AI Agents Enter the Payments Economy, Forcing Banks to Rethink Trust, Security and Accountability

Artificial intelligence is moving into a new phase of commercial activity. Instead of simply recommending products, answering questions or helping customers compare prices, AI agents are increasingly being designed to take actions on behalf of users—including selecting products, initiating transactions and completing purchases.

That shift could fundamentally change the relationship between consumers, banks, payment networks and online businesses.

The emerging concept of “agentic commerce” means an AI system can potentially understand a user’s instructions, search across services, make decisions within predefined rules and complete a transaction without requiring the customer to manually approve every individual step.

For financial institutions, this creates a problem that traditional banking systems were not designed to solve: How does a bank establish trust when the entity initiating a transaction is software acting on behalf of a human?

The issue is now becoming a major topic across the global payments industry. Financial technology leaders have begun discussing the need for a new “Know Your Agent” approach alongside existing customer-identification and fraud-control systems.

From AI Assistant to Autonomous Buyer

The most important change is the movement from AI that advises users to AI that acts for them.

A conventional shopping assistant might tell a customer that a particular laptop is available at a lower price.

An agentic system could potentially go further.

A user could instruct the agent to find a laptop meeting certain specifications below a specific price. The agent could search different retailers, compare products, evaluate delivery options and, if permitted, complete the purchase.

The human remains the ultimate customer, but the software becomes the active participant in the transaction.

That distinction has major consequences for financial institutions.

Banks have spent decades developing systems around a relatively straightforward assumption: a person or business initiates a payment and the institution verifies the transaction according to established rules.

AI agents complicate that model.

Why “Know Your Agent” Is Becoming Important

Traditional financial compliance relies heavily on identifying customers and understanding their transactions.

With autonomous AI, banks may also need to determine:

• Which AI agent initiated the transaction?
• Who authorized that agent?
• What permissions were granted?
• What spending limits apply?
• Which company operates the agent?
• Can the agent be authenticated?
• Was the transaction consistent with the user’s instructions?
• Who is responsible if the agent makes an unauthorized purchase?

These questions are becoming more important as AI systems gain the ability to interact directly with payment infrastructure.

At the Fortune Leaders Forum in Macau, Ant Digital Technologies President Zhuoqun Bian highlighted the emerging compliance challenge and the need to identify and understand AI agents involved in financial transactions.

The underlying issue is simple but significant: a bank may know the customer, but it also needs confidence that the software acting on that customer’s behalf is legitimate and operating within authorized boundaries.

The Payments Industry Is Already Building New Standards

This is not merely a theoretical discussion.

Visa, Mastercard and Ant International have already begun collaborating on a framework designed to help payment networks, digital wallets, AI platforms and online marketplaces identify and verify AI agents across different systems.

The initiative builds on existing technologies developed by the companies, including Visa’s Trusted Agent Protocol, Mastercard’s Verifiable Intent and Ant International’s Agentic Mobile Protocol.

The significance of such cooperation is that AI agents may eventually operate across multiple platforms.

An agent could discover a product on one website, interact with a merchant on another platform and use a payment network operated by a third company.

Without common standards, every participant could require a separate authentication mechanism.

That would make agentic commerce expensive and difficult to scale.

A common framework could potentially reduce that complexity.

The Biggest Question: Who Is Responsible?

One of the most difficult issues is liability.

Imagine an AI agent is authorized to spend up to a certain amount each week.

The system mistakenly purchases an expensive product outside the user’s actual intention.

Who should bear the financial loss?

Possible parties could include the consumer, the AI provider, the merchant, the payment company or the financial institution.

Traditional payment systems already have rules covering unauthorized transactions and fraud.

AI agents create additional complexity because a transaction may technically be authorized by software while still being inconsistent with what the customer intended.

This makes the concept of verifiable intent particularly important.

Payment systems may increasingly need evidence showing what the user permitted the AI agent to do and whether the final transaction stayed within those permissions.

