India's NPCI Is Building a Protocol to Let AI Agents Pay on UPI

India's National Payments Corporation of India (NPCI) has built a framework called the Unified Agent Protocol (UAP) that will let AI agents execute payments on the Unified Payments Interface (UPI) without human approval for each transaction, according to Reuters, citing three sources. NPCI is expected to unveil UAP at the Global Fintech Fest in Mumbai from September 9 to 11, with initial use cases including groceries and subscription renewals, and future versions potentially enabling agents to act on flash sales or investment thresholds. The protocol builds on existing UPI Circle and Reserve Pay mechanisms, which currently cap delegated spending at Rs 15,000 per month and pre-authorized debits at around Rs 10,000 for up to 90 days, though these caps may be revisited for agentic use.

India's payments authority is about to let AI agents spend your money on UPI without asking permission for every transaction, betting that spending caps and audit trails can stand in for the split-second judgment a human normally provides. Reuters reported on September 1, 2026, citing three people familiar with the plans, that the National Payments Corporation of India has built a framework called the Unified Agent Protocol, or UAP, letting AI agents execute payments on the Unified Payments Interface without a human approving each one. NPCI is expected to unveil it at the Global Fintech Fest in Mumbai, running September 9 to 11 at the Jio World Centre. That's happening right now. The protocol isn't starting from scratch. It leans on two tools NPCI already runs. UPI Circle lets a primary account holder delegate payment authority to a secondary user, and banks currently cap full delegation at Rs 15,000 a month. Reserve Pay lets a customer block funds in advance for multiple future debits to a specific merchant, a mechanism banks have capped at around Rs 10,000 for up to 90 days. Under UAP, an AI agent would sit in that secondary or pre-authorized slot instead of a person, spending against a limit the user sets once rather than a button the user taps every time. Those caps, according to Reuters' sources, could be revisited specifically for agentic use. The first use cases are deliberately boring. Groceries. Subscription renewals. The kind of routine digital order nobody wants to sit through a checkout flow for. NPCI expects the scope to widen once the plumbing holds. Reuters reported that future versions could let an agent jump on a flash sale the moment a discount activates, or execute an investment instruction the instant an asset crosses a price threshold, tasks that reward speed over deliberation. To make that tolerable, NPCI is building in spending limits, identity verification, and audit trails, and it's separately working on a liability framework for when something goes wrong, though who eats the loss on a rogue agent transaction hasn't been disclosed. Nvidia's $40 Billion Spending Spree Revives Dot-Com Bubble Fears https://startupfortune.com/nvidias-40-billion-spending-spree-revives-dot-com-bubble-fears/ Nvidia has spent over $40 billion this year on stakes in OpenAI, SpaceX, Intel, CoreWeave and other companies that also buy its chips. Analysts including Wedbush's Matthew Bryson are drawing direct comparisons to the vendor financing that inflated telecom valuations before the dot-com crash. - nvidia spending 40 billion on company investments https://startupfortune.com/nvidias-40-billion-spending-spree-revives-dot-com-bubble-fears/ - dot com bubble fears nvidia vendor financing https://startupfortune.com/nvidias-40-billion-spending-spree-revives-dot-com-bubble-fears/ NPCI already has a live test case to point to. In October 2025, Razorpay, NPCI and OpenAI launched a pilot letting users complete UPI payments directly inside ChatGPT conversations, built on the same Reserve Pay and UPI Circle mechanics UAP now formalizes. Axis Bank and Airtel Payments Bank signed on as banking partners, and BigBasket, the Tata-owned grocery platform, and telecom operator Vi came in as the first merchants. That pilot is the reason NPCI can talk about grocery orders as a starting use case instead of a hypothetical. It's already been running one. Everyone in payments wants this to work India isn't alone in racing toward agent-run checkout, it's just moving faster than most. Visa introduced its Trusted Agent Protocol in October 2025 with more than ten partners, aimed at helping merchants tell a legitimate shopping agent apart from a bot scraping a checkout page. Stripe built out its Agentic Commerce Suite to let AI agents discover products and complete purchases through a single integration, and at the Visa Payments Forum this year Visa struck a deal with OpenAI to embed its network and fraud checks directly into agentic shopping. Mastercard completed what it called its first authenticated agentic transaction in New Delhi in June 2026. Domestic fintech Pine Labs has already gone further, launching P3P, a protocol that lets an AI agent complete a UPI payment with a single authorization. Frankly, the technology to let a bot spend your money already exists in half a dozen forms. What's missing everywhere, including in NPCI's own plan, is a settled answer for who pays when the bot gets it wrong. That's the real gap in UAP as described so far. Spending caps limit how much damage a compromised or malfunctioning agent can do in one go, and audit trails let a bank reconstruct what happened after the fact. Neither answers the question a fraud victim actually asks: does the liability sit with the user who set the limit, the bank that authorized the agent, or NPCI itself for building the rails. NPCI's promised liability framework is supposed to answer that, and until it's published in detail, UAP is a payment rail with the safety mechanism still labeled coming soon. Also read: Amazon and Alphabet's AI Bond Binge Is Repricing Debt Markets Worldwide https://startupfortune.com/amazon-and-alphabets-ai-bond-binge-is-repricing-debt-markets-worldwide/ • How Does AI Agent Memory Work Across Sessions, and Why It Keeps Failing https://startupfortune.com/how-does-ai-agent-memory-work-across-sessions-and-why-it-keeps-failing/ • Copper Hits a Record High and the AI Data Center Boom Is Paying for It https://startupfortune.com/copper-hits-a-record-high-and-the-ai-data-center-boom-is-paying-for-it/ This article is posted in AI News https://startupfortune.com/category/ai/ , check it out for more related stories. Investors Are Already Pricing Anthropic As A $2 Trillion Company https://startupfortune.com/investors-are-already-pricing-anthropic-as-a-2-trillion-company/ Investors trading Anthropic shares on secondary markets are already positioning for a valuation near $2 trillion, months before any IPO. The bet rests on a $47 billion revenue run-rate, over 1,000 million-dollar enterprise customers, and a $9.1 billion power deal with Bitcoin miner Riot Platforms. - Anthropic's valuation reaches two trillion dollars https://startupfortune.com/investors-are-already-pricing-anthropic-as-a-2-trillion-company/ - Claude maker public market debut expectations https://startupfortune.com/investors-are-already-pricing-anthropic-as-a-2-trillion-company/ Join the discussion Open in the community → /community/ Almost there. Sign in and your reply posts straight away.