not screen for sanctions. x402, AP2, ACP, and Coinbase AgentKit are payment or protocol layers, not compliance layers. A developer who assumes the rail screens will be wrong, and under strict liability intent is not required.

Screening for agents

sanctionsai.dev is built for exactly this: a screening API an agent calls before paying. One HTTP call returns ALLOW or BLOCK in under 100 ms, checking 19,218 SDN names, 947 OFAC-listed wallets, and 16 jurisdictions. Integrations cover Claude Code, Cursor, Windsurf, LangChain, CrewAI, Eliza, the OpenAI Agents SDK, and Coinbase AgentKit, so the check fits inside existing agent stacks. The 4-Gate protocol, SCREEN, SCORE, STOP, and STAMP, gives the agent a repeatable decision path instead of ad hoc checks.