Who Gets Paid When AI Shops? Banks and Tech Giants Clash Over the Future of Payments

What if the next major threat to banks' payment revenue isn't another bank, but an AI agent? As AI-powered shopping evolves from product recommendations to autonomous purchasing, technology companies are beginning to influence who consumers buy from and how they pay. The stakes are substantial: Agentic Commerce could reshape the economics of digital payments, putting interchange fees, payment routing and customer relationships under pressure. In September 2026, six major banks, including Bank of America, Capital One, ING and NatWest, jointly called for safeguards covering transparency, fraud, privacy and consumer choice in AI-led shopping. Meanwhile, Mastercard is expanding its agentic payment infrastructure, highlighting how rapidly the industry is adapting. The emerging battle between banks and AI companies is therefore about more than checkout convenience. It is about who controls AI Payments, who captures transaction revenue and who remains accountable when an AI agent spends a consumer's money. For financial institutions, this is becoming a critical strategic issue as agentic commerce moves from experimentation towards real-world transactions, and a significant topic for discussion at Finov AI Summit Europe.

What happens when consumers no longer choose how they pay, because their AI agents make that decision for them?

That question is becoming increasingly important as AI companies move from product recommendations to autonomous shopping and payments. An AI agent that can compare products, select a merchant, choose a payment method and complete a purchase could fundamentally change the economics of digital commerce. For banks, the implications extend well beyond the loss of a traditional checkout interaction. They could affect interchange revenue, payment relationships, customer data and control over the transaction itself.

The friction is already visible. In September 2026, six major banks, including Bank of America, Capital One, ING and NatWest, jointly called for principles governing AI-powered shopping. Their concerns include fraud, privacy, transparency, consumer choice and accountability. Meanwhile, technology companies are developing AI agents that could become the primary interface between consumers and merchants.

The commercial stakes are significant. Banks have historically participated in transactions through card issuing, acquiring and payment infrastructure. Agentic commerce introduces another powerful intermediary: the AI platform that decides what consumers buy and potentially how they pay.

The central question is no longer whether AI can complete a purchase. It is who controls the purchasing decision, who bears the risk and who captures the revenue when an AI agent completes the transaction.

For the financial services industry, this emerging contest could redefine the economics of digital payments.

Traditional e-commerce places the consumer at the centre of the purchasing journey. A shopper visits a website, compares products, selects a payment method and authorizes the transaction. Merchants, banks, card networks and payment processors each have established roles.

Agentic commerce changes that sequence. A consumer can delegate some or all of the shopping process to an AI agent. Depending on its permissions, the agent may search for a product, compare offers, select a merchant and initiate payment.

This change introduces a new layer of commercial influence. An AI platform may control product discovery and influence the merchant and payment method selected, even when a bank continues to issue the card and process the payment through an established network. That distinction matters because control over the customer interface can translate into influence over transaction economics.

For example, an AI agent instructed to find the lowest total price could compare not only product prices but also shipping charges, loyalty rewards, payment discounts and available payment methods. If the agent is permitted to choose a payment method, its selection could affect the revenue earned by the financial institutions involved.

The result is a potential redistribution of influence across the payment value chain. Banks may continue to process transactions while losing some control over the decisions that generate them.

The immediate commercial concern is interchange, but the issue is broader than a single fee. Interchange is generally paid by an acquiring institution to the issuing bank as part of a card transaction. It is distinct from network assessments, processing charges and the merchant service charge. Although the amount varies by market, card type, transaction and applicable regulation, interchange is an important part of the economics of card issuing.

An AI agent that chooses a different payment method could change this economic arrangement.

The key distinction is between transaction volume and revenue per transaction. An increase in agent-led purchases does not necessarily translate into higher bank revenue if the payment method or commercial arrangements change.

Equally, reduced interchange per transaction does not necessarily mean that banks will lose overall revenue if AI agents generate substantial additional payment activity.

This makes the economic impact uncertain. The outcome will depend on adoption, payment preferences, merchant acceptance, regulatory constraints and the extent to which AI platforms influence routing. For banks, the strategic concern is that payment selection could increasingly take place outside their own digital environments.

