API Security

API Attacks on BFSI and NBFCs: What’s Happening and How Indusface Helps

8 min read Updated

In October 2026, attackers moved crores out of Indian NBFCs in under two hours each. They did not break encryption or phish customers. They called the lenders’ own disbursal APIs, and the APIs paid out. Within a month, CERT-In had issued two advisories and the RBI had begun asking banks to audit their API links with NBFC and fintech partners.

Our CEO’s companion post covers what this means for your vulnerability programme and the conversation with your board. This post is for the people who run the stack. I’ll walk through the attack chain stage by stage, show the requests involved, and be specific about which stages AppTrana stops at the edge and which ones stay with your own controls.

Date (2026) What happened Impact
5 Oct Malware on a Delhi-area NBFC’s server took over its loan-disbursal account (Ministry of Cyber Affairs) Rs 3.93 crore in 149 debits
8–9 Oct Two Mumbai NBFCs reportedly hit through the API used for loan-disbursal instructions (Elets BFSI) About USD 2 million in 700+ transfers
9 Oct CERT-In warns NBFCs and fintechs that payment and disbursement APIs are being targeted (Digital Terminal) Sector-wide alert
19 Oct A Hyderabad NBFC’s systems and API were accessed without authorisation (Siasat) Rs 3.3 crore in 183 transfers to 52 accounts, in 87 minutes
19 Oct RBI advisory: treat environments as compromised until forensically cleared after any credential exposure (Elets BFSI) Raises the bar for incident response
3 Nov RBI advisory reportedly requires banks to have NBFC API clients audited by CERT-In-empanelled auditors (MediaNama) Existing clients have until 3 Feb 2027
4 Nov CERT-In names CRAFTY FALCON, an AI-assisted campaign that re-attacks firms after remediation (CERT-In CIAD-2026-0058) Escalates the 9 Oct alert

Every incident has the same three ingredients: speed, automation, and an API that trusted whoever was calling it.

How the Attacks Work

Trace a loan through a modern lending stack. The approval in the mobile app becomes a disbursal instruction, sent over an API to a partner bank or payment aggregator, often with no human in between. Whoever can make that call can move the money.

CERT-In’s advisory CIAD-2026-0046 describes a six-stage chain. The executive post compresses it into two steps: getting in and paying out. The six are worth keeping separate here, because each one is stopped by a different control. The first is almost always an internet-facing weakness.

Stage 1

Initial access

Attackers exploit unpatched middleware, misconfigured app servers, or deprecated and unauthenticated APIs, some allowing unrestricted file upload.

Stage 2

Foothold

They drop web shells and backdoors, then map payment databases and API integrations.

Stage 3

Credential theft

They harvest payment aggregator, gateway and partner-bank API keys from code, config files and databases.

Stage 4

Evasion

They delete or truncate logs and tamper with endpoint protection.

Stage 5

Fraud

They call disbursal and fund-transfer APIs from the lender’s own whitelisted servers, skipping loan creation, KYC, beneficiary and limit checks.

Stage 6

Cash-out

Hundreds of IMPS and NEFT transfers with dummy reference IDs land in mule accounts and are withdrawn at once.

 

If you track findings against the OWASP API Security Top 10, this is where each weakness lands:

Weakness OWASP API 2023 Where it shows up
Deprecated, undocumented or test APIs left online API9 Improper Inventory Management Initial access
Unauthenticated or weakly authenticated endpoints API2 Broken Authentication Initial access, fraud
Disbursal calls accepted without a matching approved loan API6 Unrestricted Access to Sensitive Business Flows Fraud
Callers acting on records they do not own API1 Broken Object Level Authorization Fraud, data theft
No per-transaction or velocity limits API4 Unrestricted Resource Consumption Fraud, cash-out
Unpatched middleware and file-upload flaws API8 Security Misconfiguration Initial access

How Indusface Helps

The cheapest place to stop this chain is stage 1: the exposed app, API or middleware the attacker walks in through. AppTrana is built to close that door and to make abuse of what stays open visible and blockable. It runs as one loop: Discover, Scan, Protect, Monitor. The executive post makes the case for the loop, so I’ll go straight to what each step does.

