How India's AI Agents Will Actually Pay With Your UPI, Explained

Earlier this month, reports said NPCI was testing something called a "Unified Agent Protocol" that would let AI agents make UPI payments on your behalf. What wasn't clear then was exactly how it would work under the hood. A new Bloomberg report published September 8, 2026 fills in that gap: India's agentic payments plan reportedly leans on two UPI tools that already exist for other purposes -- and, separately, a payments startup has already had AI agents paying through UPI in production since June. Here's the actual mechanism, and who's already using it.

What's actually new here

This site covered NPCI's proposed Unified Agent Protocol on September 5, when reports described the concept but not the mechanics. Bloomberg's follow-up, also carried by Business Standard and Reuters via Business Recorder, adds the missing detail: rather than inventing an entirely new payment rail, NPCI is reportedly building agentic payments on top of two UPI mechanisms that already exist for human users. That matters because it means the plumbing is largely already there -- what's being added is a way for an AI agent, instead of a person, to be the one authorized to use it.

The two UPI tools doing the work

The first is UPI Circle, a feature that currently lets a primary account holder delegate payment authority to a secondary person -- for example, a parent letting a family member make payments from a shared arrangement. Extended to agentic payments, that secondary party would be an AI agent instead of a person. The second is Reserve Pay, which lets you block a set amount of money upfront to cover multiple future debits, rather than authorizing each one individually. Banks currently cap Reserve Pay blocks at around ₹10,000 for up to 90 days -- small enough for repeat purchases like groceries, not large one-off spends -- though sources say these limits could be adjusted specifically once the agentic framework is finalized. On top of both, NPCI is reportedly planning merchant-integration tools that let you set rule-based instructions -- spending caps, identity verification, and audit trails -- governing exactly what an agent is allowed to do.

A startup already beat NPCI to it

While NPCI's own framework is still in the works, India-based payments company Pine Labs didn't wait. It launched a protocol called P3P (Pine Labs Payment Protocol) on top of UPI on June 11, 2026, arguing that UPI -- built for humans tapping in an MPIN, and already processing around 23 billion transactions a month -- had no way for an AI agent to get past that authentication step on its own. P3P works by having a person authorize a UPI mandate once upfront; after that, the agent can browse, select, negotiate, and pay without asking again each time. It's supported by two pieces: Grantex, which handles identity verification, spend controls, and keeps an audit trail so mandates can be revoked or updated anytime, and HTTP 402, an existing open web standard for machine-readable payment requests. Pine Labs' first live partner is Gullak, a digital gold savings app, where users set a rule like "buy ₹500 of gold if the price drops below ₹16,000/gram" and approve it once; the agent then executes automatically whenever the condition is met. Electronics retail chain Vijay Sales, which runs more than 150 stores, is reportedly running a proof-of-concept next. Separately, Mastercard says it completed its first authenticated agentic transaction in New Delhi back in June 2026, underscoring that private players are moving well ahead of any official, India-wide framework.

The guardrails, and what's still missing

Every version of this described so far is opt-in and rule-based, not a blank check: you authorize a mandate once, set spending limits and conditions, and can revoke it at any time. NPCI is reportedly planning a liability framework to sit alongside the technical rollout -- covering who's responsible if an agent makes a mistake -- but that framework hasn't been publicly detailed yet, and NPCI has not officially confirmed the specific mechanics reported by Bloomberg and other outlets. Until it does, exact spending caps, eligible transaction types, and the liability rules remain provisional.

Why this matters beyond India

Agentic commerce -- AI agents that browse, decide, and pay without a human clicking "confirm" each time -- is being built in pieces around the world, and companies like Mastercard are already testing it across markets. What makes India's approach notable is that it's reusing UPI, a single national payments rail that already processes billions of transactions monthly, rather than each app or bank building its own separate agentic payment system. If NPCI's framework lands, India would be among the first countries with that kind of shared, national infrastructure for AI-driven payments -- a pattern this site has tracked all year as apps hand AI agents more direct access to real accounts, from WhatsApp connecting third-party AI agents to individual chats to Meta's own Hatch agent needing extra safeguards after mishandling permissions during internal testing.

Delegated Payment State Machine & Biometric Step-Up Authentication

The mechanism governing autonomous AI agent-driven UPI transactions follows a rigorous state machine that transitions payments through multiple approval checkpoints. Each delegated payment begins in an "Authorized Intent" state after the user's primary UPI Circle mandate is verified, progressing through "Risk Assessment", "Context Validation", and finally "Execution" states where the transaction is cryptographically signed against the payer's device-bound keystore.

Frequently asked questions

Related articles

Sources: Bloomberg (via Business Standard and Business Recorder / Reuters, September 8, 2026), and Pine Labs' official announcement (June 11, 2026). NPCI has not officially confirmed the specific mechanics reported for its own framework.

Comments