Explainer · India agentic payments
India's reported Unified Agentic Protocol is about letting AI agents spend on UPI without giving them an unlimited wallet.
India's National Payments Corporation is preparing a framework for AI-agent payments on UPI, according to Reuters reporting and public comments at Global Fintech Fest. The important design idea is not simply “AI can pay.” It is that an agent should be identifiable, bound to a user mandate, constrained by limits, observable after the transaction and accountable when something goes wrong.
Published 12 September 2026 · current status and source boundaries below
Reported / emerging
No public NPCI technical specification located as of publication
UPI first
Reporting says low-value, frequent UPI payments are the expected starting point
Registry layer
NPCI is reported to be building identity and monitoring for payment-capable AI agents
What is actually public?
Separate confirmed UPI building blocks from the reported new protocol
Reuters reported on 10 September that NPCI is building a registry to verify and monitor AI agents making transactions on UPI, as part of what the report calls the Unified Agentic Protocol. The report says the registry could later extend beyond UPI to cards, bill payments and other payment methods.
Separately, public NPCI documentation already shows that UPI has been moving toward bounded delegation. An October 2025 UPI Circle circular extended full delegation to software profiles including AI profiles for limited users, while NPCI's Reserve Pay product allows a user to reserve a defined amount and make multiple payments from that pool.
Those existing controls do not prove the final architecture of the new agentic framework. They do show that delegation, bounded funds and software identities are not starting from zero.
Likely control stack
Agentic payments need more than a payment credential
Agent identity
Know which software agent is acting, who operates it and whether it is permitted to transact.
User mandate
Bind the agent to a user-approved task, merchant/category scope, amount ceiling and expiry rather than broad account access.
Funding boundary
Use a bounded delegated or reserved-payment mechanism so the amount at risk is explicit before autonomous execution begins.
Transaction evidence
Preserve the instruction, merchant, amount, decision path and final payment state so disputes and retries can be reconciled.
Liability
Define what happens when an authorised agent makes the wrong purchase, exceeds intent, or encounters an ambiguous payment outcome.
Revocation
Let the user or payment provider pause or revoke authority without having to dismantle the rest of the agent workflow.
Why UPI Circle and Reserve Pay matter
They provide two primitives an autonomous buyer needs: delegated authority and bounded funds
UPI Circle lets a primary user delegate payment capability to a secondary user. NPCI's October 2025 addendum extended that idea to IoT devices and software profiles, explicitly naming AI Profiles for limited users under Full Delegation. The same circular still required explicit user action for IoT debit transactions, so it should not be read as evidence that unattended autonomous spending was already generally enabled.
Reserve Pay takes a different approach: the user reserves a defined amount from a supported funding source with a merchant, then multiple payments can draw from that bounded amount. For autonomous commerce, that kind of pre-bounded budget is useful because an agent can receive spending authority without receiving open-ended access to the account.
Reuters' September reporting says the planned agentic model could use these kinds of mechanisms to support small, frequent purchases under pre-set rules, with more complex conditional purchasing potentially following later.
UAP vs UPI Circle vs MCP vs x402
They solve different layers of the machine-commerce stack
| Surface | Primary contract | Best-known job | Current steward / ecosystem |
|---|---|---|---|
| Unified Agentic Protocol (reported) | UPI agent authorisation + payment governance | Registry, mandates, limits, auditability and liability boundaries | NPCI / India payments ecosystem |
| UPI Circle | Delegated UPI payment authority | Primary user delegates controlled payment capability | NPCI / UPI |
| UPI Reserve Pay | Reserved balance for multiple merchant debits | Pre-bounds available funds for a continuing transaction relationship | NPCI / UPI |
| MCP | Model/runtime ↔ tool contract | Tool discovery and invocation | Agent/tool ecosystem |
| x402 | HTTP resource payment | Machine-readable payment challenge and settlement | Open machine-payment ecosystem |
An Indian commerce agent could eventually use MCP to reach a merchant tool, a UPI agentic framework to prove authority and identity, and UPI itself to move funds. A machine-native API elsewhere might instead use x402. The useful architecture question is not which acronym wins; it is which system owns identity, mandate, payment, delivery and retry state.
zFinia interpretation
The important shift is from approving every payment to approving a constrained economic policy
Agentic commerce scales when a user can define what an agent may do in advance and the payment system can enforce those limits independently of the model. India's reported direction is strategically important because UPI already operates at national infrastructure scale. If NPCI combines agent identity, delegated authority, bounded funds and transaction evidence in one governed framework, it would turn the agent from a recommendation layer into a constrained economic actor without treating the model itself as the final security boundary.
Sources