Governing AI-Initiated Payments at Your Bank

Your commercial clients’ treasury systems are already making AI-initiated payments through FedNow and RTP without a human approving each transaction in the moment. Most bank fraud controls were designed for human behavior. When an AI agent transacts at midnight at institutional velocity, it looks like fraud, and it may not be. Here is how senior leaders at community and regional banks can tell the difference, using data the rail already delivers.

Why fraud controls misfire on AI-initiated payments

Fraud models flag what does not match historical behavior: the time a payment occurs, how fast it moves, its size relative to account history. AI-initiated payments break those patterns by design. The flag is often correct to fire. The problem is that behavior alone cannot separate a legitimate AI-initated payment from an attack, because at the behavioral level they look the same.

The signal that tells them apart is already in the message

FedNow and RTP run on ISO 20022, a structured data standard that carries far more than the amount and account number. The initiating party field names who initiated a payment on behalf of a client, the relationship a treasury platform has with the business it serves. Purpose codes declare why the payment is being made. Ultimate debtor and creditor fields identify the real parties behind an agent or intermediary. Structured remittance and party data arrive in discrete fields rather than free text. Most institutions are not yet reading these as fraud signals.

From fraud detection to counterparty classification

Read together, those fields turn an anomalous-looking payment into a legible one. The question shifts from “does this AI-initiated payment break the pattern” to “who authorized this agent, under what boundaries, and is this transaction within them.” That is counterparty governance, not fraud detection, and it requires reading the message context the rail already delivers.

The commercial relationship consequence

The institution that classifies AI-initiated commerce cleanly retains and deepens relationships with clients running automated treasury and payables operations. The institution that creates friction on legitimate AI transactions, or misses the ones that warrant scrutiny, loses both the relationship and the risk posture.

For the full leadership framing on agentic commerce governance and the intelligent bank, The Instant Edge publishes every Wednesday at 7:30 AM ET. Subscribe at https://substack.com/@instantpaymentsmaven