A bank or NBFC learns that customer, applicant or loan-related information has appeared on an external platform or criminal forum. The exposed material might contain identity details, contact information, internal customer identifiers, lead references, application metadata or supporting documents — and at the outset, nobody knows whether it came from an in-house application, a lending or onboarding platform, a cloud repository, an employee account, an outsourced vendor, a direct selling agent, a fintech integration or an old, previously exported dataset.
That uncertainty is the real difficulty. The objective of the response is not to find one vulnerable server — it is to determine whose data was exposed, where it originated, how it left the controlled environment, who accessed it and what harm could follow. That requires a coordinated response across cybersecurity, fraud risk, privacy, legal, vendor management and regulatory compliance, run by one accountable owner with every material decision logged.
1. Fingerprint the exposed schema
Before anything else, work out which application, database, business process or vendor most likely produced the exposed records. Acquire the sample through a threat-intelligence provider, CERT-In or law enforcement — never over ordinary email — and handle it in an isolated forensic environment: hash every file, preserve the original as read-only evidence, and examine only verified working copies.
Then build a field-level inventory of the sample: column names and order, data types, date formats, identifier conventions, null-value handling, filenames, PDF metadata, logos and form versions, API field names, and any recurring formatting quirks. Compare that fingerprint against production databases, CRM and loan-origination systems, onboarding and KYC applications, document stores, analytics exports, decommissioned applications, and the equivalent records held by vendors, DSAs and fintech partners. Score each candidate system from confirmed down to excluded based on how many unique characteristics actually match.
The step is complete once every exposed field has been mapped to a system of record, a derived field, an external source, or an honestly unresolved category — not before.
2. Preserve evidence across the ecosystem
Issue a joint evidence-preservation notice — legal, compliance and cybersecurity together — to every internal team, application owner and external party that might hold relevant evidence, including subsidiaries, DSAs, loan service providers, fintech partners and cloud vendors. Start with a broad investigation window: at least 90 days before the earliest suspected exposure, or the full lifetime of the affected application version, whichever is longer, and widen it as evidence demands.
Preserve application and database logs, API gateway and WAF records, authentication and privileged-access logs, VPN and endpoint records, email and OAuth-consent logs, cloud control-plane logs, export histories, change-management records and backups — exported in native format, hashed at collection, and held in access-controlled, immutable storage with a documented chain of custody. Take system snapshots before rotating credentials or changing configurations wherever it is safe to do so: remediation must never accidentally erase the evidence needed to establish the cause.
RBI’s outsourcing directions already require regulated entities to maintain effective access to vendor data, records, logs and alerts, and to retain or integrate cloud-provider logs for incident handling — this step is largely about exercising a right the institution already has.
3. Review the complete loan and onboarding pathway
Architecture diagrams describe the intended flow; working sessions with operational staff and vendors describe what actually happens. Reconstruct the real journey from advertisement or lead generation through DSA or branch intake, KYC and credit assessment, underwriting, disbursement, servicing and eventual deletion — and at every step, record who holds a copy, where, for how long, and whether it ever leaves the system of record through a spreadsheet export, an email attachment or a vendor’s own support copy.
Pay particular attention to rejected and abandoned applications, bulk CSV exports, test environments, OCR and document-conversion services, and manual exception handling — these are the places uncontrolled copies of customer data tend to accumulate quietly. Any system, transfer or copy that doesn’t appear in the application inventory, vendor register or records of processing should be logged as a control exception, not quietly fixed and forgotten.
4. Hunt for bulk-access and extraction patterns
Turn the incident into a short list of testable hypotheses — sequential API enumeration, mass document download, a compromised mailbox, a stolen service-account credential, vendor-side extraction, or an older copy taken before current monitoring existed — and assign each an owner, the evidence needed, and a conclusion. Establish what normal activity actually looks like (daily record views, exports per user, API requests per identity) so detection is based on genuine deviation rather than one arbitrary threshold.
Then hunt for the patterns that tend to indicate extraction: sequential customer or lead identifiers in requests, unusually high successful-response rates, repeated CSV exports, access from unfamiliar locations or devices, new API keys or OAuth applications, disabled logging, or cloud-storage listing followed by bulk retrieval. A single query rarely tells the whole story — correlate it against logins, privilege changes and outbound transfers on a common timeline before drawing a conclusion.
5. Audit service accounts, privileged identities and APIs
Build an inventory of every privileged employee account, vendor administrator account, service account, API key, OAuth application and scheduled job that touches customer data, with an owner, purpose and credential age for each. Where practical, preserve authentication logs, active sessions and key-use history before rotating anything, then revoke suspicious sessions and rotate credentials in a sequence that doesn’t disrupt critical services.
