India is preparing to open the world's largest retail fast-payment network to autonomous AI agents, a move that would make UPI among the first national infrastructures purpose-built for agentic commerce. The proposed Unified Agent Protocol, expected to be unveiled at the Global Fintech Fest in Mumbai next week, would let software agents execute small transactions within user-defined limits, without per-transaction approval1,4,5. ANALYSIS The timing is deliberate: by layering agent capabilities onto a system that already processed 24.51 billion transactions worth 29.82 trillion rupees ($314.21 billion) in August alone, NPCI is positioning India's public digital infrastructure as the default rail for a category of commerce that private networks in the U.S., Europe, Singapore, and Australia are only beginning to address through proprietary protocols.
Why it matters
UPI is the world's largest retail fast-payment system by transaction volume, according to a 2025 IMF report. Google Pay and Walmart's PhonePe account for around three-fourths of monthly transaction volumes. ◆ Any protocol change at this scale does not merely add a feature; it redefines what kinds of economic actors can participate in the network. If the Unified Agent Protocol ships as described, verified software agents would join the roughly half-billion human users already on UPI, creating a transaction layer where machines initiate payments on behalf of people under pre-set constraints.
The big picture
The proposed framework is expected to build on two existing UPI capabilities2,3. UPI Circle lets a primary account holder delegate payment authority to a secondary user, a mechanism that could be adapted to designate a verified AI agent. Reserve Pay lets customers block funds for multiple future debits; banks currently cap such blocks at 10,000 rupees ($105.44) for up to 90 days, though both the amount and validity period could be revised.
Low-value, frequent purchases such as groceries are likely to be among the first use cases, with e-commerce platforms well-placed to capture early demand. More advanced applications could follow: AI agents identifying sale offers and discounts, placing orders based on user-defined instructions, and potentially making investments when specified price thresholds are met. NPCI is also expected to provide infrastructure that merchants can integrate directly.
The safeguards reportedly under consideration include spending limits, identity checks, and audit trails. NPCI plans to build a liability framework governing who bears responsibility when an agent-initiated transaction goes wrong. ANALYSIS That liability question is the hardest design problem in the protocol: delegating payment authority to software that acts autonomously requires a clear chain of accountability that existing consumer-protection rules, written for human-initiated transactions, do not contemplate.
Between the lines
NPCI is not operating in a vacuum. Mastercard completed its first authenticated agentic transaction in New Delhi in June. Visa is separately building agentic capabilities for payments in India. Earlier this year, fintech firm Pine Labs launched its own agentic protocol, P3P, letting AI agents complete UPI payments after a single upfront authorization. And deep-tech startup TRUSTNOW unveiled TRUSTNOW.EKAM, described as India's first sovereign AI governance platform for autonomous enterprise agents, offering air-gapped security, real-time runtime guardrails, and a kill-switch for agent actions6.
ANALYSIS The convergence of a public payment protocol, private card-network experiments, a fintech-built agent protocol, and a governance platform all arriving within the same quarter suggests that the Indian ecosystem is treating agentic commerce not as a research agenda but as near-term infrastructure. NPCI's decision to embed agent capabilities at the rail level, rather than leaving them to application-layer intermediaries, follows the same pattern India used with UPI itself: build the public standard, then let private players compete on top of it.
The TRUSTNOW.EKAM launch is particularly telling in context. Its governance layer, designed to identify every AI agent, define its authority, and intervene before high-risk actions execute, addresses precisely the trust gap that a protocol like UAP would create. The fact that a startup is already shipping enterprise-grade agent governance tooling indicates that the market expects autonomous agents to be transacting at scale soon enough to justify the investment now.
What's next
The Global Fintech Fest in Mumbai next week is the earliest window for a formal NPCI unveiling of the Unified Agent Protocol. NPCI has not confirmed the timeline or detailed specifications. Several operational and liability provisions remain undisclosed. ANALYSIS What matters most in the next disclosure is not the transaction-size cap or the grocery use case but the liability framework: whether NPCI assigns fault to the user who set the rules, the agent developer whose software acted on them, or the bank that processed the debit will determine how aggressively the ecosystem builds on the new rail.