Discover: know every API you expose

  • Continuous API discovery finds documented, shadow and zombie APIs from live traffic, including old app versions and retired partner endpoints, and flags the ones that carry PII, account or loan data.
  • East-West discovery (launching mid-November 2026) will use lightweight sensors on your app servers to map internal APIs and outbound partner calls, showing which services can reach disbursal and payment interfaces, the path the fraudulent calls took in stage 5.

Scan: find what an attacker would find

  • AI-powered DAST and API scanning test for the OWASP Top 10 and OWASP API Top 10, including broken authentication and misconfiguration.
  • AI pentesting, validated by experts, goes beyond signature scans. AI agents chain steps and probe authorisation and business logic the way an attacker would. In customer engagements, AI pentesting finds 2 to 3 times more vulnerabilities than traditional manual pentesting.

Protect: patch at the edge, fix in the pipeline

  • Block mode from day zero stops exploit attempts against middleware, file-upload flaws and web shell drops at the edge.
  • SwyftComply AI turns every finding from scanning and AI pentesting into an app-specific custom policy at the platform edge. AI generates and tests each policy for false positives, and experts verify it before enforcement. Critical findings are protected in hours and the full backlog in days, not months, while code fixes follow the normal release cycle.
  • Zero-day coverage within hours of a critical CVE disclosure, verified by experts.
  • API positive security enforces your API schema, so calls to undocumented endpoints or with unexpected parameters are rejected, while rate limits catch bursts of automated calls.

Monitor: experts watching 24×7

  • A 24×7 managed security team watches for attack surges, abuse patterns and false positives, and tunes protection as your apps change.
  • When a campaign like the October NBFC attacks breaks, the same team checks customer applications for exposure and pushes protections out without waiting to be asked.

In practice: stopping BOLA and business logic abuse

Signature-based WAFs miss these attacks because every request looks valid. The payload is clean and the caller is authenticated. What is wrong is who is asking for what, and in what order. You find that by reasoning about the application, not by matching signatures. Here are four patterns we see in lending APIs, with how we find and block each one.

1. Reading another borrower’s loan (BOLA)

A logged-in borrower changes one number in the URL and gets someone else’s loan statement, PAN and bank details.
GET /api/v2/loans/LN-1048821/statement
Authorization: Bearer <token for customer C-20931>

  • Find: AI pentesting replays every object ID with tokens from different users to check whether the API enforces ownership, at a scale manual testing cannot match.
  • Block: Where the request carries the caller’s identity, for example a customer ID in the token and in the URL, a SwyftComply AI custom policy rejects mismatched requests at the edge.
  • Detect: Elsewhere, behavioural detection blocks enumeration, such as one session requesting many or sequential loan IDs, until the ownership check ships in code.

2. Disbursal with no approved loan behind it

An attacker calls the disbursal API directly with a made-up reference ID. The API pays out because it never checks for a sanctioned loan. This is the core pattern in CERT-In’s October advisory.
POST /api/disburse
{"loan_ref": "TMP-000000", "amount": 195000, "beneficiary": "XXXX4471"}

  • Find: AI pentesting replays disbursal calls with invalid, reused and out-of-sequence references.
  • Block: Positive security enforces the API schema, so reference IDs that do not match the expected format, or fields the API should never accept, are rejected at the edge.
  • Detect: Velocity rules flag bursts of disbursal calls, new beneficiaries and repeated amount bands, the 183-transfers-in-87-minutes pattern.

3. Changing the amount or beneficiary in flight (mass assignment)

The app sends a loan-acceptance request. The attacker adds fields the client was never meant to set.
PUT /api/loans/LN-1048821/accept
{"accepted": true, "sanctioned_amount": 900000, "disbursal_account": "XXXX9902"}

  • Find: Scanning and AI pentesting test whether the API accepts extra or overridden fields.
  • Block: Schema enforcement drops requests with parameters outside the documented contract. A targeted SwyftComply AI custom policy can lock specific fields, such as amount and beneficiary, on endpoints that should never change them.

4. Skipping a step in the loan journey

A loan flow should run KYC, then sanction, then e-mandate, then disbursal. An attacker calls the final step directly.

