Agentic Payments
What Are Agentic Payments? Definition and How They Work
Definition
Agentic payments are financial transactions authorized, initiated, and executed by an autonomous AI agent acting under delegated authority from a human user. Rather than direct human input at checkout, agentic payments use scoped credentials, such as tokenized payment methods, cryptographic mandates, or signed wallet authorizations, to move funds while maintaining a verifiable chain of user consent.
How it works
Every agentic payment runs through four stages: discovery, authorization, payment, and reconciliation.
In the discovery phase, the agent identifies a merchant, product, or service that meets the user's stated intent. Merchants that expose payment-ready APIs or MCP servers are directly accessible; others require the agent to simulate browser interactions, which most protocol frameworks discourage.
Authorization is where agentic payments depart most significantly from standard card-not-present flows. The agent must prove two things simultaneously: that the buyer's payment provider scoped a credential to this merchant and amount (handled by a Shared Payment Token or Agentic Token issued by the card network), and that the user explicitly consented to this specific purchase (handled by a cryptographically signed mandate such as a Google AP2 Cart Mandate or an OpenAI ACP purchase token). Together these create a non-repudiable consent record.
Payment execution on card rails routes the scoped token through the merchant's processor to the card network and issuing bank, settling through normal interchange, no changes to downstream clearing infrastructure. On stablecoin rails, a wallet-signed transfer authorization (EIP-3009 or similar) is submitted to the chain, with settlement finality at block confirmation.
Reconciliation attaches agent identity metadata, mandate hashes, and user consent signals to the transaction record. Card networks are building agent-specific data fields into their authorization messages so that issuers receive visibility into whether an AI agent was involved, enabling more accurate fraud scoring and dispute attribution.
Why it matters
Agentic payments force a redesign of authentication assumptions across the entire payment stack. Traditional SCA models, 3DS, biometric confirmation, one-time passwords, are predicated on a human being present at the moment of payment. When an agent pays on behalf of a sleeping user for a standing grocery order, none of those mechanisms apply in the usual way. Issuers and networks are building new trust models, agent registries, mandate verification, spend-scope enforcement, to fill the gap.
Fraud risk profiles shift materially. Stolen credentials used by an agent look different from stolen credentials used by a human; conversely, legitimate agent purchases can trigger false positives in models trained on human behaviour. Risk engines need agent-aware signal sets.
Liability allocation is unsettled in the near term. When an agent completes a purchase the user disputes, the chain of consent evidence, mandate signatures, token scope records, agent registry status, determines who bears the chargeback. Merchants, networks, and processors are actively building the data infrastructure to support this.
With PXP
PXP supports merchants and partners across the payments value chain. To talk through agentic payments as part of your payment strategy, get in touch with our team.
Frequently asked questions
Are agentic payments the same as merchant-initiated transactions (MITs)?
There is overlap but they are not the same. Merchant-initiated transactions are payments triggered by the merchant under a pre-agreed stored credential framework, recurring billing, instalment payments, where the cardholder is not present. Agentic payments are initiated by a third-party AI agent acting for the cardholder, not the merchant. New protocol standards (ACP, AP2) define the credential and consent model specifically for the agent-initiated case.
How do card networks handle agent identity?
Visa's Trusted Agent Protocol (TAP) and Mastercard's Agent Pay framework both include agent registry systems. Agents register with the network, receive a verifiable identity credential, and sign their payment requests with a network-issued key. Merchants verify the signature against the registry before accepting the transaction. This separates agent identity verification from cardholder identity verification.
What happens when a user disputes an agentic payment?
Under card-rail protocols, the merchant's dispute evidence is the signed mandate, a tamper-proof record of the user's explicit consent to the specific transaction. If the mandate signature verifies against the user's registered key, the merchant has strong evidence that the user authorized the purchase. Dispute outcomes depend on how well the mandate evidence was captured and stored at transaction time.
Revolutionize your business with PXP
Take complete control of your commerce and payments with one platform.
Get Started