India recently unveiled an AI model that possesses a “vocabulary for how Indians transact”, however, recent reports suggest agents might be making payments soon.
India has reportedly started working on adding agentic payment capabilities to its Unified Payments Interface (UPI).
According to Reuters, the National Payments Corporation of India (NPCI) is developing a framework that would allow AI agents to spend money on users’ behalf while following instructions set by the customer.
UPI is the world’s largest real-time payments system, processing 23.66 billion transactions worth 29.88 lakh crore ($313.47bn) in July 2026. A vast number of transactions are day-to-day payments, such as shopping for food, fuel or sending money to a friend or relative.
If plans go ahead to launch the capabilities, it is believed that many of these transactions could be made by agents. The user would set rules around how much can be spent, what types of purchases can be made and the conditions that need to be met before a transaction is completed.
Reuters also reported that NPCI is considering agents that could respond to price changes or make investments when predefined conditions are met.
NPCI is reportedly looking to provide the framework to merchants, which would provide them with the ability to support agent-led transactions.
UPI has the foundations for this technology, with UPI Circle enabling a primary user to let another person make payments from their account with permission. Additionally, Reserve Pay allows money to be earmarked in advance for a series of future debits.
Fraud challenge of Agentic payments
These reports come days after Razorpay launched Vulcan, an AI foundation model made specifically for India’s payments ecosystem. Announced last month, Vulcan is used across Razorpay’s network to improve payment routing, detect fraud and personalise checkout experiences.
Razorpay said the model has developed “a vocabulary for how Indians transact”, helping it identify patterns across the almost four billion customer-to-merchant payments processed through its network annually.
Using AI to allow agents to initiate payments, however, introduces a plethora of fresh challenges.
Traditional anti-money laundering and fraud controls, including new technologies like Vulcan, are designed around human behaviour, with systems looking for changes from customer routines.
Garima Chaudhary, VP of Financial Crime and Compliance AI at Thetaray, told Payment Expert earlier this year that agentic payments challenge such assumptions.
“Agentic payments completely shift the assumptions AML systems are designed on,” she said. “Instead of thinking what a normal human behaviour is, we need to think about what a dedicated normal behaviour is.”
Agents can make payments without sleep cycles or spending routines and authorisation can also become less straightforward when a customer gives an AI permission to act within a wider mandate rather than approving each transaction.
This leads to questions around how institutions distinguish legitimate automation from a compromised agent and how consent can be shown if a payment is later disputed.
Chaudhary said the change “changes everything they have known about AML”, adding that institutions capable of managing agentic payments could gain an advantage in compliance and customer trust.
UPI builds its AI infrastructure
NPCI’s reported work on agentic payments follows an effort to embed AI within India’s payments infrastructure.
In February, NPCI announced a collaboration with NVIDIA to develop what it described as a “sovereign, payments-native AI foundation” for the country’s digital payments ecosystem.
The organisation plans to use NVIDIA Nemotron as part of developing AI models capable of operating within high-volume, real-time payment environments.
NPCI CTO Vishal Kanvaty said the initiative is a move away from individual AI use cases towards a “foundational and scalable AI layer” tailored around India’s regulatory, security and data sovereignty requirements.
AI is already being tested through NPCI’s UPI Help Assistant pilot, which uses a financial language model to support grievance handling.
NPCI has said its AI infrastructure could eventually support areas including trust frameworks, operational intelligence and grievance redressal and also provide technology that banks and fintechs can use.