22 September 2026, six institutions jointly published Building Trust in Agentic Commerce, outlining principles for the emerging AI-led shopping ecosystem. The report is an important development because it takes the debate beyond individual banks and places it within a wider industry discussion. The banks identify five principles:

The report identifies several risks that become more significant as agents receive greater autonomy. An agent could purchase the wrong product, spend more than intended, expose sensitive credentials or direct a consumer toward a payment option that offers weaker protections. It also raises questions about incentives. An agent may prioritize a product or payment method because of a commercial arrangement rather than because it best meets the consumer's requirements.

These concerns have direct economic implications. Transparency about payment selection, for example, could reveal how commercial incentives influence an agent's recommendations. Interoperability could make it easier for consumers to move between competing services rather than remain dependent on one platform.

However, the report should not be interpreted as proof that banks are lobbying solely to protect interchange. Its published principles explicitly address consumer safety, privacy and choice. Commercial incentives are relevant to the debate, but they are not a substitute for examining the evidence behind the stated concerns.

The banking debate is unfolding as major technology companies extend AI from conversation into commerce.

The competition is not limited to who builds the most capable AI model. It is also about who owns the infrastructure connecting product discovery, merchant systems, payment credentials and transaction authorization.

Recent disputes illustrate the challenge. Amazon asked Meta's Muse shopping assistant to stop using its website, citing concerns about how the agent accessed its services and handled customer information. Meta also disclosed that it had fixed a vulnerability affecting Muse.

These incidents demonstrate that agents do not automatically have permission to interact with every merchant platform. Access, authentication and commercial agreements will influence which agents can complete transactions and under what conditions.

For AI companies, the opportunity is to make purchasing a natural extension of conversational experiences. For merchants, the challenge is to retain control over their products and customer relationships. For banks, it is to remain a meaningful part of the payment journey even when another company controls the interface.

The industry needs to distinguish between the growth of AI-assisted shopping and the adoption of fully autonomous payments. A particularly useful early indicator comes from British retailer John Lewis. According to Reuters, searches originating from AI agents rose from 0.3% to 2.5% over a year, as reported in September 2026.

This is an early indicator of changing shopping behavior, not proof that consumers are already delegating large volumes of payments to AI agents. Searches can originate from AI systems without the agent ever completing a transaction.

Consumer trust presents another important adoption constraint. Visa's September 2026 Trust Index reported that only 23% of surveyed US consumers trusted generative AI to handle payment transactions on their behalf. In the same research, 61% said they would trust Visa to handle agentic transactions. These are different measures of trust in different entities. They nevertheless illustrate a significant distinction: consumers may be comfortable using AI to discover products while remaining reluctant to authorize it to handle their money.

This gap matters to the banking industry. AI agents may gain acceptance as shopping assistants well before consumers are willing to delegate payment authority. Payment networks and banks could therefore remain central to building the trust necessary for adoption.

30 September 2026, Mastercard's latest announcement provides an important development in the evolving debate. The company expanded its Agent Pay programme with new trust and intelligence services designed to help banks and merchants understand AI-initiated transactions.

The services combine agent identity, consumer intent, behavioural signals and fraud-related insights. Mastercard also introduced a probability score intended to help identify transactions likely to have been initiated by an AI agent, initially rolling out for testing in the United States.

This is commercially significant because banks need to distinguish legitimate agent-led transactions from suspicious activity. An unusual payment pattern might reflect a consumer authorizing an AI agent to book an entire trip, rather than a fraudster making multiple purchases.

A system that provides additional context could help issuers make more informed authorization decisions. It may also reduce unnecessary transaction declines when legitimate agent-led purchasing becomes more common. Mastercard's move illustrates that payment networks are developing infrastructure to participate in the emerging ecosystem rather than simply waiting for AI platforms to define it.

However, these services are still developing. A probability score or intelligence signal does not guarantee that a transaction is legitimate. Its effectiveness will depend on the quality of underlying data, how banks use the signals and how well the system handles new patterns of agent behaviour.

The announcement also demonstrates that agentic commerce is becoming an infrastructure competition. Recognizing the agent, understanding its instructions and evaluating the resulting transaction may become essential capabilities for payment providers.

The economic impact of agentic commerce will not be uniform across the financial ecosystem. Different participants have different sources of revenue and different forms of exposure.

