OFAC screening for AI agent developers

Agents that pay on a user's behalf must screen the counterparty first, using the 4-Gate protocol.

TL;DR

TL;DR: AI agents that move money must screen the counterparty before paying, because rails like x402, AP2, ACP, and Coinbase AgentKit move funds but do not screen, so the agent developer owns the check.

Agents now move money

Autonomous agents are becoming payment actors, paying for services and subscriptions on a user's behalf. The rails they use (x402, AP2, ACP, Coinbase AgentKit) transfer value but do not perform OFAC screening. That leaves the agent developer responsible for making sure the agent does not pay a sanctioned counterparty, under strict liability with penalties starting at $356,000.

The 4-Gate Agent Payment Protocol

The 4-Gate protocol fits the agent loop directly: SCREEN the counterparty, SCORE the risk, STOP if flagged, and STAMP the decision into a record. Screening is a single HTTP call under 100 ms, so it does not slow the agent down, and it returns ALLOW or BLOCK the agent can act on before any payment is authorized.

Integrations

• Editors: Claude Code, Cursor, Windsurf.
• Frameworks: LangChain, CrewAI, Eliza, OpenAI Agents SDK.
• MCP: pip install sanctions-mcp.