TRENDING
Reddit Says Google's AI Mode Is Cutting Its Search TrafficMistral Releases Open Weights for Its Trillion-Parameter Large 4OpenAI Launches GPT-6.1 Terra, a Cheaper Computer-Use AI ModelGoogle DeepMind's Genie 4 Builds 3D Worlds With SoundAnthropic Files for IPO on Nasdaq Under Ticker ANTOpenAI Models and Codex Now Available on Google Cloud29 Open-Source Alternatives to Paid Software on GitHubSpaceship Promo Codes October 2026: Cheap Domains for AI StartupsPHP2Shell Explained: The PHP Flaw Putting 70% of Websites at RiskClaude Duet Explained: Anthropic's New Proactive AI AgentElice IPO Crash and CoreWeave Bond Stress Expose AI Debt RiskAnthropic Quietly Pitches Congress Its Own AI Safety BillJio AI Classroom: Free 4-Week AI Course With CertificateA Hacker Used AI Agents to Breach 100+ Companies Including DeereMicrosoft Launches Guardian Agents to Police AI Agents at Work

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

In early September 2026, 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. Reporting since then by Reuters and Bloomberg 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. We lay out the actual mechanism, and who's already using it.

⚡ Quick facts

  • The building blocks: NPCI is reportedly extending two existing UPI tools: UPI Circle (delegating payment authority to a secondary party) and Reserve Pay (blocking funds upfront for multiple future debits)
  • Current cap: Reserve Pay blocks are currently limited to about ₹10,000 for up to 90 days, though this could change specifically for agent use
  • Already live: Pine Labs' "P3P" protocol has let AI agents pay via UPI in production since June 11, 2026, with digital gold app Gullak as its first live partner
  • The timing: This landed just before Global Fintech Fest 2026 in Mumbai (September 9-11). At the event NPCI described the protocol's design principles but did not launch it or announce limits, liability rules or a rollout date, according to MediaNama
Advertisement

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. Reuters, citing sources, and a Bloomberg follow-up added 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 same two tools were already tested in a 2025 pilot in which NPCI, Razorpay and OpenAI let users pay inside ChatGPT with UPI.

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, Inc42 reported, NPCI is 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 processing billions of 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. According to YourStory, it extends UPI's existing mandate tools (One Time Mandates and Reserve Pay) to agent-triggered payments, and is 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, the web's status code for machine-readable payment requests, according to Pine Labs. Its first live partner is Gullak, a digital gold savings app, where users set a rule such as buying a fixed amount of gold when the price falls below a target, and approve one UPI mandate; the agent then executes automatically whenever the condition is met. Electronics retailer Vijay Sales is running a proof-of-concept, with more pilots in retail, fintech and travel. 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 Reuters, Bloomberg and other outlets. At Global Fintech Fest in September, MediaNama reported, NPCI described the protocol's design principles but did not launch it, and as of late October we found no official launch. Until NPCI announces one, exact spending caps, eligible transaction types, and the liability rules remain provisional.

Our take: a worked example, and a checklist before you say yes

The ₹10,000 Reserve Pay cap sounds small, but it shapes how an agent can be used. Take a grocery agent that does a ₹2,200 weekly order:

Before you authorize any AI agent to pay from your UPI account, under P3P today or NPCI's framework later, check five things:

What it means 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. Related: JioFinance’s AI personal CFO.

Sources

Frequently asked questions

How much of my money does a Reserve Pay block tie up?

The full blocked amount, up to about ₹10,000 today, is reserved in your account until it is spent, expires or you revoke it, so you cannot use it for other payments meanwhile. Unspent money should be released when the block ends.

Is UPI actually letting AI agents pay for things right now?

In a limited way, yes. Pine Labs' P3P protocol has been live since June 2026, and Gullak already lets an AI agent execute pre-approved gold purchases automatically. NPCI's own, official framework for all of UPI is still being prepared.

How would an AI agent actually be allowed to pay from my account?

Reports point to two existing UPI tools being extended: UPI Circle (delegating payment authority to a secondary party) and Reserve Pay (blocking funds upfront for multiple future debits), combined with a one-time mandate and spending limits you set.

What's the maximum an AI agent could spend without asking me each time?

Reserve Pay is currently capped around ₹10,000 for up to 90 days, though sources say limits could change specifically for agentic use once NPCI's framework is finalized. Nothing about final caps is officially confirmed.

Is this safe, and can I opt out?

The systems described are opt-in, you set the rules and limits, and can revoke a mandate anytime. NPCI is reportedly planning identity verification, audit trails, and a liability framework, but it's worth reviewing exactly what you're authorizing before turning it on.

Related articles

Comments