STACKWIRE

SaaS and AI

How can businesses let AI agents make payments safely?

Published by Stackwire. Published 10 November 2026.

AI agents are starting to handle business payment tasks, not just research and repetitive admin. But before a company lets an agent pay an invoice, it needs to decide exactly what the agent is allowed to do.

Business professional reviewing financial documents at a desk with a digital tablet
Business payment workflows still involve people checking documents and approving transactions. Photo by Tima Miroshnichenko on Pexels.
Quick answer

Businesses should give AI agents limited permissions, not unrestricted access to company money. Set payment limits, verify suppliers, separate payment preparation from approval, protect credentials and record every transaction.

AI agents are moving into business payments

Imagine a company that receives 200 supplier invoices every month. An employee usually checks each invoice, confirms the supplier, opens a payment portal and enters the payment details.

An AI agent could help with several of these steps. It could find an invoice, compare it with purchase records, prepare a payment and submit it after the required approval.

This is moving beyond theory. On October 20, 2026, Visa, BILL and Billtrust announced a live pilot in which trusted AI agents can initiate and complete buyer-approved invoice payments through supplier portals.

The pilot uses Visa's Trusted Agent Protocol, tokenized payment credentials and controls connected to the buyer's approved payment instructions. Visa says the agent executes eligible payments that have already been approved by the buyer; it does not make the payment decision itself.

Read Visa's announcement.

How should an AI agent payment work?

A business should decide what an agent can do before connecting it to a payment system. For example, it may allow an agent to read invoices from approved suppliers, match them against purchase orders, prepare payment requests and record completed transactions.

That does not mean the agent should also be able to change supplier bank details, increase its own spending limit or approve unusual payments.

  1. STEP 1Set the rulesDefine suppliers, limits and approval requirements.
  2. STEP 2Check the invoiceMatch supplier, amount and invoice details.
  3. STEP 3AuthorizeCheck the request against approved instructions.
  4. STEP 4Pay or reviewProceed if eligible; send exceptions to a person.
A simple example of a controlled invoice-payment workflow. Actual steps depend on the payment platform and company policy.

What if a supplier changes its bank details?

Imagine a company normally pays a supplier $2,000 each month. The next invoice has the same supplier name, but a different bank account number.

The agent should not treat the familiar name as proof that the new details are genuine. It should pause the payment and request verification through a trusted contact method already on file, rather than relying on the email or invoice that requested the change.

The same rule should apply to duplicate invoices, unexpected amounts and requests to pay a new recipient. An instruction inside an email or invoice is information to check, not automatic permission to move money.

Five controls every business should consider

1. Set clear payment limits

Define maximum amounts per transaction and over a set period. A company might allow an agent to prepare any payment but only execute transactions that meet a pre-approved policy.

2. Verify suppliers and payment details

Maintain approved supplier records. A change to a bank account, payment destination or supplier identity should trigger an independent verification process.

3. Separate preparation from approval

An agent that prepares a payment should not automatically have permission to approve it. Keep approval authority separate, especially for large or unusual transactions.

4. Protect payment credentials

Avoid giving agents unrestricted access to bank accounts, card details or employee login credentials. Use appropriately restricted credentials and payment infrastructure. In Visa's announced pilot, tokenized payment credentials are used instead of sharing supplier-portal logins.

5. Keep an audit trail

Record the invoice, payment request, authorization, agent identity, approval and final transaction result. Finance teams should be able to understand what happened without relying only on the agent's explanation.

What are Google's AP2 and Visa's Trusted Agent Protocol?

Companies are developing different approaches to make agent-led payments more trustworthy.

Google's Agent Payments Protocol (AP2) is an open protocol designed to represent payment instructions and authorization. On April 28, 2026, Google announced that it was donating AP2 to the FIDO Alliance and releasing AP2 v0.2, which adds support for autonomous transactions based on instructions authorized in advance, including payments when the user is not present.

Read Google's AP2 announcement.

Visa's Trusted Agent Protocol focuses on helping merchants recognize and authenticate legitimate AI agents. Its developer documentation describes how merchants can validate signed agent requests.

Read Visa's developer documentation.

These initiatives address related but different parts of the problem. AP2 represents authorized payment intent, while Visa's protocol focuses on agent recognition and authentication. Neither replaces a company's own fraud checks, approval procedures or financial controls.

What should businesses do before adopting AI payments?

Businesses do not need to start by allowing an agent to control every payment. A safer starting point is one limited workflow with a small group of verified suppliers.

The PCI Security Standards Council published additional guidance on securing AI in payment environments on October 7, 2026. The guidance discusses securing AI deployments while maintaining appropriate payment security controls.

Read the PCI SSC announcement.

Final thoughts

AI agents could make business payments faster by reducing manual work in accounts payable. But faster payments are not automatically safer payments.

A reliable system needs clear authorization, verified suppliers, restricted credentials and a way to stop unusual transactions.

The goal is not to give an AI agent control over company money. It is to let the agent complete specific, approved tasks while the business keeps control.

Frequently asked questions

Can AI agents make payments without human approval?

Some systems can execute payments under instructions authorized in advance. Whether a payment can proceed without a person depends on the payment system and the business's approval rules.

What is an AI agent payment?

It is a payment initiated or completed by an AI agent acting on behalf of a person or business, within the permissions it has been given.

How can businesses prevent AI payment fraud?

They can verify suppliers, restrict payment permissions, set transaction limits, independently check changed bank details and monitor payment activity.

What is Google's AP2 protocol?

AP2 is an open protocol designed to help represent authorization and trust for payments initiated by AI agents.

Does Visa's Trusted Agent Protocol approve payments?

Its role is to help merchants recognize and authenticate trusted agents. Payment authorization and business spending rules remain separate considerations.

Should small businesses use AI agents for payments?

They can explore limited, low-risk automation, but should first establish supplier verification, approval procedures, secure payment access and transaction monitoring.