  • Find: AI pentesting maps the journey and tries each step out of order, the kind of flaw scanners cannot see.
  • Block: Once the finding is confirmed, a SwyftComply AI custom policy blocks the out-of-sequence call at the edge.

In all four, the edge policy buys you time. The ownership check, the loan lookup and the sequence check still need to land in your code.

What this covers, and what it does not

Attack stage Where AppTrana helps Controls you still own
1. Initial access Discovery, scanning, virtual patching, edge blocking of exploits and unsafe uploads Patch management, secure configuration
2. Foothold Blocks web shell uploads and command injection at the edge EDR on servers
3. Credential theft East-West map of which services hold and use partner API access (from mid-November 2026) Secrets vault or HSM, key rotation, no hard-coded keys
4. Evasion Edge logs held outside your servers, out of the attacker’s reach Tamper-proof central logging, 180-day retention
5. Fraud via API Positive security and rate limits on APIs AppTrana fronts; AI pentesting for logic flaws, protected by SwyftComply AI Maker-checker, matching every payout to an approved loan, per-day limits
6. Cash-out Velocity alerts on APIs AppTrana fronts Beneficiary validation, real-time reconciliation, outbound allow-listing

No single product stops this attack chain on its own. AppTrana removes the easiest way in and gives you visibility and enforcement at the API layer. Your fund-flow controls remain the last line of defence.

The Evidence Your Team Will Be Asked For

According to Elets BFSI, the RBI is asking banks to prove that cyber directives are implemented, not just documented, and to probe weaknesses in their API links with NBFC and fintech partners. For an NBFC on the other end of those links, that comes down to evidence that controls work, not policy documents. The readiness check below lists the specific asks.

  • RBI: Evidence that controls work, not policy documents. Expect to show audit results for partner APIs, a plan for AI-driven attacks, and a process for treating an environment as compromised after any credential exposure.
  • CERT-In: An API inventory you can produce on request, proof that unused APIs are actually removed, logs retained for at least 180 days, and prompt incident reporting.
  • DPDP Act: Loan and KYC data exposed through an API is a personal data breach with its own notification duties, so you need to know which endpoints carry that data before an incident, not after.

AppTrana supports these asks with a live API inventory, AI pentest and SwyftComply AI reports that auditors accept, and protection status you can show in a board or regulator review.

A 10-Point Readiness Check for BFSI Security and DevOps Teams

The executive post has five questions for leadership. These are the checks underneath them, written so your team can answer each one with evidence.

  1. Can we export a current inventory of every internet-facing API, including old app versions and test endpoints, and does it match what live traffic shows?
  2. Is every API that no partner or app still uses actually switched off, not just dropped from the documentation?
  3. Is every disbursal and payment API authenticated per session, with no long-lived static tokens?
  4. Are partner-bank and aggregator keys stored in a vault or HSM and rotated on a schedule, with none left in source code or config files?
  5. Does the disbursal API reject a call with no approved, sanctioned loan record behind it, even from a whitelisted server?
  6. Are per-transaction, per-day and velocity limits enforced at the API layer?
  7. Have our APIs been pentested for business-logic and BOLA flaws in the last 12 months?
  8. For the last critical CVE that affected us, how long did it take to get protection in place: hours or weeks?
  9. Are logs stored centrally, out of reach of a compromised server, for at least 180 days?
  10. Would someone notice 100 transfers in 30 minutes, at 10:30 on a weekday morning?

Every “not sure” is a work item. If you have more than two, you have the gaps the October attacks exploited.

Proof of Concept

Indusface protects 850+ customers in 45+ countries, including banks, NBFCs and insurers across India. If you would rather test this on your own applications than take my word for it, a POC includes managed onboarding with block mode from day zero, API discovery from your live traffic, and SwyftComply AI custom policies on your findings.

Ask your Indusface account manager to set up a POC. – Manage your POC →

Sources

Vivek Gopalan

Vivek Gopalan is the Chief Product Officer at Indusface. With 18 years of experience designing and building technology products, he has a keen eye for solutions that solve real-life problems. At Indusface, Vivek leads product direction across AppTrana and the Web Application Scanner, and drives the company's AI initiatives, bringing AI into how applications and APIs are discovered, tested, and protected. Support and Services team also report to him, so that what Indusface builds and how it serves customers come together as a single experience.