NPCI Chairman: AI Agents Should Not Approve Payments
Ajay Kumar Choudhary, chairman of India's National Payments Corporation (NPCI), stated at GFF 2026 that AI agents can identify user intent but should not be authorized to approve payments. Decision-making and execution must remain separate.

Ajay Kumar Choudhary, non-executive chairman of India's National Payments Corporation (NPCI), declared at the Global Fintech Fest 2026 in Mumbai on September 10, 2026, that artificial intelligence (AI) agents should identify user intent but must not be authorized to approve payments. Choudhary emphasized that decision-making and execution must remain separate, with AI recommending actions while authentication and final settlement adhere to deterministic, auditable rules.
The NPCI is exploring protocols to identify and authorize digital agents within the Unified Payments Interface (UPI) ecosystem, aiming to maintain interoperability, auditability, and settlement finality. UPI processed over 24.51 billion transactions in August 2026. Currently, no specific framework governs AI agents, although CERT-In has proposed human-in-the-loop controls for transactions exceeding defined financial thresholds.
Choudhary highlighted the need for payment infrastructure to remain deterministic and auditable, contrasting with the adaptive and probabilistic nature of AI. A sound architecture, he explained, separates intent recognition, authorization, and settlement. While an agent can read intent, verification of identity, mandate, and limits falls under the authorizer's responsibility.
Furthermore, Choudhary cautioned against over-reliance on a single AI stack or ecosystem, which could diminish national choice over time. He advocated for portability, enabling entities to switch providers without rebuilding, allowing businesses to adapt to changing cost, risk, or regulatory requirements.