Agentic payments are moving closer to India’s UPI system just as merchant pricing begins to change. Together, these shifts raise a practical question: who will pay for autonomous commerce?
From October 15, 2026, selected UPI merchant payments above ₹2,000 will attract a merchant discount rate (MDR). Meanwhile, Reuters has reported that NPCI is preparing a framework for AI agents to make UPI payments for users.
The two developments are connected. Safe payments by AI agents need identity checks, spending rules, fraud controls, and strong evidence. Building that layer will cost money. Therefore, the new MDR framework may help fund the next stage of UPI.
What Is Changing in UPI Pricing?
For more than six years, zero MDR helped UPI grow across India. Merchants could accept digital payments at little direct cost. Consumers also gained a fast and simple way to pay.
However, banks, payment apps, and other participants earned little direct revenue from the system. They still had to operate it, secure it, and support users.
The new framework changes that balance. Specified person-to-merchant payments above ₹2,000 will attract MDR of 0.4%. For payments of ₹75,000 or more, the charge will be capped at ₹300.
Some sectors will follow a different structure. Payments for fuel, railways, telecommunications, insurance, and agricultural inputs will carry a flat ₹5 charge. Capital-market payments will attract 0.02%, capped at ₹300.
Customers will not pay MDR directly. Person-to-person transfers remain free. Payments to qualifying P2PM small merchants also remain free when their monthly UPI receipts do not exceed ₹1 lakh.
The scale explains the policy debate. UPI processed 24.51 billion transactions worth ₹29.82 lakh crore in August 2026, according to NPCI’s official statistics. Infrastructure at that scale needs continuing investment in capacity, cybersecurity, fraud prevention, and customer support.
The Ministry of Finance has linked merchant pricing to long-term sustainability. Yet the industry still needs to decide how revenue will be shared among banks, payment apps, aggregators, and other providers.
How Agentic Payments Differ From Automation
Automated payments already exist. Customers use standing instructions, direct debits, and UPI AutoPay. Those tools follow precise commands about the recipient, amount, and timing.
Agentic payments can involve a much broader task. For example, a user might ask an agent to renew an insurance policy if the terms remain acceptable. Another instruction could ask it to buy a flight within a budget.
The agent may search, compare options, interpret conditions, and decide whether a purchase fits the customer’s preferences. As a result, valid payment credentials may not prove that it respected the user’s instructions.
Consider a request for the “best reasonably priced morning flight.” The agent might select a cheaper nonrefundable ticket with a long layover. The payment could be technically authorized even though the purchase fails the customer’s real goal.
Financial services make the problem harder. An insurance agent must notice changes in coverage, exclusions, and premiums. An investment agent may need to interpret risk limits, cash needs, and product suitability.
These decisions combine payment execution with product choice. Therefore, payment authority should not automatically grant decision-making authority.
Why the Customer Mandate Matters
Before an AI agent can spend, the system must identify its owner and software. It must also know the exact scope of the agent’s authority.
A mandate may set a limit for each transaction or period. It can restrict merchants, product categories, locations, and times. In addition, it should define whether subscriptions are allowed and when human approval is required.
Customers also need a simple way to change or withdraw that authority. They should not have to visit several platforms to stop one agent.
Reuters reported on September 10, 2026, that NPCI was building a registry to identify and monitor AI agents using UPI. The report cited three sources involved in the discussions. However, a registry alone cannot prove that a transaction followed the customer’s instructions.
A safe agentic-payment flow connects customer consent, spending limits, secure payment, and an auditable record.
A safe system must connect three records. These are the agent’s identity, the customer’s mandate, and the evidence created when the agent acts. Together, they help payment providers decide whether to process a transaction. They also matter when the customer raises a complaint.
Who Should Pay for Agentic Payments?
Today’s UPI chain may include the customer’s bank, the merchant’s bank, a payment app, NPCI, and a payment aggregator. Agentic commerce could add an AI platform, an agent developer, identity services, mandate tools, and new monitoring systems.
Each participant may add value. However, each can also create a new point of failure.
One option is to recover the cost through MDR. Merchants would pay because the agent helped complete a sale. Yet automated shopping could also create more returns, cancellations, and disputes.
