NPCI's Unified Agent Protocol Puts AI Agents on UPI's Agenda at Global Fintech Fest

As Global Fintech Fest 2026 opens in Mumbai this week, the framework drawing the most attention isn't a new app or a new bank tie-up — it's a protocol that would let software agents, not people, initiate payments on UPI. The National Payments Corporation of India (NPCI) has been developing a Unified Agent Protocol (UAP) that would register, verify and authorise AI agents to transact over India's payments rail, and the timing of the disclosure puts it squarely at the centre of this year's summit.

What the Protocol Actually Does

UAP is designed to plug into UPI's existing architecture rather than requiring a separate rail. It builds on UPI Circle, a feature NPCI introduced to let a primary account holder delegate payment authority to a secondary user — typically a family member — within preset spending limits, without handing over full account access. UAP's innovation is to let an AI agent occupy that secondary-user role instead of a person. A user could authorise an agent to make small, repetitive payments — recurring bill payments, subscription renewals, grocery reorders — or to execute purchases automatically when a tracked item hits a target price, all without approving each individual transaction.

The safeguards under discussion include hard spending limits set by the user, identity verification for the agent itself, predefined rules governing what categories of purchase an agent may make, and audit trails that let a user (or regulator) reconstruct exactly which transactions an agent initiated and why. NPCI has been developing the standard in consultation with industry, and its actual rollout is expected to require a formal regulatory nod from the Reserve Bank of India before agents can transact live on the network.

Why the Timing Is Deliberate

Global Fintech Fest 2026 runs September 8–11 at Mumbai's Jio World Centre, with Prime Minister Narendra Modi set to inaugurate the event. The summit's theme this year — "Potential to Impact: Agentic AI, Tokenisation, Quantum: Trusted, Connected, Global Systems for Inclusive Finance" — was chosen specifically to foreground the shift toward autonomous financial software, and UAP gives India's own payments infrastructure a concrete answer to showcase rather than leaving the agentic-payments conversation to foreign platforms. Domestic payments-industry commentary in the run-up to the summit has focused heavily on the fraud and consumer-protection questions UAP raises — an AI agent that can spend within a limit is still an attack surface if its credentials or instructions are compromised, and India's fraud-detection infrastructure has been built almost entirely around the assumption that a human is initiating each transaction.

The Global Race for Agentic Payments Rails

Several private payment networks in the United States, Europe, Singapore and Australia have already launched frameworks that let AI agents transact on their behalf, typically through card-network protocols layered on top of existing rails. What distinguishes NPCI's approach is scale and ownership: UPI already processes the overwhelming majority of India's digital retail payment volume, and a national, NPCI-operated agent protocol — rather than a patchwork of bank-specific or app-specific agentic features — would make India one of the first countries to build agentic payments as core public payments infrastructure rather than a private-sector add-on. If UAP clears RBI's regulatory review and rolls out at meaningful scale, it would give India a claim to having built the world's largest agentic payments network essentially overnight, simply by virtue of UPI's existing user base.

What's Still Unresolved

NPCI has not published a public specification for UAP, and several open questions remain unanswered even as the protocol is discussed at this week's summit: how liability is apportioned when an agent makes an erroneous or fraudulent payment, whether merchants will be required to signal that a transaction originated from an agent rather than a human, and how the spending-limit and identity-verification mechanisms interact with UPI's existing fraud-monitoring systems. RBI's own posture on agentic payments has not been made public in detail, and its regulatory review — rather than NPCI's technical readiness — is likely to be the actual gating factor on when, or whether, UAP moves from protocol design to live transactions on India's payments network.

The Fraud Question Nobody Has Fully Answered

Payments-industry commentary ahead of the summit has focused heavily on a structural problem that UPI's existing fraud infrastructure was never built to handle: almost every fraud-detection heuristic UPI banks and NPCI rely on today — unusual transaction timing, atypical merchant categories, velocity of repeated payments — assumes a human is making a judgment call about each transaction, however quickly. An AI agent authorised to act within a spending limit could, in principle, generate exactly the kind of rapid, repetitive, round-the-clock transaction pattern that today's fraud models would flag as suspicious if a human account did it, forcing a redesign of what "normal" behaviour looks like on the network. There is also the compromised-agent scenario, distinct from a compromised human credential: if an attacker manages to manipulate the instructions or data an agent relies on rather than stealing a password outright, the agent could execute payments that are technically authorised under UAP's rules but never intended by the user — a failure mode current UPI fraud tooling has no equivalent for.

A Test Case for Public Digital Infrastructure

UPI's rise from a domestic real-time payments system to a template other countries have sought to license or replicate has largely rested on NPCI treating the rail as public infrastructure rather than a proprietary product — a posture that let banks, fintechs and merchants build on top of it without licensing fees. UAP extends that same logic to autonomous agents, but the stakes of getting it wrong are different in kind: a bug in a merchant-facing UPI feature typically affects that merchant's transactions, while a flaw in agent-authorisation logic could, at UPI's transaction volumes, create a systemic vulnerability across every bank connected to the network simultaneously. How carefully NPCI and RBI sequence UAP's rollout — starting with narrow, low-value use cases before expanding scope, as UPI Circle itself did — will likely determine whether agentic payments becomes UPI's next genuine leap in utility or its first major public trust incident.

Sources

• Why UPI agentic plans must answer fraud challenges — Payment Expert
• India to Pioneer Agentic AI Payments on UPI, Redefining Digital Commerce — IndianWeb2.com
• India's Global Fintech Fest 2026 to Highlight Quantum Technology, AI and Tokenization — The Quantum Insider
• PM Modi To Inaugurate Global Fintech Fest 2026 In Mumbai Tomorrow, Spotlight On India's Digital Finance Growth — Free Press Journal