ET World Leaders Forum 2026: AI will reshape payments, but trust will be the real battleground

“Agents are today going out and doing the purchase for you. They are looking at variants, creating the carts and doing checkouts.”

That comment from Suresh Sethi, Group Country Manager, India & South Asia, Visa, captures a potentially fundamental shift coming to payments: the buyer may soon not be human at all.

AI agents are increasingly being designed to discover products, compare options, create carts and eventually complete purchases and payments on behalf of users. But as machines begin spending human money, the biggest question for banks, payment networks and merchants will be whether they can trust the transaction.


Also Read: Indian CEOs should stop ‘just asking questions’ and start building with AI: Anthropic India MD

Speaking at The Economic Times World Leaders Forum 2026 during a panel, Sethi said the internet was built largely for human interaction, but commerce now needs to become “agent-ready”.

The panel also featured Madhav Kalyan, MD & Head of Payments – Asia Pacific, J.P. Morgan; PD Singh, CEO, India & South Asia, Standard Chartered Bank; and John O'Loghlen, MD – APAC, Coinbase, and was moderated by Rahul Jain, India Head, BCG.

That means websites and payment systems must be able to recognise legitimate AI agents, allow them to navigate transactions and, crucially, establish that an agent is acting on the user's actual intent.

The new payment password may be intent

For years, payment security has centred on authentication and payment credentials. In an agentic world, that may not be enough.

“If I am going on a journey and actually purchasing something, I need to be sure that the site where the purchase is happening is able to understand that this was the intent,” Sethi said.

That is where tokenisation and new forms of payment credentials could become important. The system needs to distinguish between an AI agent acting on your instructions and a malicious bot trying to exploit the same infrastructure.

And AI is likely to make that battle harder.

Also Read: ET World Leaders Forum 2026: Smriti Irani on politics, power and the price of being a woman

Fraudsters can use AI to create convincing phishing messages, automate attacks and operate at much greater scale. But payment networks can also use AI to identify suspicious behaviour.

Sethi said the industry will need to move away from static fraud rules towards real-time, risk-based intelligence, looking at factors such as transaction velocity, timing, value and behavioural patterns.

The result could be what he called “dynamic friction”: legitimate transactions move through seamlessly, while suspicious ones are slowed down so systems can establish whether the intent is genuine.

The payment rail could disappear

The shift could also change the role of banks and payment companies.

Madhav Kalyan, MD & Head of Payments – Asia Pacific at J.P. Morgan, said consumers ultimately don't care which payment rail is being used. They care whether their money moves quickly, reliably, cheaply and safely.

Cards, UPI, wallets, tokenised deposits and stablecoins could therefore become increasingly invisible to the customer.

But that does not mean the institutions behind them become irrelevant.

Banks will still need to verify identity, establish whether an agent has the authority to act, settle transactions, move data and ensure the entire process remains auditable, Kalyan said.

PD Singh, CEO, India & South Asia, Standard Chartered Bank, said AI could fundamentally reshape banking, with banks potentially looking very different over the next three to five years as adoption accelerates.

For Coinbase's John O'Loghlen, the growing interaction between machines makes guardrails even more important. Human oversight, regulation, governance and security will remain critical as AI agents begin making decisions with real financial consequences.

The next payment revolution, then, maybe about figuring out when a machine should be allowed to spend your money — and how everyone else can be sure it is doing exactly what you asked it to do.