Banks and payment apps could fund some controls from their share of MDR revenue. They already manage authentication, monitoring, and many complaints. If they must absorb agent-related fraud, they will demand strict access controls.
AI companies could pay for access to the payment layer. An agent provider may earn subscription revenue or commissions from the transaction. Therefore, it should bear part of the verification and compliance cost.
Users may fund some premium services. Basic UPI payments could remain free while customers pay for an agent that manages travel, bills, or household purchases. Still, users may resist another fee when the agent already comes with a banking or technology subscription.
In practice, the model may combine merchant charges, revenue sharing, and fees from agent providers. The commercial arrangement should follow responsibility. A participant that controls the agent’s behavior should bear the cost when that behavior causes harm.
Where Agentic Payments May Begin
The ₹2,000 threshold makes low-value purchases a natural testing ground. Groceries, food orders, local transport, and inexpensive digital services offer frequent but limited transactions.
These use cases let the industry test mandates, notifications, revocation, and agent identity. At the same time, they limit the customer’s immediate financial exposure.
Travel, insurance, investments, and expensive goods offer more commercial value. However, they also create harder questions.
Paying an insurance premium is simple after the customer chooses the policy. Allowing an agent to decide whether that policy should be renewed is different. Likewise, an investment agent may transfer money into a chosen product. Selecting the product creates a regulated financial decision.
The first systems should preserve this boundary. Otherwise, a tool designed for payment convenience could drift into advice, suitability assessment, or regulated decision-making.
A New Contest for Control
Agentic payments will create a strategic layer above the UPI rail. NPCI and regulated institutions will control access to accounts and settlement. Banks and payment apps will remain central to authentication and fraud management.
AI companies will seek to own the interface where customers express their intentions. Meanwhile, merchants will decide which agents can access their catalogs, prices, and checkout systems.
The strongest position may belong to the platform that converts an informal request into an enforceable mandate. That function has commercial consequences. It establishes who understands the customer’s intent, who can influence the transaction, and who holds the evidence during a dispute.
Agent identity, permission management, and transaction records may become sources of market power. If one platform holds a customer’s preferences and spending history, moving to another agent may become difficult.
India should avoid placing a closed agent layer over an open payment system. Related systems already use AI to choose payment paths. FinTech Central’s guide to AI payment routing explains how those models balance approvals, fraud, and cost.
Liability Rules Must Come First
Several issues must be settled before autonomous UPI transactions move beyond limited uses. Who is responsible when an authorized payment conflicts with the customer’s request? What evidence will prove the user’s intention?
The rules must also explain how a customer can withdraw a mandate during a purchase. In addition, users need a clear complaint path when the bank, merchant, and AI provider blame one another.
Sometimes the agent will follow its instructions exactly, but those instructions will rely on incomplete information. Responsibility may then depend on the agent’s role. It may have executed a mandate, offered financial advice, or made a choice for the customer.
Existing payment rules usually assume that a person initiated the transaction or created a conventional automated instruction. An AI agent adds interpretation between the instruction and the payment. Most of the new risk lies in that gap.
India Can Set the Rules Early
India has an interoperable, national-scale, real-time payment system. That gives it a strong base for common agentic-payment standards.
Elsewhere, agentic commerce is developing across cards, wallets, merchant platforms, and private technology networks. UPI could support shared rules for agent identity, delegated authority, revocation, and transaction evidence.
Scale alone cannot produce trust. Customers need to understand the authority they granted. Banks need to know who monitors each agent. Merchants must be able to recognize legitimate agent traffic.
The complaint process should also give users a clear point of responsibility. No customer should have to move among the bank, payment app, merchant, and AI provider to find help.
UPI’s new merchant-pricing framework starts to answer how the system can finance its next stage. Agentic commerce adds a second question: how should the costs and risks of autonomous transactions be divided?
Whether an AI agent can make a UPI payment is becoming a technical question with a technical answer. Yet the rules on cost, control, and liability will determine whether the system earns the right to scale.
Explore more analysis in FinTech Central’s AI in Payments and Lending section.