Regulators, researchers and payments leaders are all saying the same thing: AI agents are about to move money, and fraud models tuned to human behavior may not recognize what they see.

For decades, every payment has had a person at the end of it: someone who clicked, tapped, signed or swiped. That assumption is starting to break.

At Sibos on Sept. 29, Federal Reserve Governor Christopher Waller gave a speech titled “Payments in the Age of AI Agents.” He drew a line between two models. In the agent-assisted model, a buyer uses AI mostly for search and discovery, but “they make the decisions, and they handle the payment.” In the agent-delegated model, “a buyer grants authority to an AI agent to shop and make payments on their behalf,” and the agent operates autonomously.

He named the challenges plainly: “authentication, liability, and fraud.” And on fraud, he was direct: “Existing fraud detection and prevention systems, calibrated to human behavior, may not translate well to agents.”

That’s a Fed governor telling the industry that the controls most institutions run today were built for a different kind of customer.

From click-to-pay to decide-to-pay

The day before Waller spoke, the Filene Research Institute published The New Money Movement Landscape (Report #680). It gave the shift a name. Agentic payments, Filene wrote, may move commerce from “click-to-pay” to “decide-to-pay.” Agents could act for a member within defined permissions, without the member approving every purchase.

Filene is measured about timing. The technology is early, and credit unions “do not need to treat agentic payments as an immediate product-launch imperative.” But the report is clear that the hard questions aren’t about the payment button. They’re about identity, authorization, authentication, fraud, transaction controls and accountability. One of its open questions for credit unions goes straight to the point: can existing monitoring tell legitimate agent activity apart from unauthorized or malicious transactions?

Filene’s near-term advice is readiness, not rapid deployment.

The rails are already fast. The members already expect it.

Agents won’t arrive on slow rails. On Oct. 5, Credit Union Times reported that credit unions must offer instant payments to keep up with Gen Z. The pressure to move money in seconds, around the clock, is coming from members, not just technology vendors.

Put those together. Waller warned that small machine-to-machine payments “could reach a much larger scale if shopping agents are working around the clock at machine speed,” and that agentic transactions “could materially change the frequency and timing of payments.” Instant rails mean those payments settle fast. A fraud model that expects a member to pay bills on Friday afternoon will see a very different pattern when software is buying on the member’s behalf at 3 a.m.

His conclusion: “fraud models and rules will need to be recalibrated to account for agent payment patterns.”

The blind spot is upstream

On Oct. 7, FinTech Global covered a payments panel (summarized by Vivox AI) with participants from Bank of London, Mastercard, PXP, CGI and formerly Starling Bank. The panel’s takeaway was uncomfortable: controls built around human behavior may be looking in the wrong place.

A payment can pass every spending limit and still “stem from a manipulated decision made long before checkout.” The examples were concrete: an agent paying a fake merchant with the customer’s authority, or attacks planted in fake merchant sites, fabricated catalogs and malicious instructions hidden in the content agents read. And once attackers find a weakness, automated agents could exploit it repeatedly and quickly.

Liability is still unsettled. Waller asked it directly (“Who is on the hook if an agent makes the wrong purchase?”), and the panelists disagreed on how responsibility should be shared. Their practical advice: test now whether fraud controls can separate legitimate automated activity from misuse, instead of waiting for a finished model.

Real-time detection isn’t the same as real-time protection

Most fraud programs today can spot something unusual. The problem is what happens next.

An alert lands in a queue. An analyst reviews it in the morning. By then, an instant payment has settled, and an agent that was manipulated once may have done it a dozen more times. When money moves at machine speed, a human review step measured in hours is a record of the loss, not a defense against it.

The old model: Detect → Alert → Wait for a human.

The model decide-to-pay demands: Detect → Assess → Act.

Where RembrandtAi® fits

RembrandtAi® was built for exactly this gap. It ingests events across every channel (card, ACH, remote deposit, online banking, ATM) and scores each one in real time using multi-stage AI, including neural networks, statistical analysis and machine learning scoring. When risk crosses the thresholds your team sets, it acts directly in your banking core: member alerts, card restrictions, transaction blocks, account freezes and access suspension, 24/7/365, with no human in the loop required.

That matters when the payer might be software. RembrandtAi® doesn’t need to know whether a member, a fraudster or a manipulated agent started the session. It judges what’s actually happening in the account, as it happens, and responds before the money is gone.

Your people stay in charge. They set the thresholds, work the exceptions and own the member relationship. RembrandtAi® makes sure the response doesn’t wait for business hours.

It has a track record. Launch Credit Union adopted RembrandtAi® at the end of 2020 and saw a 97% reduction in fraudulent transactions, saving $1.8 million in 2021, American Banker reported. As CFO Kevin LeBeau put it: “If you’re uploading three hours later, that money is already gone.”

The takeaway

In one week, a Fed governor, the credit union movement’s research institute and a panel of payments practitioners all reached the same place: agents are coming to payments, and fraud controls calibrated to people may not see the risk. Filene is right that credit unions don’t need to launch agentic payments tomorrow. But readiness means having a defense that runs at the same speed as the payments it protects.

AI must move at the speed of payments.

See how RembrandtAi® acts in real time, in your core. Request a demo.

Sources

1. Christopher J. Waller, “Payments in the Age of AI Agents,” speech at Sibos 2026, Miami, Board of Governors of the Federal Reserve System, Sept. 29, 2026. https://www.federalreserve.gov/newsevents/speech/files/waller20260929a.pdf
2. McKaye Black, “The New Money Movement Landscape,” Report #680, Filene Research Institute, Sept. 28, 2026. https://www.filene.org/reports/the-new-money-movement-landscape
3. Joyce Moed, “Credit Unions Must Offer Instant Payments to Keep Up With Gen Z: Report,” Credit Union Times, Oct. 5, 2026. https://www.cutimes.com/2026/10/05/credit-unions-must-offer-instant-payments-to-keep-up-with-gen-z-report
4. FinTech Global, “Agentic payments expose blind spots in fraud controls,” FinTech Global, Oct. 7, 2026. https://fintech.global/2026/10/07/agentic-payments-expose-blind-spots-in-fraud-controls/
5. “Florida credit union overhauls its fraud prevention with big assist from AI,” American Banker, May 6, 2022. https://www.americanbanker.com/news/ai-revolutionized-how-this-credit-union-prevents-fraud