India has quietly launched a pilot where users can shop and make payments directly through ChatGPT, in partnership with the National Payments Corporation of India (NPCI) and Razorpay. This “agentic payments” test is built on new NPCI protocols—Reserve Pay (where funds are pre-blocked for future debits) and UPI Circle (which delegates UPI authentication)—so that users don’t have to leave the chat interface during checkout. NPCI and Razorpay say OpenAI leads the rollout, with integrations planned for Google‘s Gemini and Anthropic‘s Claude. The pilot includes merchant partners like BigBasket and Vi, and banking backing from Axis Bank and Airtel Payments Bank, while OpenAI has confirmed it won’t access raw payment data. The move signals a push to embed conversational AI more deeply into India’s digital payments infrastructure.
Sources: Reuters, Times of India
Key Takeaways
– India is experimenting with AI-driven commerce, letting users shop and pay seamlessly through ChatGPT using the UPI infrastructure and new NPCI protocols.
– The pilot caps intrusiveness: OpenAI and AI models won’t access raw payment data, and users must pre-authorize transactions with authentication.
– NPCI leadership warns of systemic risk if too much AI infrastructure and control consolidate among a few global tech firms.
In-Depth
Over the past decade, India’s Unified Payments Interface (UPI) has become a bedrock of digital payments, processing over 20 billion transactions per month and routinely touted as one of the world’s most efficient real-time payment systems. In that context, the new pilot to integrate conversational AI and payments marks a bold attempt to leapfrog from “payments via apps” to “payments via dialogue.”
At its core, the pilot combines OpenAI’s ChatGPT with NPCI (India’s payments regulator) and fintech firm Razorpay. The innovation is largely enabled by two new NPCI initiatives: Reserve Pay, which allows users to earmark funds in advance for future debits to known merchants, and UPI Circle, which lets AI agents delegate authentication under controlled parameters. This means users might browse, select, and pay for groceries or mobile recharges entirely within a chat interface—without switching apps or copying payment links.
Initial merchant partners include BigBasket for grocery orders and telecom firm Vi for recharge plans. On the banking side, Axis Bank and Airtel Payments Bank are supporting the backend infrastructure. According to participants, OpenAI will not have access to raw payment data—transactions are handled through pre-authorized tokens and two-factor authentication. Google’s Gemini and Anthropic’s Claude are expected to join the pilot soon, expanding the AI ecosystem participating in the experiment.
From a strategic vantage, this move is both opportunistic and risky. On one hand, embedding AI into payments could dramatically simplify buyer journeys, reduce friction in commerce, and lock in user engagement. On the other hand, having AI firms interface directly with payment rails raises concerns over data privacy, control, and dependency. NPCI’s own chairman recently voiced unease about too much AI infrastructure being consolidated in the hands of a few global entities, warning of risks to financial stability, sovereignty, and economic resilience.
The success of this pilot will hinge on both user trust and operational robustness. It must contend with potential errors, fraud, liability in failed transactions, and resistance from regulators wary of handing over too much decision power to autonomous agents. But if it works, it could redefine what “checkout” means—not as a separate step, but as a natural part of conversation.