India's digital payments system, which has moved from PINs and OTPs to fingerprint and facial authentication, could be headed toward a new phase - hands-free transactions where AI agents recognise users, evaluate purchases and complete payments with little to no manual intervention.
According to a Reuters report, the government may soon permit AI agents to process small UPI transactions without requiring the user's approval for every single payment. The framework, called the Unified Agent Protocol, could reportedly be unveiled at the Global Fintech Fest in Mumbai next week. The National Payments Corporation of India (NPCI) has not formally confirmed the proposal.
What the framework would allow
If implemented as reported, the protocol would let users grant AI agents restricted payment authority based on pre-set conditions, replacing the need to approve each transaction individually. The early rollout is expected to centre on small, frequent purchases such as groceries, with e-commerce platforms likely among the first areas of adoption. Reuters reported that the scope could eventually widen to cover more complex transactions.
Per the report, AI agents could be capable of spotting deals and discounts and placing orders automatically based on instructions set by users. Further down the line, potential applications could include triggering investments once asset prices hit predetermined thresholds - though this would hinge on how the final framework is designed and regulated.
Built on existing UPI infrastructure
Rather than building an entirely new system, the reported framework would draw on tools already within UPI. This includes UPI Circle, which lets a user delegate payment authority to another party, and Reserve Pay, which allows funds to be earmarked in advance for specific payments. NPCI is also expected to build infrastructure that lets merchants integrate directly with the protocol, according to the report. Users would be able to define rules around how much an AI agent can spend and under what circumstances.
Safeguards under consideration
Per Reuters' sourcing, the safeguards being weighed include spending limits, identity verification checks and audit trails to track agent-led transactions. NPCI is also said to be working on a liability framework to determine accountability in case something goes wrong, though specifics have not been made public. Importantly, the design would not hand AI agents unrestricted control. Users would set the rules within which an agent can act, and the system would enforce those boundaries.
How other payment networks are approaching agentic commerce
India is not alone in pursuing this shift. Mastercard has built Agent Pay, an infrastructure layer meant to let AI agents transact while preserving controls around trust, security and visibility into transactions. Visa is developing similar infrastructure for agent-led commerce. Broadly, payment companies are working toward a model where an AI agent can search for products, make selections based on user instructions, and initiate the transaction - while authentication and other checks remain in place.
Within India, Mastercard has already run pilot agentic commerce transactions with businesses, and fintech firm Pine Labs is also reported to be building infrastructure to support AI-led payments.
What happens next
As of now, the Unified Agent Protocol remains a proposed framework. Its final shape - including transaction limits, safeguards, liability provisions and rollout timeline - will depend on what NPCI formally announces.