Spending Limits Could Become a Core Safety Mechanism

One possible approach is to give AI agents carefully defined financial permissions.

Instead of providing unrestricted access to a bank account or credit card, consumers could establish rules such as:

• maximum amount per transaction;
• maximum daily spending;
• approved merchants;
• approved product categories;
• geographic restrictions;
• recurring-payment permissions;
• transactions requiring additional confirmation.

This would make an AI agent resemble a digital financial assistant operating inside a controlled environment.

The approach could allow automation while reducing the risk of unlimited financial access.

However, limits alone would not solve every problem.

A sophisticated fraudster could potentially attempt to manipulate an agent, exploit vulnerabilities in an AI system or trick it into interpreting an instruction incorrectly.

That means authentication, monitoring and transaction-level risk analysis would remain necessary.

India Is Preparing for the Same Transformation

The issue is particularly relevant to India because of the enormous scale of the country’s digital payments infrastructure.

The National Payments Corporation of India has been working on a proposed framework for AI agents interacting with UPI.

Reports indicate that the proposed system is intended to identify and authorize AI agents and could initially focus on relatively small, frequent payments.

The development shows how the transition toward agentic payments is becoming a global phenomenon rather than a technology trend limited to Silicon Valley or Western financial markets.

India’s approach also highlights an important principle: introducing autonomous payments at scale requires governance mechanisms to develop alongside the technology.

The proposed UPI framework has faced additional scrutiny around safeguards and regulatory approval, illustrating how financial authorities are approaching the technology cautiously.

Why Small Payments May Come First

The first widespread use of autonomous payments may not involve large purchases.

Instead, AI agents could initially handle repetitive, low-value transactions.

Consider everyday purchases such as groceries, household supplies, subscriptions or routine services.

A customer might tell an AI agent to keep certain household products stocked and remain within a weekly budget.

The agent could monitor prices and purchase items when necessary.

This type of transaction has several advantages for early adoption.

The financial risk per transaction is relatively limited, the user’s preferences can be clearly defined and the value of automation is easy to understand.

More complicated applications—such as financial products, large purchases or investment decisions—could require significantly stronger controls.

The Security Challenge Could Be Larger Than the Convenience Benefit

AI-powered commerce offers obvious convenience, but security will determine how quickly consumers accept it.

Traditional online fraud often involves stolen passwords, compromised cards or deceptive messages.

Agentic commerce introduces another potential attack surface: manipulation of the AI agent itself.

If an attacker can influence an agent’s instructions, environment or data, the system could potentially make a legitimate payment for the wrong reason.

This creates a different category of financial-security problem.

The transaction may look technically valid to a payment network even though the AI was manipulated before it reached the payment stage.

The International Monetary Fund has also highlighted concerns surrounding the use of agentic AI in payment processes, including authorization, execution, settlement and the nondeterministic nature of some AI systems.

That is why financial institutions are likely to require more than simple identity verification.

AI Agents Could Change the Economics of Online Shopping

The rise of autonomous shopping could also alter competition among retailers.

Today, companies invest heavily in advertising, search-engine optimization, product placement and brand visibility.

But an AI agent may not shop in the same way a human does.

Instead of browsing dozens of advertisements, it could compare prices, specifications, delivery times, ratings and return policies automatically.

This could shift competitive power from websites designed to attract human attention toward systems optimized for machine-readable product information.

Retailers may therefore have to ensure that their product data is accurate, structured and accessible to AI systems.

The future online marketplace could become less about persuading a person to click and more about convincing an AI agent that a product satisfies a specific set of requirements.

Banks May Become Infrastructure Providers for AI Commerce

The banking industry’s role could also evolve.

Instead of interacting directly with customers for every transaction, banks may increasingly provide the financial infrastructure that allows authenticated AI agents to operate within controlled permissions.

That could create new banking services built around:

• agent identity;
• transaction authorization;
• spending controls;
• fraud detection;
• real-time monitoring;
• permission management;
• automated dispute handling;
• digital identity verification.

