menu_open Columnists
We use cookies to provide some features and experiences in QOSHE

More information  .  Close

Can your applications handle transactions initiated by AI agents?

24 0
06.11.2026

Suppose your company sells software subscriptions. A customer asks an AI agent to renew a plan, provided the annual cost remains below $1,000. The agent sends a renewal request to your application. Its access token is valid, the payment method works, and your system processes a $1,200 charge.

From your application’s perspective, the renewal succeeded. But the customer never approved that amount. The spending limit existed in the instructions given to the agent, and your application never received it as an enforceable condition.

So, if your business plans to accept agent-initiated transactions, how will your application verify what the customer has authorized the agent to do? And if that authorization is missing, will your system reject the request or ask for additional approval?

Let’s discuss this aspect in more detail.

What an Access Token Can and Can’t Authorize

When someone uses your application directly, the process is familiar. Your application presents options, collects information, requests confirmation, and applies business rules.

An AI agent adds another decision-making layer. It interprets a person’s instructions and chooses how to carry them out. Your application may receive an authenticated API request without knowing the conditions attached to the customer's original instructions.

An access token granting subscription-management permissions may establish that an agent can request a renewal. That permission alone doesn’t establish the customer’s spending limit, approved plan, or confirmation requirements.

Existing standards and emerging agent-payment protocols provide ways to communicate more detailed permissions – for instance:

OAuth 2.0 Rich Authorization Requests (RFC 9396) supports fine-grained authorization information, including transaction amounts and recipients

Google’s Agent Payments Protocol (AP2) introduces verifiable mandates to help establish what a customer approved for an agent-initiated payment

Your system must verify that the authorization presented covers the requested action. If the necessary proof is unavailable, the workflow........

© Forbes India