EMVCo Framework Targets Consumer Intent in Agentic Payments

Artificial intelligence agents add information to a card payment that merchants, issuers and networks don’t routinely exchange today: what the consumer authorized the agent to do and whether the transaction falls within that authority.

EMVCo is developing a technical framework for carrying that information through card payments.

The standards body closed public comments Sept. 30 on its Agentic Payments — Framework for Specifications. EMVCo’s Sept. 1 request for comment described a “common foundation” for card-based agentic payments that could support consumer intent, agent identification and signals that an agent was involved in a transaction.

EMVCo develops technical specifications used across card payments. Its framework could standardize how information about an agent and its delegated authority moves through the payment system, while decisions about approving transactions and applying commercial rules remain with networks, issuers and other participants.

“It does not define the business rules, in other words, how a product is used,” Oliver Manahan, director of engagement and operations for EMVCo, told PYMNTS.

Why Payments Need This Data

A conventional card payment starts with a consumer deciding to transact. An agentic payment can start with instructions the consumer gave software hours, days or potentially weeks earlier.

A consumer could tell an agent to buy a specific product, spend no more than a certain amount, make recurring purchases or operate within a budget across several transactions. The agent may possess a valid payment credential even when a particular purchase falls outside those instructions.

Authentication alone doesn’t convey all that information. A valid credential can establish that the payment method may be used, while the agent’s authority can carry separate limits on what it was permitted to buy and under what conditions.

EMVCo’s framework proposes “Intent Services” to make information about that authority available across the payment flow. Its request for comment described a shared layer through which participants could register, reference, retrieve and manage consumer-authorized intent before, during and after a transaction. Consumer intent may also need to remain available beyond the initial transaction. Manahan said entities in the payment ecosystem may need to retrieve historical information for dispute resolution.

“EMVCo does recognize that intent must be continuously managed from a lifecycle perspective, as entities within the ecosystem may need to retrieve historical information for dispute resolution,” Manahan said.

For purposes of the framework, a payment credential is selected and associated with authorized intent before the agentic payment begins.

Identifying the Agent Behind the Payment

EMVCo said in the request for comment that it may also consider know your agent (KYA) capabilities that could let participants consistently identify an agent and associate it with stable, interoperable metadata.

The metadata could support payment processing, an accountability chain, risk evaluation, lifecycle coordination, auditability and operational transparency, Manahan said. KYA could eventually be addressed through future EMVCo specifications, profiles, operating frameworks or ecosystem programs.

Agentic transaction indicators could provide a separate signal that an agent was involved in a transaction.

The indicators could support “risk assessment, fraud prevention, dispute resolution, operational transparency, auditability, reporting and ecosystem accountability,” Manahan said.

They could also help participants interpret payments in which consumer authentication had been delegated to an agent operating within previously established limits, he said.

The amount of authority consumers are prepared to delegate varies depending on the task.

The PYMNTS Intelligence report “Global Digital Shopping Index: The AI-Powered Shopper Has Arrived” found in June that 56% of consumers would allow an AI agent to search and compare products. Meanwhile, 37% would allow an agent to authorize payments, 36% would allow predictive automatic buying and 35% would give an agent access to saved payment methods.

AI already has influence before checkout. PYMNTS Intelligence’s “The 50 Million Consumer Migration: The Data Behind Retail’s Shift Toward AI Discovery” found in August that 19% of consumers in the United States begin retail product research with AI. Among consumers using AI for retail research, 43% found a better price, 27% changed brands and 26% changed products.

Consumers could give agents different levels of authority. One might use an agent only to compare products. Another could approve each purchase before checkout. A third could authorize transactions within limits on price, merchant, product or time.

Payment participants would need to distinguish among those permissions when an agent presents a credential.

Comments Move Into Review

The public consultation closed without a regulatory-style docket showing what each respondent submitted. PYMNTS asked EMVCo about the roadmap ahead.

“As the public review period has only just ended, the Agentic Payments Task Force is now working to collate the feedback and will review next steps,” Manahan said.

The framework will help EMVCo determine whether new specifications are needed and whether existing technologies require changes. Manahan identified EMV 3-D Secure, EMV Payment Tokenization, EMV Secure Remote Commerce and the EMV Digital Payment Credential as possible areas for future enhancements. EMVCo may also consider KYA capabilities and agentic transaction indicators.

The resulting specifications could determine what information about an agent and its delegated authority accompanies a card payment and what merchants, issuers and networks can see when handling it.

For all PYMNTS AI coverage, subscribe to the daily AI newsletter.