India is building an AI-agent registry for UPI. Liability is the unresolved part

Reuters reported on September 10 that the National Payments Corporation of India is developing a registry to identify, verify and monitor AI agents that could transact through UPI. The registry is being designed as part of a wider agentic-payments framework, with early use cases expected to focus on small, routine transactions. NPCI had not announced a final public rollout or complete rulebook at the time of the report.

The technological idea is straightforward: a user delegates limited authority to software, and the software can act within pre-set conditions. The harder question is accountability. If an agent misunderstands an instruction, is compromised, or makes an unauthorised payment, the payment system needs a clear chain of responsibility among the user, agent provider, bank, merchant and network.

For India, this matters because UPI is already a systemically important retail-payment rail. Adding autonomous software to that infrastructure creates a new layer of convenience, but also raises questions about consent, authentication, fraud control, data governance and consumer protection. The reported registry should therefore be read as infrastructure under development, not as a fully notified liability regime.