- Delhi Police told The Indian Express that 12 NBFCs in Delhi-NCR lost over Rs 65 crore in three months, after hackers diverted approved but undisbursed loans.
- Police say the fraudsters sent payout requests from rented cloud servers whose addresses the lenders' firewalls already trusted. Police say they found traces of AI use.
- Leaders should ask whether a firewall trusts an address range or a verified identity, and how quickly they would spot a changed payout account.
A guard who checks the address, not the person
Many security systems trust a place instead of a person. If a request comes from the right address, it passes. Delhi Police say fraudsters have exploited that habit at non-banking financial companies (NBFCs), the lenders that sit alongside banks in India.
The Indian Express learned that a dozen NBFCs in Delhi-NCR reported losses above Rs 65 crore over three months. The money was not customer savings. According to a senior police officer, hackers located loans that had been approved but not yet paid out, and sent the money elsewhere.
What police say happened
A senior Delhi Police officer said these 12 cases sit within more than 150 complaints from NBFCs nationwide in the same period. Losses across all of them exceed Rs 300 crore.
Police have made 20 arrests in the Delhi-NCR cases. According to the officer, each one concerns mule accounts, the accounts that take in stolen money. Officers say Rs 18 crore of the total was caught and seized while still sitting in those accounts, before it was withdrawn or turned into crypto.
The officer called this a new pattern. Earlier fraud tricked customers into so-called digital arrests. Here, the lender's own systems were the target.
How the trick works
Police gave this account. A lender typically contracts a fintech firm to run its website from a data centre of a major cloud provider, with Amazon Web Services named as one example. The lender's firewall then keeps an approved list, a "whitelist", of the outbound addresses of those data centres. A payout request is accepted only from one of them.
The weakness is that anyone can open an account on those same cloud services. According to the officer, fraudsters register accounts and use addresses from the trusted range. Malware in the lender's app front end stays for seven to eight days. It shows which commands set off a payout. The fraudsters then write their own command and swap the loan applicant's account number for a mule account.
The request reaches the payout interface, passes the firewall and goes on to the bank. The officer said it gets instant approval. To the lender's systems, it looks routine.
This is a police description relayed by a newspaper. The Indian Express report gives no technical documentation, and the account rests on unnamed police officers. Police describe many targets as small firms running software with weaknesses that let the malware in.
How large the payouts were
The Indian Express listed individual cases. Police said Hero FinCorp lost Rs 12 crore on August 17 in more than 300 transactions, using at least 165 addresses from the whitelist. Hero FinCorp declined to comment.
The list also includes a lender that filed a complaint on October 4. It put its loss at Rs 10.59 crore, spread over 1,131 transactions across three days. Counts like these show how many separate payouts were diverted. Police have not said AI was used in either case.
Where AI fits, and where it does not
DCP Gaurav Tyagi of Delhi Police's cyber fraud unit said: "We have detected traces of AI being used in these frauds." A senior officer added that fraudsters are using AI tools to work "faster and in bulk".
The one named example is an October case in which an NBFC lost Rs 3 crore. Investigators there found that a Chinese AI platform called CyberStrikeAI had been used to send command scripts.
Police did not say how much of any intrusion the AI performed. The word "traces" is the limit of the claim.
Here is our argument, not a police finding. If police are right that AI is being used faster and in bulk, the lesson is that attackers need no new flaw. A known gap, once found, can be worked repeatedly. A firewall that trusts a whole address range may struggle to tell one such request from another. That is our inference. The sources do not test it.
Slow to notice
A senior officer said many smaller NBFCs check loan status only weekly or fortnightly. Only then do they learn that a customer never received the money. Malware has days to work before anyone looks.
MediaNama's coverage adds the regulator's view. It reported that CERT-In, India's national cyber agency, issued an advisory on October 9. It described attackers stealing payment gateway and partner-bank API credentials, then using payout interfaces directly to bypass identity checks and per-transaction limits. A second advisory on October 15 warned of AI-native offensive tools being used against Indian entities.
MediaNama noted that the first advisory named no affected entities or threat actors and gave no incident counts or loss totals. The second named neither the tools nor the entities.
Questions to put to your team
Additional CP Sanjay Bhatia urged smaller NBFCs to follow the Delhi Police and CERT-In advisories and have their systems audited. Leaders can start with plainer questions.
Does our firewall trust an address range, or a verified caller? Could someone else rent an address in that range? Can a payout go out with a changed account number and no second check? How often do we reconcile approved loans against money actually paid? Will our hosting partner tell us when payout commands arrive in unusual bursts?
An approved address says where a request came from. It does not say who sent it.
Produced by the WebPulse Newsroom with AI assistance from the original reporting credited below, and checked against that source by our editorial review. How we use AI.
Original reporting: The Indian Express.





