The National Payments Corporation of India (NPCI) is quietly assembling the regulatory and technical scaffolding required to let software agents spend human money.
At the center of this push is a planned Unified Agentic Protocol (UAP) and a dedicated NPCI AI Agent Registry. This identity and security layer will verify, accredit, and monitor artificial intelligence agents initiating autonomous UPI payments on behalf of users. The goal is a system where software can execute small, frequent digital transactions—such as reordering milk when a smart fridge detects a shortage or purchasing flight tickets the moment prices drop below a pre-set threshold—without requiring a manual, per-transaction OTP or fingerprint scan.
The Architecture of Autonomous UPI Payments
The proposed registry acts as a gatekeeper. By establishing a formalized identity layer, the NPCI aims to ensure that AI-driven transactions remain bounded, accountable, and auditable.
Rather than building these capabilities from scratch, the UAP builds upon existing structural mechanisms within the Unified Payments Interface (UPI) framework:
- UPI Circle: Launched in August 2024, this delegated payments feature allows primary users to authorize secondary users (or, under the new protocol, non-human agents) to make transactions within defined limits.
- Reserve Pay: A mechanism capable of blocking funds up to ₹10,000 for up to 90 days, providing a financial sandbox within which an AI agent can operate without risking a user's entire bank balance.
This structural approach aligns with broader market trends in India, where infrastructure providers are racing to support autonomous systems. For instance, the sovereign-backed infrastructure of the IBM and Yotta Sovereign Agentic AI Platform is already positioning the country to host localized, high-security agentic workflows.
The Core Design Constraint: Deterministic Execution
Allowing non-human entities to touch real-time payment rails introduces massive security risks. Addressing these concerns at the Global Fintech Fest on September 10, 2026, Ajay Kumar Choudhary, non-executive chairman of NPCI, emphasized a strict separation of powers.
Under the UAP, decision-making and execution must remain entirely separate. An AI agent may analyze data and recommend a transaction, but the final authentication and settlement must follow deterministic, auditable rules. The AI cannot "negotiate" or alter the payment terms dynamically outside of the hardcoded boundaries set by the account holder.
This focus on containment mirrors global efforts to build guardrails around autonomous software, such as the Nvidia Open Agent Safety Platform, which seeks to prevent AI agents from escaping their designated software sandboxes.
Pilots and the Competitive Landscape
The NPCI is not the only player testing these waters. The race to capture AI payments in India has already drawn major fintech and technology companies into active pilots:
- Pine Labs "P3P": Launched in June 2026, this stands as India’s first agentic payment protocol built directly on top of the UPI network.
- Mastercard "Agent Pay": Demonstrated in February 2026 in partnership with Axis Bank, RBL Bank, Swiggy, and Zepto to test automated merchant settlements.
- Razorpay & OpenAI Pilot: A early-stage 2025 integration that tested ChatGPT-driven payment initiations within controlled merchant environments.
- Google Gemini & Flipkart: Google's ongoing testing of a direct checkout flow within its AI assistant, detailed in Google's Gemini-Flipkart checkout pilot, highlights how consumer-facing frontends are preparing for agentic shopping.
+-----------------------------------------------------------------+
| UNIFIED AGENTIC PROTOCOL (UAP) |
+-----------------------------------------------------------------+
|
v
+-----------------------------------------------------------------+
| NPCI AI AGENT REGISTRY |
| - Verifies Agent Identity |
| - Accredits Developer Credentials |
| - Monitors Transaction Behavior |
+-----------------------------------------------------------------+
|
v
+-----------------------------------------------------------------+
| UPI EXECUTION ENGINE |
| - UPI Circle (Delegated Limits) |
| - Reserve Pay (Blocked Funds up to ₹10,000) |
+-----------------------------------------------------------------+
Scale and Regulatory Hurdles
The stakes for this protocol are incredibly high. UPI processed 24.51 billion transactions worth ₹29.82 lakh crore in August 2026 alone, representing roughly 49% of global real-time payments volume. Introducing autonomous agents into a network of this scale requires extreme caution.
While the NPCI is actively developing the registry, the entire framework remains subject to formal regulatory approval from the Reserve Bank of India (RBI). The central bank will have the final say on transaction limits, liability frameworks for when an AI agent makes an unauthorized purchase, and the specific security protocols required to prevent automated API exploitation.
The registry will initially vet AI agents operating strictly on UPI before the NPCI considers expanding the protocol to credit cards, utility bill payments, and other retail payment methods.
