Payments In The Age Of AI Agents

TL;DR

Prime Big Deal Days · Oct 6–7Offer from Amazon

Get smart everyday buys delivered free — and shop member deals

• Fast, free delivery on millions of items
• Access to Prime Big Deal Days deals on October 6–7
• Prime Video, Amazon Music and more included

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

Federal Reserve Governor Christopher Waller said AI could help improve cross-border payment screening and routing, while AI agents may eventually shop and pay on a consumer’s behalf. Speaking at Sibos on Sept. 29, he stressed that agentic commerce remains at an early stage and that delegated purchases need stronger safeguards.

Federal Reserve Governor Christopher Waller said artificial intelligence could help payment providers screen cross-border transactions and optimize routing, while AI agents may eventually make purchases and payments for consumers. In a speech at Sibos 2026 in Miami on Sept. 29, he described agentic commerce as an early-stage development and said more trust mechanisms and safeguards are needed before agents can shop autonomously at scale.

Waller said payment companies have used machine learning for years to detect fraud and have adopted large language models for tasks such as payment reconciliation. He described AI agents as systems that can use language models to plan and carry out multistep processes, including transactions. The speech set out potential applications and risks; it did not announce a Federal Reserve policy change or a new payment product.

For cross-border payments, Waller pointed to sanctions screening and anti-money-laundering work. He said research has found that large language models can reduce false-positive alerts, allowing staff to focus on higher-risk cases and potentially reducing delays for legitimate payments. He expects these models to complement faster, traditional anomaly-detection methods, rather than replace them.

Waller also described AI as a possible tool for choosing payment routes and managing foreign-exchange conversion and liquidity. In commerce, he distinguished between agent-assisted shopping, where a person retains control and pays, and agent-delegated shopping, where a person authorizes an agent to buy items and pay within set limits. He said consumer-to-business purchases are expected to be an early use case, with assisted shopping likely to come before more autonomous delegation.

Safeguards for Agent-Made Purchases

If AI agents begin making purchases, payment systems will need to establish how an agent’s authority is granted, limited and checked. A delegated agent could act quickly without a person reviewing each step. That makes spending limits, authorization and dispute handling central practical issues, not just technical details.

AI’s potential benefits also come with security trade-offs. Waller warned that attackers could use AI to develop more sophisticated cyberattacks, while payment networks are highly interconnected and vulnerable to disruption. He said operators must use AI to strengthen defenses while accounting for the imbalance between attackers, who may need one exploitable weakness, and providers, who must defend a broad attack surface.

For consumers, the difference between an agent suggesting a product and an agent completing a transaction is material: the second model transfers some purchasing authority to software. For payment providers and merchants, wider use could require new ways to identify the agent, verify its permissions and resolve unauthorized or mistaken transactions. Waller raised these issues as questions for the industry, rather than describing settled standards.

From Fraud Detection to Agentic Commerce

Waller spoke at Sibos, a payments and financial-services industry conference, building on a speech at the previous year’s event about emerging technologies in global payments. He said the payments sector has long applied machine learning to fraud prevention and was an early adopter of large language models for reconciliation and related operational tasks.

In summer 2026, Waller hosted a Federal Reserve industry roundtable on agentic commerce. He said participants broadly viewed it as being in an early phase, though adoption could reshape commerce and payments if it grows. The models he outlined range from agents that help people search for products to agents authorized to purchase them, potentially using a pre-funded virtual card and instructions such as shopping based on past grocery purchases.

“I expect LLMs to augment, rather than replace, faster, more traditional anomaly detection methods.”

— Christopher Waller, Federal Reserve governor

Adoption and Liability Remain Open

Waller did not give an adoption timetable, estimate how many consumers or merchants are using payment agents, or set out a standard for delegated purchasing. The speech also did not specify who would bear responsibility when an agent exceeds instructions, makes an unintended purchase or is compromised. Those questions will matter as companies develop authorization and consumer-protection rules.

The potential gains in screening and cross-border routing were presented as possibilities, with Waller citing research on false-positive alerts but not giving a figure in the speech. How well AI performs in live payment systems, and how its use affects fraud, costs and transaction times, remains to be established. Waller also warned that AI could strengthen cyber defenses while making attacks more capable.

Industry Work on Payment Guardrails

Waller’s speech did not announce a specific follow-up decision or deadline. The next developments will depend on payment firms, merchants and AI providers testing agent-led purchases and deciding how to limit and verify an agent’s authority. The Federal Reserve roundtable he described offers one venue for industry discussion, but he did not say whether another session is planned.

For cross-border payments, providers may continue exploring AI-assisted screening, routing and liquidity management. Any assessment of progress will need to distinguish research findings and pilot results from performance in routine operations. Waller’s central message was that potential efficiency gains must be weighed against cybersecurity risks and the safeguards required when software is allowed to transact for people.

Key Questions

What did Waller say about AI and payments?

He said AI could support cross-border payment screening and routing, and that AI agents may eventually make purchases for consumers. He also warned about cyber risks and the need for safeguards.

What is the difference between agent-assisted and agent-delegated commerce?

In agent-assisted commerce, an AI helps a shopper find products but the person decides and pays. In agent-delegated commerce, the person authorizes the agent to shop and complete payment within specified limits.

Are AI agents already making payments at scale?

The speech did not establish how widely these systems are used. Waller said market participants regard agentic commerce as being in an early phase.

How could AI affect cross-border payments?

Waller said AI could help reduce false-positive alerts in sanctions and anti-money-laundering screening and could assist with payment routing, currency conversion and liquidity decisions. These were described as potential applications, not guaranteed results.

What safeguards are needed for AI agents that pay?

Waller said delegated commerce needs more extensive trust mechanisms and guardrails because it carries a greater risk of unintended purchases. His speech did not specify a final set of industry standards or rules.

Source: primary