UPI processed 22.71 billion transactions in June 2026, worth Rs. 28.92 trillion. The person behind most of these transactions was a human. NPCI is now designing the infrastructure for a version where that may not always be the case.
They are developing a Unified Agent Protocol (UAP) which includes allowing AI agents to make UPI payments on a user’s behalf without per-transaction approval. The protocol has not yet launched and requires RBI regulatory approval before it goes live.
This article highlights the developments made on this in recent times and the three things it does differently, along with what responsibilities would AI agents actually handle.
Key Takeaways
- The user defines what the AI can perform before it starts doing anything. The AI operates within those boundaries
- The first use cases are expected to be low-consideration, recurring purchases where the decision is already effectively automated by habit
- An AI agent operating within pre-set permissions executes what the user has already decided; it does not make a new spending decision
- It builds on UPI Circle’s delegated framework, adds a registration and verification layer for AI agents, and requires RBI approval before launch
Most people hear “AI pays your bills” and picture an AI system with full access to a bank account. The UAP works differently.
The protocol is a trust and authorisation layer that is positioned above the existing UPI payment guidelines. It doesn’t change how UPI works underneath. It simply adds a layer that registers, verifies, and authorises AI agents before they can transact.
The user defines what the AI can perform before it starts doing anything. The AI operates within those boundaries. The bank account and UPI infrastructure remain unchanged.
Digital payments companies in India have chosen not to build rival protocols. They will work within the NPCI framework, per Business Standard.
The user control framework is the most important part of the protocol.
| User-Defined Permission | What It Controls |
|---|---|
| Merchant category | Which types of merchants the AI can pay |
| Spending limit per transaction | Maximum amount for a single payment |
| Spending limit per period | Total the AI can spend in a day, week, or month |
| Permitted circumstances | Conditions under which the AI may execute a payment |
For instance, if a user sets up an AI agent for grocery payments, with a limit of Rs.500 per transaction and Rs. 3,000 per week, the AI cannot on its own pay a travel merchant, nor can it exceed the payment limit. This is bounded access, not open action.
The first use cases are expected to be low-consideration, recurring purchases where the decision is already effectively automated by habit.
Bill payments are the clearest example. Electricity bills, mobile recharges, and subscription renewals happen on a schedule. The amounts are predictable. The merchants are known.
An AI agent operating within pre-set permissions executes what the user has already decided; it does not make a new spending decision.
Daily grocery purchases follow the same logic. The AI acts on an instruction the user made once at the permission stage.
NPCI runs UPI. It also runs the UPI Circle, which already allows a primary account holder to keep limited payment permissions for a trusted person. UAP extends this existing delegated payments model to include AI agents as authorized delegates.
India would be among the first countries to build national infrastructure for agentic payments. Similar frameworks are being developed in other markets, but as private-sector initiatives. India’s approach through NPCI creates an interoperable national standard.
Before things proceed, two conditions must be met. The protocol must complete the full development cycle. NPCI is working on it in consultation with industry stakeholders.
The Reserve Bank of India must grant regulatory approval. Until both happen, UPI transactions continue as they do today: user-initiated and manually confirmed.
NPCI’s Unified Agent Protocol would let AI agents make UPI payments within boundaries the user defines. It builds on UPI Circle’s delegated framework, adds a registration and verification layer for AI agents, and requires RBI approval before launch. The payments it would handle first are the ones users already make on autopilot. The AI executes the action. The user makes the decision once, at the permission stage, before anything moves.
Q1) What is the role of UAP in the integration of AI in UPI systems?
Ans: The following are the three things UAP is tasked to do:
Q2) Will the AI agents have full access to my bank account?
Ans: No, the agents only work on top of the existing UPI system, where they’d require permissions to gain limited access before even proceeding with an automated payments system.
Q3) Can it directly make new, unrelated payments?
Ans: No, an AI agent operating within pre-set permissions only executes what the user has already decided; it does not make a new spending decision.
