Launched for Autonomous UPI Transactions: What It Means and What to Do Next

So this happened fast. On June 11, 2026, Pine Labs officially launched for autonomous UPI transactions a protocol called P3P that lets AI agents complete UPI payments without you tapping a PIN, approving an OTP, or doing anything mid-transaction. You set the rules once. The agent handles the rest.

If you’re trying to figure out what this actually changes for your business, your wallet, or your next app build, here’s the real breakdown not the press-release version.

• Pine Labs’ P3P protocol is the most complete autonomous UPI system launched so far, anchored on existing UPI mandate frameworks (SBMD and One Time Mandate).
• This matters most for merchants, fintech builders, and shoppers who already use AI agents for recurring purchases — casual UPI users can mostly ignore it for now.
• The single biggest thing to get right is the upfront mandate: set spend limits and merchant scope carefully, because that’s the only checkpoint before the agent acts.
• The biggest mistake to avoid is treating this like a “set and forget” feature review your active mandates weekly, especially in the first month.
• If you want a lighter-touch version without full autonomy, Razorpay’s “Proceed to Pay” flow with Sarvam AI still requires one manual tap, which some businesses prefer for liability reasons.

What “Autonomous UPI Transactions” Actually Means Right Now

Here’s the thing “autonomous” doesn’t mean an AI agent has your bank password and goes shopping unsupervised. Pine Labs launched P3P, a payment protocol for agentic AI on UPI, allowing AI agents to complete transactions without human authentication after an initial mandate. That initial mandate is everything. It’s a one-time setup where you tell the system: this agent can spend up to X amount, with Y merchant, for Z purpose.

After that? The consumer authorises once, upfront. After that, the agent browses, selects, negotiates, and pays. No interruption, no second approval screen, no PIN popup.

This builds on something that’s been brewing since February. At the India AI Impact Summit in New Delhi on February 20, 2026, Razorpay and the National Payments Corporation of India officially announced a partnership to bring Agentic Payments to Claude, allowing AI agents to transact securely within user-set spending limits without repeated PIN prompts. Pine Labs took that groundwork and pushed it further with a full protocol stack.

What’s actually new with P3P versus earlier pilots: Grantex provides verifiable identity, delegated authorization, spend controls and auditability, and together with HTTP 402 a standardized protocol for discovering and requesting payment — the framework enables trusted, auditable agent-to-agent transactions without human intervention. That auditability piece is the part most coverage glosses over, and it’s the part that matters if you’re a business owner thinking about liability.

Why This Is Happening in India First (And Why That’s Not an Accident)

You might be wondering why India keeps showing up first for this stuff. It’s not hype it’s infrastructure. Very few markets globally have the infrastructure to support AI-led payments at scale, and UPI’s real-time, mandate-enabled framework provides exactly what agentic commerce requires: speed, security, transparency, and consent-driven control.

Most countries’ payment rails weren’t built with machine-speed transactions in mind. UPI was already real-time and mandate-based before AI agents entered the picture, so extending it to agents was less of a redesign and more of a layer on top. That’s a structural advantage companies like Pine Labs, Razorpay, and NPCI are clearly leaning into.

And Pine Labs isn’t shy about positioning itself as the standard-setter. Pine Labs CEO Amrish Rau said India should not look to the West to determine the protocols governing agentic payment transactions, and the company was the first to launch an agentic payments protocol within India’s regulatory framework, positioning it to potentially emerge as a de facto market standard as adoption grows. Whether that prediction holds up in six months is anyone’s guess, but right now P3P has a head start.

How This Compares to What Razorpay and Sarvam AI Already Have Live

Before you assume P3P is the only game in town, it’s worth knowing what’s already running. Razorpay and Sarvam AI built a voice-first ordering system, and it’s been live since late March. Here’s the key difference, and it’s bigger than it sounds:

In the Razorpay-Sarvam demo, the agent applies coupons within the conversational flow without the user leaving the chat, but at the final step a “Proceed to Pay” button appears and the user taps it manually the agent does not complete the transaction autonomously. Razorpay has said full autonomy is technically possible on their stack too, but the live demo stops one step short.

One detail that doesn’t get enough attention: if the agent orders the wrong thing, the merchant bears the liability, with authentication happening upfront and the spending limit capped at Rs 10,000. That Rs 10,000 cap is doing a lot of work — it’s a safety net that makes the whole system tolerable for merchants. P3P’s mandate-based approach is more flexible but puts more weight on you setting that initial mandate correctly.

Who’s Actually Using This Already

Real talk — most “AI agent payment” announcements are demos that never ship. This one has actual merchants live, which is rare enough to flag.

Digital gold savings platform Gullak is live on P3P, allowing users to use its AI agent to make purchases if the gold hits their target price, with the company describing this as moving from manual savings to autonomous wealth creation. That’s a genuinely useful use case you set a price target once, and the agent executes when conditions are met, without you needing to be online watching gold prices at 2am.

Retail electronics chain Vijay Sales is in active proof of concept on the protocol, which suggests this is moving beyond niche fintech apps into mainstream retail. If that POC goes well, expect other large retailers to follow within a quarter or two — that’s the usual pattern with UPI feature rollouts in India.

The honest truth: adoption at the merchant level always lags the tech by 6-12 months. Don’t expect every UPI-enabled store to support this by Diwali. But the rails are there now, which is the hard part.

What Nobody’s Telling You About the Risks

Here’s what surprised me digging into this almost every article covering the P3P launch focuses on the convenience angle and skips the boring-but-critical stuff. So let’s cover it.

Mandate scope creep. When you set up a one-time mandate with a spend limit, that limit applies to the agent’s full activity within that merchant relationship not per transaction. If your agent is doing recurring small purchases (say, automated grocery restocking), a Rs 5,000 monthly limit could get eaten up faster than you expect if prices fluctuate or the agent makes more frequent orders than anticipated.

Auditability isn’t the same as dispute resolution. Grantex provides verifiable identity, delegated authorization, spend controls and auditability great for proving what happened, but the actual dispute resolution process (who refunds you, how fast) still runs through your bank and NPCI’s existing UPI grievance channels. P3P doesn’t replace that; it just generates better records for it.

This is genuinely new regulatory territory. NPCI’s executive director for growth, Sohini Rajola, said at the India AI Impact Summit 2026 that the organisation plans to work with more AI platforms to expand UPI payments through AI agents — which is good news for ecosystem growth, but also means the rules are still being written as adoption happens. If you’re a business integrating this, build in manual override capability now. You’ll likely need it as RBI guidance catches up.

If you’re a developer or fintech founder: pull Pine Labs’ P3P documentation and test the mandate-setup flow with a sandboxed spend limit before touching production. Pay close attention to how Grantex’s audit logs structure data — that’s likely to become a compliance requirement, not just a nice-to-have.

If you’re a merchant evaluating this for your store: start with Razorpay’s manual-tap flow if you want lower liability exposure, and treat full P3P autonomy as a phase-two move once you’ve seen how the Vijay Sales POC plays out.

If you’re a regular UPI user curious about agent-powered shopping: it’s not mainstream-ready for you yet. Keep an eye on apps like Gullak that are already live that’s where you’ll get hands-on experience with this before it shows up in everyday shopping apps.

For more on how AI systems are being secured and governed as they take on more autonomous roles, our piece on AI agent identity security using voice biometrics and decentralized verification covers the identity-layer side of this same shift. And if your organization is trying to figure out where autonomous payment agents fit into broader risk policy, our framework for AI risk classification in organizations is a useful starting point before you greenlight anything customer-facing.

For official protocol details and regulatory updates, check NPCI’s official site.