Banks that successfully integrate these systems could become important infrastructure providers for the emerging agent economy.

At the same time, banks will have to balance automation with regulatory obligations.

Trust Could Become the Most Valuable Currency

The central challenge for agentic commerce may ultimately be trust.

Consumers need to believe that their AI agent will follow their instructions.

Merchants need confidence that transactions initiated by agents are legitimate.

Payment networks need to know that agents have appropriate authorization.

Banks need to determine whether transactions fall within acceptable risk parameters.

Regulators need mechanisms to establish accountability when something goes wrong.

If any one part of this chain fails, consumer adoption could slow.

Technology alone cannot solve that problem.

Clear rules, transparent permissions and reliable authentication will be equally important.

The Human May Approve the Rules, Not Every Transaction

One of the most significant changes could be psychological.

Today, consumers often think of authorization as a transaction-by-transaction activity.

Agentic commerce moves toward a different model.

A user may authorize a set of rules rather than each individual purchase.

For example, instead of approving ten separate grocery transactions, a customer could authorize an AI agent to spend up to a fixed amount on groceries during a particular period.

That could make digital commerce dramatically more automated.

But it also changes the meaning of consent.

The question becomes not simply “Did the user approve this transaction?” but “Did the user’s previously granted instructions reasonably authorize this transaction?”

That distinction could become central to future financial regulation.

What Happens When AI Agents Start Competing With Each Other?

Another potentially important development is the emergence of machine-to-machine commerce.

Imagine thousands of AI agents representing consumers, retailers and businesses negotiating simultaneously.

One agent searches for the cheapest product.

Another agent represents a retailer and automatically adjusts offers.

A third agent manages delivery.

Payment could occur automatically once predefined conditions are satisfied.

Such an economy could operate much faster than traditional human commerce.

However, it would also require highly reliable digital identity and authorization systems.

Without those systems, large-scale autonomous transactions could create new systemic risks.

The Regulatory Challenge Will Expand

Regulators are likely to face questions that did not exist when digital banking was designed.

Rules may eventually need to address:

• AI-agent identity;
• consumer consent;
• transaction limits;
• liability for autonomous decisions;
• data protection;
• fraud;
• dispute resolution;
• cross-border transactions;
• transparency of AI decision-making.

Different countries could develop different frameworks, creating another challenge for global payment companies.

That is one reason interoperability standards are becoming important.

An AI agent operating in India may eventually purchase a product from a merchant in another country using an international payment network.

The technology therefore needs rules that can function across borders.

Businesses Are Preparing for a Different Kind of Customer

The rise of AI agents could also force businesses to rethink the definition of a customer.

Historically, companies have optimized their websites for people.

In the emerging agent economy, software may increasingly become the first point of interaction.

This could influence product descriptions, pricing structures, loyalty programs, payment systems and customer service.

Businesses may need to create interfaces that allow AI agents to interact safely while preventing malicious automated activity.

The companies that adapt early could potentially gain access to a rapidly changing form of digital commerce.

The Road Ahead

The transition to agentic commerce is unlikely to happen overnight.

Consumers will probably continue making many purchases themselves.

But the technology is moving toward a model in which AI systems can increasingly handle repetitive decisions and transactions.

The important question is therefore not whether AI agents will participate in commerce, but how much authority they will eventually receive.

The financial industry is now confronting that question directly.

The emergence of “Know Your Agent” frameworks shows that payment companies recognize that identifying the human customer may no longer be sufficient when software is increasingly acting between the customer and the financial system.

The next phase of digital commerce will depend on whether convenience can be combined with control.

If AI agents can operate within clearly defined permissions, authenticate themselves reliably and provide an auditable record of what they were instructed to do, autonomous payments could become a major part of everyday commerce.

If security, accountability and consumer trust fail to keep pace, adoption could face significant resistance.

Either way, the financial system is entering an era in which the question is no longer only “Who is the customer?”

Increasingly, it will also be “Which AI agent is acting for that customer, what authority does it have, and who is responsible for what it does?”