India is preparing for a major change in how people use UPI: instead of approving every purchase themselves, users could eventually give an artificial intelligence agent permission to make certain payments on their behalf. The idea is not to hand an AI unrestricted access to a bank account. It is to create tightly defined permissions, spending limits and audit trails that allow an agent to complete routine transactions within boundaries set by the user.

Reuters reported on September 1 that India is preparing a rollout of agentic payments on the Unified Payments Interface, with a new Unified Agent Protocol expected to provide a common framework for AI-led transactions. The development matters because UPI already operates at extraordinary scale: it processed 24.51 billion transactions worth ₹29.82 trillion in August 2026, according to figures cited by Reuters.

What are agentic UPI payments?

An agentic payment is a transaction initiated or completed by an AI agent under authority previously granted by a user. Today, a typical UPI purchase is deliberately human-led: a person chooses a merchant, checks the amount and authenticates the payment. An agentic system changes the workflow by allowing software to perform some of those steps after receiving a goal and a defined set of permissions.

A user might tell an AI assistant to buy a regular grocery basket when prices fall below a certain level, for example. The agent could compare eligible merchants, select products and pay, but only if the transaction fits the user's approved rules. That is fundamentally different from giving an AI a UPI PIN. NPCI's existing public guidance says UPI PINs are used to authorise bank transactions and should never be shared.

How India's proposed Unified Agent Protocol could work

The proposed Unified Agent Protocol, or UAP, is intended to create a common trust layer rather than leave every AI company, bank and merchant to build incompatible systems. Earlier reporting on the protocol described a framework for registering, verifying and authorising AI agents on UPI while maintaining spending controls and records that can be audited if something goes wrong.

Reuters says the framework is expected to use existing UPI capabilities including UPI Circle and Reserve Pay. That is important because India does not need to rebuild its payment network from scratch. The new layer can build on mechanisms designed for delegated authority and reserved funds, then add identity and controls suited to software agents.

The user would still define the boundaries

The most important distinction is between delegation and unlimited autonomy. A useful agentic payment system needs explicit rules covering how much an agent can spend, which merchants it can use, what type of purchase it can make and how long the permission remains valid. Users also need a clear way to revoke access. Reuters reported that safeguards under consideration include spending limits, identity verification and audit trails.

Why groceries are a natural first use case

Routine, low-value purchases are a logical starting point because the user's intent is often predictable. Someone who buys the same milk, vegetables and household supplies every week may value an assistant that can find suitable options and complete checkout without requiring a separate app journey each time.

India has already tested pieces of this model. Razorpay and NPCI announced an agentic-payment initiative with OpenAI in 2025 using UPI Circle and UPI Reserve Pay, with BigBasket among the initial merchants. Razorpay described a flow in which an AI could discover products and place an order after user confirmation, while users retained tracking and revocation controls. A separate 2026 pilot brought agentic UPI payments to Claude for selected food and grocery services.

Why this is a bigger change than conversational payments

UPI already supports conversational interfaces. NPCI's Hello! UPI, for example, is designed to let people navigate payments using voice and Indian languages. But the existing model still centres the human as the transaction initiator and authenticator. Agentic commerce goes further because software can make decisions and initiate an authorised payment within predefined limits.

That shift is particularly significant given UPI's rapid growth into India's dominant digital-payment rail. A common national protocol could mean that an authorised agent works across participating apps and merchants rather than being locked into one company's checkout system.

What could change for consumers

For consumers, the immediate attraction is convenience. An AI assistant could potentially move from recommendation to transaction without forcing the user to repeat searches, open multiple apps and approve each small step. Over time, agents could help manage recurring purchases, travel bookings or other tasks where the user can express clear budget and merchant constraints.

The risk is that a mistaken recommendation becomes a real financial transaction. That makes reversibility, dispute handling and transparency essential. Users need to know which agent acted, what rule authorised the payment, which merchant received it and how to challenge an incorrect transaction. Convenience cannot come at the cost of losing visibility over bank-account activity.

What banks, payment apps and merchants will need to solve

Banks and payment providers will need to distinguish a legitimate delegated transaction from fraud. That means agent identity, user consent and transaction scope must be machine-verifiable. Merchants, meanwhile, will need systems that can expose product information and checkout actions reliably to authorised agents without creating new security gaps.

There is also a competitive question. If AI assistants become a new interface for commerce, the software choosing between merchants could influence where billions of rupees of consumer spending goes. Rules around ranking, sponsored recommendations, conflicts of interest and disclosure may become increasingly important as agentic shopping expands.

India is not starting from zero

India's advantage is the scale of its existing digital infrastructure. UPI already connects banks, payment apps and merchants through an interoperable network, while India's broader digital economy continues to attract large technology investments, including developments such as the Jio Platforms IPO process. Agentic payments attempt to add a software-agent layer on top of infrastructure consumers already use rather than establish an entirely separate payment method.

Private companies are also experimenting with their own approaches. Pine Labs introduced its P3P protocol for agentic UPI payments in 2026, using existing mandate infrastructure. Globally, major payment networks are working on related technologies. A national interoperable framework could therefore become strategically important if India wants AI commerce to remain as open as UPI itself.

What happens next

The key point for users is that unrestricted AI access to UPI is not suddenly being switched on. The framework is being developed around controlled delegation. Reuters reported that the Unified Agent Protocol is expected to be unveiled at the Global Fintech Fest in Mumbai, and the initial use cases are expected to focus on relatively small purchases before more complex activities are considered.

The details will determine whether the idea succeeds. Spending limits, agent verification, merchant controls, dispute resolution and clear consent need to work together. If they do, agentic UPI could turn AI assistants from tools that merely suggest what to buy into software that can complete tightly controlled financial tasks. If the safeguards are confusing or weak, users may reasonably prefer the certainty of approving payments themselves.

For India, the stakes are unusually high because digital payments are already woven deeply into everyday commerce and the economy is continuing to expand rapidly, as shown by the latest GDP growth figures. A successful agentic-payment standard could become another exportable piece of India's digital public infrastructure. The next milestone will be the formal protocol details and the rules governing which agents can transact, under what limits and with what protections for the person whose money is ultimately being spent.