RBI-supervised lenders are leaning harder on National Payments Corporation of India machine-learning monitors as official tallies show Unified Payments Interface incident volumes remain far above pre-pandemic levels, even after a modest dip in reported cases last fiscal year.

Parliament’s five-year table

In a Lok Sabha answer dated 15 December 2025, the government published annexed data on domestic UPI payment incidents: 4.07 lakh cases in 2021-22 involving ₹242 crore, rising to 13.42 lakh cases and ₹1,087 crore in 2023-24, then 12.64 lakh cases and ₹981 crore in 2024-25. For 2025-26, the table listed 10.64 lakh incidents and ₹805 crore through November—still large absolute numbers on a rail that now handles the majority of retail digital transactions. Ministers cited the figures while describing layered defenses, not to alarm users away from UPI, but to show why banks keep upgrading analytics.

The same answer outlined non-technical measures: device binding between a customer’s mobile number and handset, PIN-based two-factor authentication, daily limits, and restrictions on certain use cases. It also pointed to the National Cybercrime Reporting Portal and the 1930 helpline as reporting channels, alongside Department of Telecommunications tools such as the Chakshu complaint window for suspicious SMS and calls.

NPCI’s federated scoring

NPCI operates a shared monitoring stack that scores transactions and devices in real time, sending alerts to issuers and, where authorized, declining suspicious payments. The payments body has described the setup as federated: banks contribute customer-risk scores built from demographics and account history, while NPCI adds transaction velocity, device profiling, and cross-bank trend data. Public remarks from NPCI risk officials in 2025 emphasized reducing false positives so call centers are not flooded while still catching mule-account patterns early.

Real-time application programming interfaces also feed the Ministry of Home Affairs’ Indian Cyber Crime Coordination Centre, allowing banks to mark accounts that appear in multiple incident reports. That feedback loop helps models retrain when criminals shift from QR code tricks to remote-screen-sharing tactics. Issuers retain final authority on whether to block a payment, but switch-level declines are available when policies allow.

RBI’s parallel intelligence work

The Reserve Bank’s 2025 payment-system report notes that its innovation hub built Mulehunter.AI to help banks spot mule accounts quickly and is prototyping a Digital Payment Intelligence Platform to give remitting banks beneficiary risk scores before execution. Those projects complement—not replace—NPCI’s switch-level tools. Awareness campaigns through SMS, radio, and branch posters continue alongside technology, because many losses still start with social engineering rather than stolen credentials.

Together with device binding, PIN authentication, transaction limits, and telecom risk indicators now reaching wallets, the stack reflects a policy choice to fight digital financial crime with data sharing rather than throttling UPI growth. For account holders, the visible outcome remains SMS alerts, in-app warnings, and occasional blocked transfers when models disagree with a payment pattern—signals that the system is working even when headlines focus on rising incident counts.

What supervisors expect from banks

RBI inspection teams have increasingly asked lenders to document how they tune model thresholds, escalate alerts, and refund customers when errors occur. A false decline on payday can erode trust as quickly as a successful impersonation scheme, so banks pair analytics with human review desks for high-value exceptions. NPCI’s shared scores help smaller private banks that lack the data science benches of the largest issuers, leveling the field as UPI penetration reaches tier-two cities.

Parliament’s annex also reminds lawmakers that amount involved per incident can fall even when case counts stay high, suggesting faster reporting and smaller average ticket sizes rather than a single mega-heist trend. Supervisors read that as progress on detection speed, not complacency. The next policy tests will be whether beneficiary-name confirmation and telecom risk feeds reduce repeat victimization among seniors, a cohort regulators singled out in recent awareness scripts aired on All India Radio.