Test whether each identity can read, search or export more than its role genuinely requires, and remove dormant accounts and excessive access outright. APIs handling customer information should carry strong authentication, object-level authorisation, short-lived tokens, rate limits, and complete request logging without unnecessarily logging sensitive content — no privileged or service identity should be left without a named owner and a reviewed scope.
6. Notify and support affected customers
Classify individuals as confirmed, probably, potentially or not affected, and when meaningful uncertainty remains, favour protective communication over waiting for perfect attribution. Notify only through channels the institution already has registered with the customer — authenticated banking apps, registered email or SMS, or direct branch contact — and never through a clickable shortened link or a request for a password, PIN or OTP.
Say plainly what happened, what information was involved, what is still unknown, what the institution has done, how the data could realistically be misused, and what the customer should do next. Stand up a dedicated helpline and trained support scripts, and resist the temptation to say “there is no risk” simply because passwords weren’t exposed — identity and contact details are enough to enable convincing fraud on their own.
India’s notified DPDP framework carries direct breach-notification obligations, and its substantive provisions follow the government’s own phased commencement schedule — building this notification capability now means the institution is ready regardless of exactly when each provision takes effect.
7. Apply enhanced fraud monitoring
Load confirmed and probable affected customers into fraud monitoring under an internal risk indicator that increases scrutiny without visibly labelling the customer or freezing their account outright. Apply closer attention to registered-number changes, SIM swaps, password resets, new beneficiaries, new loan applications and remote account recovery — and treat accurate knowledge of a customer’s date of birth, address or loan details as exactly what it is after an exposure: not proof of identity.
Depending on risk, step up verification with a callback to the registered number, a cooling period on sensitive profile changes, or branch verification, and keep enhanced monitoring in place until the fraud-risk function can show that incident-related risk has genuinely returned to normal — not on a fixed calendar date.
8. Minimise retained customer and applicant data
For every category of data the institution holds, record its business purpose, retention basis, system of record, minimum required fields and approved deletion method — starting with the highest-risk material: KYC documents, income documents, abandoned and rejected applications, exported customer lists, and anything a vendor holds as a duplicate. Incident-related legal holds take precedence over routine deletion, but once preservation requirements are satisfied, unnecessary copies should be deleted under an auditable process, not left to accumulate again.
9. Measure vendor controls using evidence
Certifications support assurance; they don’t substitute for testing the controls that actually matter for the outsourced service in question. For every DSA, loan service provider, fintech platform, KYC vendor, document processor and cloud provider in scope, request access-control lists, privileged-access reviews, penetration-test results, log-retention evidence, backup and restoration results, and a current subcontractor inventory — and confirm the contract actually gives the institution, its auditors and its regulators effective access to the relevant systems, records, logs and personnel, not just a right on paper.
Track every vendor finding centrally with a severity, an accountable owner, a due date and an escalation route for anything overdue. The regulated entity keeps the responsibility for protecting customer data even when the processing itself is outsourced.
10. Treat data-rich non-core systems as high-risk banking systems
System criticality should follow the harm the data can cause, not only whether the system moves money. A lead-management platform may never touch a transaction, but the identity and contact information it holds can enable convincing impersonation and fraudulent credit applications on its own — it deserves the same strong authentication, privileged-access management, segmentation, monitoring and board visibility as a core banking system, not a lighter standard because it sits outside the payments estate.
Request a walkthrough of Practitioner Toolkit →
Recommended activation timeline
CERT-In requires specified cyber incidents — including data breaches and data leaks — to be reported within six hours of noticing or being informed of them, so the first six hours matter disproportionately: appoint the incident commander, open the register, validate the alert without redistributing the exposed data, preserve anything at immediate risk of expiry, make the required CERT-In and sectoral reports, and instruct vendors to preserve their own evidence.
| Window | Priority actions |
|---|---|
| 0–6 hours | Incident commander appointed, register opened, CERT-In/sectoral reports made, vendor preservation notices issued, confirmed unauthorised access blocked. |
| 6–24 hours | First schema fingerprint, probable systems/vendors identified, logs preserved, fraud monitoring activated, holding statements prepared. |
| 24–72 hours | First data-flow reconstruction, credentials rotated after evidence preservation, risk-based notification begins, board committee briefed. |
| 3–7 days | Source confirmed or narrowed, affected-individual count validated, permanent containment implemented, independent validation of key findings commissioned. |
| Within 30 days | Root-cause analysis complete, customer remediation confirmed, vendor/retention/monitoring controls revised, residual risk reported to the board. |