India’s Next UPI Leap: Letting AI Agents Pay Within Rules You Set

India is preparing a framework that would allow artificial intelligence agents to make small digital payments on a user’s behalf without asking for approval at every checkout. The plan, reported by Reuters and carried across Indian business outlets, would run on the Unified Payments Interface rather than on a new network. If it is launched as described, UPI would become one of the world’s largest rails for agentic payments: software that does not only suggest a product but can also complete the purchase when the user’s own limits are met.

The proposed standard is called the Unified Agent Protocol. People familiar with the matter say it is likely to be shown next week at the Global Fintech Fest in Mumbai. NPCI did not immediately comment when asked. The timing still fits the calendar the payments body has already advertised. The 2026 festival runs from 9 to 11 September at the Jio World Centre, and agentic AI is one of its stated themes. NPCI began piloting an agentic layer last year and demonstrated it at the 2025 edition. A public protocol would turn those trials into a common rulebook for apps, merchants and agents.

Groceries and e-commerce likely first use cases for AI-driven payments. | India Today

What an agent would be allowed to spend

The design is meant to give limited authority, not a blank cheque. A user would set rules in advance: when an agent may pay, how much it may spend, and possibly what kind of purchase it may make. The first cases are expected to be frequent, low-value orders such as groceries. An online shopper could authorise an agent once and let later payments go through if they stay inside the budget. Over time the same machinery could watch for discounts and place an order, or invest only when a price crosses a threshold the user has already chosen.

The protocol is not intended to replace UPI. Earlier reporting described it as a trust layer on top of the existing system: a way to register, verify and authorise agents so they can transact across UPI apps without changing the rails underneath. In that model, NPCI would confirm that an agent is trusted and acting with consent. The money would still move through UPI. Privacy is part of the argument. The operator would verify the payment, not necessarily inspect what was bought. Interoperability is the other aim. Without a shared standard, every app and every agent would invent its own handshake.

The two UPI tools already built for delegation

The framework is expected to lean on UPI Circle and Reserve Pay. Circle already lets a primary account holder give payment authority to another person, with monthly and per-transaction caps. Full delegation today can run up to ₹15,000 a month and ₹5,000 a transaction, with a cooling period after a new link is created. The agentic idea is to treat an AI agent as that secondary actor, still bound by limits the account holder sets. Reserve Pay, also known as Single Block Multiple Debit, lets a customer block funds once and allow several later debits against that pool. Banks currently cap those blocks at ₹10,000 for up to 90 days. Sources say both the amount and the validity period could be reviewed if agents start paying. Unused money is released if the mandate is cancelled.

Those two features matter because ordinary UPI was built for a person at checkout. An MPIN or biometric step stops a machine mid-purchase. Mandates already solve a version of that problem for subscriptions and recurring bills. Extending them to agents is the next step: one upfront consent, then payments that fire when the user’s rules are met.

Private firms have already started

A national protocol would arrive after months of commercial work. In June, Pine Labs launched P3P, which uses UPI one-time mandates and Reserve Pay so an agent can complete a payment without a fresh human authentication. The first live case is Gullak, a digital gold platform. A user can set a rule such as buying ₹500 of gold if the price falls below ₹16,000 a gram, approve a mandate once, and let the agent execute when the condition is hit. Vijay Sales has been in a proof of concept. Pine Labs has also said it is talking to card networks about extending the same mandate logic beyond UPI.

Razorpay has said it has been running agentic payment experiences in production with NPCI for months, including earlier work with large language-model products. Mastercard completed a fully authenticated agentic commerce transaction on its network in New Delhi, using tokenisation and existing security features with cards issued by Indian banks. Visa is separately developing agentic capabilities for the Indian market. The public protocol would not invent the idea. It would try to make those separate tracks work with one another.

Why UPI’s size changes the stakes

UPI’s reach is the strategic reason this matters. It is the world’s largest retail fast-payment system by volume, according to a 2025 IMF assessment. In August it processed 24.51 billion transactions worth ₹29.82 trillion. Google Pay and PhonePe together handled about three-fourths of that volume. An agentic layer on that base would reach hundreds of millions of users and merchants without building a new network. NPCI is expected to give merchants a way to integrate directly and to let customers set spend rules with audit trails and identity checks. Sources also say a liability framework is planned. The details have not been published.

That gap is the hard part. If an agent buys the wrong item, pays the wrong merchant, or is tricked by a prompt, it is not yet clear who bears the loss. Dispute rights, cancellation, and how much purchase data an agent or merchant should see remain open. The festival may also bring other UPI changes, including interoperability for AutoPay mandates so recurring permissions can move across apps. Those are separate from the agent protocol, but they point in the same direction: less friction, more automation, and a heavier need for clear consent and redress.

The MGMM Outlook 

India’s proposed Unified Agent Protocol marks an important step in the evolution of digital payments by combining the convenience of artificial intelligence with the trust and security of UPI. Rather than replacing the existing payment infrastructure, the framework is designed to let users authorize AI agents to make small, rule-based purchases within predefined spending limits, ensuring that automation remains firmly under the user’s control. Features such as UPI Circle and Reserve Pay provide the foundation for delegated payments while preserving consent, spending caps, and auditability.

As India continues to lead the world in real-time digital payments, a standardized agentic payment framework could make everyday commerce faster, smarter, and more interoperable across apps and merchants. Its long-term success, however, will depend on robust safeguards around liability, privacy, and dispute resolution, ensuring that innovation strengthens consumer confidence while expanding the capabilities of the country’s digital economy.

(Sources: Firstpost, India Today, Business Standard)