The most important distinction is between owning the customer interface and owning the financial infrastructure. AI companies may gain control over the interface through which consumers discover products and initiate purchases. But payment networks and banks may retain important roles in authentication, account funding, authorization and dispute resolution.

The resulting market may not have a single winner. It could create a more complex ecosystem in which different participants control different stages of the transaction.

For example, an AI assistant could choose the product and merchant, a bank could issue the payment credential, a card network could authenticate the transaction, and the merchant's acquiring institution could process the payment.

The commercial question is how much value each participant can capture as these responsibilities are redistributed.

As AI agents gain authority to initiate payments, the industry will need more than technical integration. It will need clear rules for authorization, accountability and consumer protection.

These questions have practical consequences for financial institutions. A bank authorizing an agent-led transaction may need to understand more than the card credentials and transaction amount. It may also need contextual information about the agent, its authority and the consumer's instructions.

Similarly, a merchant may need to know whether an agent has the authority to complete a purchase, return an item or change an order. The six-bank principles provide an initial framework for this discussion, while payment network initiatives are beginning to address the technical infrastructure. The regulatory and commercial details will continue to evolve as the market develops.

Financial institutions should avoid treating agentic commerce as an adoption race. The important question is whether new payment experiences create measurable business value while preserving appropriate protections. Banks can begin by establishing a baseline across six areas.

These metrics should be evaluated together. A higher approval rate may appear positive, for example, but could be accompanied by an increase in fraud losses. Similarly, transaction growth may not translate into stronger profitability if the payment mix changes.

Banks should also distinguish agent-originated traffic from agent-completed transactions. Product searches, recommendations, checkout initiation and confirmed payments represent different stages of adoption.

The same discipline should apply to revenue projections. A forecast based on potential agent-led transaction volume should not be presented as realized interchange revenue or confirmed market growth. For payment leaders, a useful first step is to establish a controlled pilot with clear authorization boundaries, measurable outcomes and a process for reviewing exceptions.

Agentic commerce raises questions that extend well beyond payment processing. It touches financial infrastructure, consumer protection, digital identity, data governance and the commercial relationships between banks and technology companies. For professionals following Finov AI Summit Europe, the development is particularly relevant to three areas of discussion.

First, the future of payment economics. Financial institutions need to understand whether AI agents will generate additional card transactions, redirect spending toward alternative payment rails or create new commercial intermediaries.

Second, the future of customer relationships. As AI agents become more prominent in shopping, banks will need to consider how their products remain accessible and relevant when consumers interact with financial services through third-party interfaces.

Third, the future of trusted AI. Secure agent identity, transparent authorization, reliable audit trails and clearly assigned liability will be essential to the development of a sustainable agentic commerce ecosystem.

These are not merely technology questions. They are strategic decisions about the future structure of digital financial services.

Banks, payment networks and AI companies will need to collaborate on standards even as they compete for influence over the customer journey. The companies that develop the necessary infrastructure may play an important role in determining how agent-led payments evolve.

The growing friction between banks and AI companies reflects a deeper change in digital commerce. AI agents are moving from recommending products toward executing purchases, creating new questions about payment selection, transaction security and the distribution of commercial value.

For banks, the potential pressure on interchange revenue is only one part of the challenge. The bigger strategic issue is whether they can retain influence over payment decisions as AI platforms become increasingly important shopping interfaces.

For AI companies, the opportunity is to make purchasing more convenient. But doing so at scale will require consumer trust, reliable authorization and cooperation with the financial infrastructure that processes payments.

Recent developments from the six-bank consortium and Mastercard demonstrate that the industry is already working on these challenges. However, neither new standards nor new technology alone will resolve the underlying commercial tensions.

The future of agentic commerce will depend on whether the industry can separate control of the shopping experience from control of the payment infrastructure, while ensuring that consumers remain protected.

The next major shift in digital payments may not be about replacing cards or banks. It may be about redefining who gets to decide when, where and how those cards are used.

Join the conversation at Finov AI Summit Europe to explore how AI Payments, autonomous agents, financial infrastructure, trust, and regulation are converging to redefine the future of financial services.

Secure Your Tickets and Be Part of the Conversation.
Get Your Tickets Now