India plans UPI framework for AI agents to make low-value payments without each approval
The design is expected to use UPI Circle delegation and Reserve Pay blocked funds, with current Rs 10,000 and 90-day limits under review.
India is preparing a framework that would let AI agents initiate small payments on UPI without requiring the user to approve every transaction. If implemented, it would push “agentic commerce” onto one of the world’s highest-throughput retail payment rails, with key safeguards and caps still not publicly defined.
Key Takeaways
- India is preparing a framework that would allow AI agents to initiate small UPI payments without asking for user approval on every transaction.
- The proposed design is expected to rely on UPI Circle for delegated spend authority and Reserve Pay for pre-blocked funds that can be debited multiple times.
- Banks currently cap Reserve Pay-style blocked funds at Rs 10,000 ($105.44) for up to 90 days, and those parameters may be revisited for agentic use.
- UPI processed Rs 29.8 lakh crore of value in August, averaging 791 million transactions per day and Rs 96,205 crore in daily value, per NPCI data.
India Preps ‘Agentic’ Payments on UPI Without Per-Transaction Approval
India is set to develop a framework that would allow AI agents to make small digital payments on the Unified Payments Interface (UPI) without requiring user approval for every transaction. The plan would put agent-initiated payments onto a national payments network that already clears near-record volumes.
UPI is the country’s real-time payments network that lets users send money instantly between bank accounts via apps. It is operated by the National Payments Corporation of India (NPCI), and a 2025 IMF report described UPI as the world’s largest retail fast-payment system by transaction volume.
The immediate significance is scale, not novelty. Agentic payment setups exist in other markets, but putting them on UPI would make India one of the first countries with national infrastructure for agentic AI payments and potentially one of the largest networks running agent-initiated consumer payments.
How UPI Circle and Reserve Pay Could Become the Agent’s ‘Wallet’
The expected architecture does not require a new payments rail. It repurposes two existing UPI mechanisms to give an agent bounded autonomy: delegation for permissioning, and pre-blocked funds for settlement.
UPI Circle is the delegation layer. It allows a primary account holder to delegate payment authority to a secondary user, which the proposed framework would treat as an AI agent acting on the user’s instructions. In practice, that means the “who can spend” decision moves to setup time, with revocation and permission controls becoming the primary safety valve.
Reserve Pay is the funding layer. It lets customers block funds in advance so multiple debits can be made against that reserved amount, rather than requiring a fresh approval flow each time. For agentic payments, that looks like a constrained wallet: the agent can transact repeatedly, but only within the amount the user has already ring-fenced.
The source material also sketches early use cases that fit this model: low-value, frequent purchases like groceries, and e-commerce actions such as placing orders based on sale offers and discounts. It also cites agents making investments based on specified instructions around price thresholds, which would raise a different class of risk questions if it extends beyond simple payments.
The Real Constraint: Rs 10,000 Blocks, 90-Day Validity, and What Might Change
The current hard limit is the part traders should anchor on, because it defines what “agentic” can mean in practice on day one. Banks currently cap Reserve Pay-style blocked funds at Rs 10,000 ($105.44) for up to 90 days, though the report said the limit and validity period may be revisited for agentic use.
At Rs 10,000, the design naturally optimizes for frequent, low-ticket spending rather than large purchases. That is consistent with the cited early demand coming from e-commerce platforms and routine consumer categories. It also means the system’s first stress test is likely to be operational and behavioral, not balance-sheet sized: lots of small debits, lots of edge cases, and lots of customer support and dispute flow.
The open questions are where the risk concentrates. The framework has not yet published what qualifies as “small” or “low-value,” how delegation is granted and revoked in detail, or what safeguards apply when an agent initiates a debit that the user claims was unauthorized or erroneous. Those are not cosmetic policy choices. They determine whether the system behaves like a controlled autopay feature or like a broad authorization that is hard to unwind once it is live across apps and merchants.
UPI’s throughput is the backdrop that turns those parameters into market-relevant plumbing. NPCI data cited put UPI transaction value at Rs 29.8 lakh crore in August, near record-high levels, with an average of 791 million transactions per day and average daily value of Rs 96,205 crore. Even a narrow “low-value only” agentic slice becomes meaningful if it is deployed widely, because small percentages of UPI volume still imply very large absolute counts of autonomous payment decisions.
What Traders Should Watch as UPI Moves Toward Agentic Commerce
The first signal is an official publication of the framework or implementation notes from NPCI and banks that pin down what transaction size qualifies as “small” for agent-initiated UPI payments. Without that threshold, the market is trading a direction-of-travel narrative rather than a spec.
The second is whether Reserve Pay constraints change from the current Rs 10,000 cap and 90-day validity window. Raising the cap expands addressable commerce and increases the blast radius of mistakes or abuse. Tightening it would keep the feature closer to “smart autopay” than general-purpose agent spending.
Third, watch the eligibility and control details for UPI Circle-style delegation: who can be delegated to, how revocation works, and whether AI agents are explicitly recognized as delegates. The practical control plane is the product here, not the model.
Finally, the safeguards and liability model will decide adoption speed. If the rules clearly assign responsibility for unauthorized or erroneous agent-initiated debits and require strong authentication at setup, the system can scale with fewer trust shocks. If liability is ambiguous or remediation is slow, the first wave of disputes will become the real rollout gate.
My Take: UPI’s Volume Makes This a Stress Test for AI Payment Guardrails
The threshold that matters is not whether an AI agent can technically click “pay.” It is whether delegation plus pre-blocked funds can bound the failure modes tightly enough that the system survives UPI-scale throughput, where even rare edge cases become a daily event.
If the framework lands with a clear “low-value” definition, explicit delegation and revocation controls, and a liability model that does not leave users guessing, the setup starts to look structural rather than narrative-driven. If those pieces stay vague while caps expand, agentic commerce on UPI becomes less a payments upgrade and more a dispute-resolution load test for the entire stack.
