If successful, this technology could empower Indian banking to intercept digital fraud before money disappears into an elaborate web of mule accounts.
According to officials with knowledge of the project, nearly a dozen banks are shortlisted in the pilot where each bank will run an NPCI-developed AI model locally on their own transaction data, generating a risk score for individual transactions that is then transmitted to NPCI.
Sitting atop India’s entire payments network, NPCI aggregates these signals from all banks simultaneously. When a transaction is flagged as suspicious, NPCI immediately tracks where the money is heading. If funds are moving abnormally fast through a chain of accounts, NPCI can detect the network effect in real time and trigger an alert or freeze before the money is fully dispersed, officials familiar with the matter said.
“The system, built on a federated learning architecture, addresses one of the most vexing challenges in combating digital payment fraud – that by the time a victim reports a fraudulent transfer, the money has already cascaded through multiple accounts across different banks and is effectively untraceable,” said an official familiar with the project. “NPCI’s AI model compresses that detection window from hours or days to seconds. The genius of the system lies in how it achieves network-wide visibility without compromising customer data privacy.”
NPCI did not respond to a request for comment.The AI tool is designed to read the classic signature of a mule network, where fraudsters rapidly layer stolen funds through multiple accounts to obscure their origin. Through this tool NPCI can trigger intervention before the funds are fully dispersed.
“If NPCI runs AI on top of the entire payment structure, it can identify networks – which means it can also move fast,” said a person familiar with the development.