India's NPCI Building AI Agents Registry to Enable Autonomous UPI Payments

NPCI Develops Registry for AI Agents on UPI

India's National Payments Corporation of India is building a registry to verify and monitor AI agents making transactions on the country's Unified Payments Interface, according to three sources involved in the discussions1

1

. The registry will form part of NPCI's planned Unified Agent Protocol and vet AI agents that make payments on behalf of users. This infrastructure aims to enable agentic payments on UPI, which processed 24.51 billion transactions worth Rs 29.82 lakh crore in August 20262

2

3

3

. The registry will initially vet AI agents making transactions on UPI and could later cover cards, bill payments and other payment methods, according to a senior industry executive1

1

.

Protocols for Authorizing Digital Agents

Ajay Kumar Choudhary, non-executive chairman and independent director of NPCI, revealed at the Global Fintech Fest 2026 that the payments infrastructure is examining protocols to identify and authorise digital agents within the UPI ecosystem while preserving interoperability, auditability and settlement finality2

2

. Choudhary emphasized a critical design constraint: AI agents may interpret intent, but the authoriser must verify identity, mandate, limits and consent3

3

. The architecture separates intent, authorisation and settlement—while an AI system could interpret what a user wants and determine how to execute it, the final settlement layer would remain rule-based, secure and legally final2

2

.

AI Agents Making Autonomous Transactions

Agentic payments on UPI would allow AI agents to make small, frequent digital payments without requiring users to approve each transaction, with groceries and other low-value purchases expected to be among the first applications1

1

. The system could eventually allow AI-driven financial agents to act on more complex instructions, such as buying products when discounts reach a specified level or making investments when prices cross set thresholds1

1

. The framework is expected to build on existing UPI capabilities such as UPI Circle, which allows a primary account holder to delegate payment authority, and Reserve Pay, which enables funds to be blocked for multiple debits2

2

.

Data-Localisation Rules Face New Challenge

Agentic payments could test India's data-localisation rules, according to Cashfree Payments cofounder Reeju Datta. Personal AI agents could store card credentials outside the country, challenging existing frameworks designed before autonomous agents emerged as a payment option4

4

. An individual could save card information in a vault connected to a personal agent even if the vault sits outside India, allowing the agent to transact with Indian merchants. Datta emphasized that payment providers need to be wary of users sharing card information out of ignorance, leading to potential leaks4

4

. To reduce this risk, Cashfree does not share sensitive payment data with third-party AI models and uses role-based access control to limit what an agent can see and do.

Liability for Unauthorized Payments Remains Critical

While agents would be indexed in the registry, elements like liability for wrong or unauthorised payments will need to be addressed through regulation1

1

. NPCI is expected to provide infrastructure for merchants to integrate with such systems, with spending limits, identity checks and audit trails forming part of the framework2

2

. A liability mechanism would be needed as responsibility shifts between users, agents, banks and merchants. Datta stated that the entity offering the agent has to be responsible for its actions, and merchants may want the system to check before issuing a refund or when money is moving out of their accounts4

4

.

Global Context for AI-Led Payment Infrastructures

India's efforts come as governments and payments companies elsewhere develop similar systems to identify and authenticate AI agents. China's Payment & Clearing Association has issued guidelines for AI-agent payments covering identification, security and risk controls, while Amazon, Google and Microsoft have developed agent registries for agentic commerce1

1

. Mastercard and Visa are developing agentic payment capabilities in India, while Pine Labs has launched an agentic protocol that allows AI agents to complete UPI payments after a single upfront authorisation2

2

. The scale of UPI makes this development particularly significant, as Google Pay and PhonePe account for roughly three-fourths of the monthly volume2

2

.

Sovereignty and Interoperability Concerns

Choudhary warned that dependence on a single national or commercial stack may over time reduce sovereign choices, particularly relevant given recent United States policy promoting an integrated American AI technology stack3

3

. His prescription was portability—the ability to move between suppliers without rebuilding. Banks and payment firms should design applications to run across multiple models and infrastructure providers wherever practicable. Interoperability must remain non-negotiable, he added, letting agents and tokenised platforms work across institutions without creating new walled gardens3

3

. For India, the challenge now is not simply allowing AI agents to make payments, but ensuring that autonomy remains bounded, accountable and auditable as UPI enters its second decade2

2

.