The National Payments Corporation of India (NPCI) is developing a Unified Agent Protocol (UAP) to enable trusted AI agents to make UPI payments, positioning India among the first countries to build national infrastructure for agentic commerce. According to reports from Business Standard, work is underway at NPCI to develop the proposed protocol in consultation with the industry, with the launch requiring regulatory approval from the Reserve Bank of India (RBI). The protocol is designed to create a trusted, common, interoperable infrastructure through which AI agents can be registered, verified, and authorized to transact across the UPI ecosystem without changing the underlying payment rails. As per Business Standard, UAP will not replace UPI but will act as a verification layer on top of the existing system, allowing trusted AI agents to securely interact with UPI while preserving its existing payment infrastructure. Currently, only users and authorized payment applications can initiate UPI transactions, with AI agents lacking the authority to independently execute payments even after helping users select products or services.
A year ago, agentic AI use cases across India's financial services and digital commerce sectors were primarily proof-of-concept experiments. Today, the industry has progressed closer to production deployment, but faces a critical infrastructure challenge. According to reports from Business Standard, the biggest obstacle to widespread AI agent adoption is no longer artificial intelligence itself, but the decades-old financial and commerce infrastructure designed for deterministic software and human interactions, not autonomous AI agents. This infrastructure transformation is becoming increasingly urgent as businesses move beyond basic automation towards Agentic AI, enabling autonomous systems to handle complex workflows and decision-making, promising Zero-Human Ops that free human talent for strategic tasks by automating mundane operations.
The focus on controls and preventive systems is particularly critical as e-commerce expects initial agentic AI use cases to center on price intelligence. According to Keshav Kumar, chief product and technology officer at BigBasket, agents acting on consumers' behalf will require clear consent and spending controls before adoption can move beyond pilots, which requires regulatory clarity rather than just better technology. Ishan Sharma, head of sales and business development at Juspay, emphasized that payments must remain deterministic, with agents capturing user intent and users themselves whitelisting their agents, requiring proactive controls to prevent unauthorized actions. As reported by Business Standard, the system is expected to work much like existing UPI features such as AutoPay and Reserve Pay, where users pre-authorise a spending limit instead of giving unrestricted access to their bank account. The proposed framework is expected to operate similarly, with users approving spending limits rather than granting unrestricted access to their bank accounts.
Companies are developing robust AI infrastructure solutions to address these challenges. Snapdeal has implemented an orchestration layer based on the Model Context Protocol (MCP), along with strong guardrails to minimize hallucinations and ensure reliable recommendations. As reported by Business Standard, MCPs are emerging standards that enable AI agents to securely interact with external software, retrieving data and executing tasks across different applications. The proposed Unified Agent Protocol (UAP) includes a trust layer for AI that verifies and authorizes AI agents on UPI, built on UPI Circle using delegated payments with spending limits, and preserves privacy by ensuring NPCI verifies transactions, not purchases. Technology service providers like Zeta are building intermediary software layers that sit between AI agents and core banking infrastructure, authenticating agents, verifying authorized access, and maintaining audit trails before information retrieval. The framework is expected to establish whether an AI agent has been authorized to act for a user, define the scope of that authority, and help determine accountability if those limits are exceeded.
AI agents for users could originate from merchant apps, consumer-facing payment apps, AI assistants such as ChatGPT or Claude, or dedicated agentic platforms such as Hermes and OpenClaw. According to Business Standard, a senior executive at a digital payments firm expects low-consideration, frequent purchases such as daily groceries, dairy, and similar repeat buys to be among the first automated as agentic transactions in India. The banking sector is implementing risk-based authentication approaches to address emerging fraud patterns, with the industry acknowledging that new fraud patterns are evolving in the AI ecosystem, requiring continued infrastructure development and authentication improvements. Industry participants emphasize that existing systems for user-initiated payments flows such as chargebacks and dispute management should be in place in the agentic world, with focus on regulation, audit trails, and dispute handling to solve concerns about controlling machines going rogue. The proposed payment flow could follow a simple sequence where users instruct AI assistants to perform tasks, AI agents compare options and generate payment requests, the protocol verifies the AI agent's registration and authorization, and the transaction is routed through UPI after user approval, with pre-authorized payments completed automatically subject to applicable rules.
