Safeguard
Incident Analysis•

State Bank of India's dark web data exposure: what loan lead records teach Indian banks about data pipelines, CERT-In's six hours and the DPDP Act

SBI says a preliminary review of about 375,000 records found on the dark web identified 3,897 customer records with personal data, including CIF identifiers, lead IDs and loan lead metadata. The cause is not public. Here is what is confirmed, what CERT-In, the RBI and the DPDP Act expect, and the engineering work any bank or lender can start this week.

Safeguard Security Team
Updated 35 min read

State Bank of India's dark web data exposure: what loan lead records teach Indian banks about data pipelines, CERT-In's six hours and the DPDP Act

Short answer. On 21 October 2026, State Bank of India (SBI) told the BSE and NSE that it "has experienced an incident involving the loss of certain data." The bank said it received alerts on 16 October 2026, including from its threat intelligence service provider and from CERT-In, that customer data was available on the dark web. A preliminary analysis of about 375,000 records identified 3,897 customer records containing personally identifiable information: names, registered mobile numbers, dates of birth, gender, email and residential addresses, branch and Customer Information File (CIF) identifiers, lead IDs, and metadata tied to loan and lead records. SBI activated its incident response, engaged an independent CERT-In empanelled cybersecurity firm, and said there was no material impact on its core business operations. On the same day, the Internet Freedom Foundation (IFF) pointed out that SBI had told investors but not, as far as it could see, the affected customers, and asked for the breach notification duty in Section 8(6) of the Digital Personal Data Protection Act, 2023 to be brought into force immediately. Two days later SBI said it had contacted affected customers individually by email or WhatsApp.

How the data left its source has not been made public. SBI's filing names no cause, no attacker and no affected system. What the field list does tell an engineering team is where to look in its own estate: lead management, loan origination, customer relationship and partner systems, and every export, feed and file share that moves those records between them. This post separates what is confirmed from what was claimed, walks through the reporting clocks that apply to an Indian bank, and lays out the work a bank, NBFC or fintech lender can do now without waiting for anyone's forensic report.

At a glance

ItemWhat is on the recordSource
Alert received16 October 2026, from SBI's threat intelligence service provider and CERT-In, among othersPSU Connect, 21 Oct
First public flagMonitoring account CyberPulse56 posted the claim on X on 16 OctoberThe Hacker News, 21 Oct
Exchange disclosure21 October 2026, under Regulation 30 of the SEBI Listing Obligations and Disclosure Requirements Regulations, 2015, titled "Incident involving Data Loss"PSU Connect, IFF
Records reviewedAbout 375,000 (3.75 lakh)CRN Asia
Records with personal data3,897, in a preliminary analysisCRN Asia
Data fields named by SBINames, registered mobile numbers, dates of birth, gender, email addresses, residential addresses, branch and CIF identifiers, lead IDs, loan and lead record metadataThe Hacker News
Not listed in the filingAccount numbers, balances, passwords, one-time passcodes, card details, PAN or Aadhaar numbersThe Hacker News
Claimed by the sellerAbout 375,000 CSV rows and 2,722 PDF documents, posted on BreachForums under the handle SiriusimFree Press Journal, 21 Oct
ResponseIncident response and containment activated; independent CERT-In empanelled firm engagedPSU Connect
Business impact"No material impact" on core business operations, financial performance or business continuity, per SBIPSU Connect
Customer noticeIFF found none on 21 October; SBI said on 23 October it had contacted affected customers by email or WhatsAppIFF, Madhyamam, 23 Oct
CauseNot disclosedThe Hacker News

What SBI actually disclosed

The disclosure went to the exchanges because SBI is a listed company. Regulation 30 of the SEBI LODR Regulations requires a listed entity to disclose material events, and the filing, dated 21 October 2026 and signed by the bank's compliance and company secretary function, is the primary statement of record (PSU Connect).

Read closely, the filing says a few specific things and leaves a lot open.

It confirms data loss, not a system compromise. The language is "an incident involving the loss of certain data." The filing refers to "the alleged compromise" and does not say whether any SBI system was breached, which system the records came from, or whether the source was internal or a third party (The Hacker News; CRN Asia).

The numbers are preliminary and narrow. About 375,000 records were examined and 3,897 were identified as SBI customer records containing personal data. CRN Asia was careful to add that the 375,000 figure "should not be treated as 375,000 confirmed affected SBI customers." The Hacker News noted that SBI had not said whether the 3,897 records map to 3,897 different people (CRN Asia; The Hacker News).

The fields are identity and relationship data. Names, registered mobile numbers, dates of birth, gender, email and residential addresses identify a person. Branch and CIF identifiers tie that person to the bank's internal customer master. Lead IDs and loan and lead metadata tie them to a sales or credit process. The filing does not list account numbers, balances, credentials, OTPs or card data (The Hacker News).

The investigation scope is stated. The independent firm is to verify the authenticity and completeness of the data, determine the source and extent of the exposure, and assess risk to customers and the bank. SBI said it is implementing remedial and preventive measures and monitoring the situation under applicable regulatory requirements (PSU Connect).

The impact statement is about the bank. "No material impact" covers core operations, financial performance and business continuity. As The Hacker News observed, it says nothing about the risk to the customers whose details are in circulation (The Hacker News).

What the seller claimed, and why it is a different thing

On 16 October, the X account CyberPulse56 flagged a listing by an actor using the handle Siriusim, who claimed to have released SBI linked data: about 375,000 CSV rows and 2,722 PDF documents, described as personal information, loan applications, supporting documents and account opening forms (The Hacker News). The Free Press Journal reported that the listing was on BreachForums, a forum used to trade stolen data, and that very little is known about Siriusim, which has no known prior history (Free Press Journal).

The Free Press Journal examined four sample records posted with the listing. The phone numbers were valid, and in one case a WhatsApp profile photo matched the passport size photograph on a KYC document the actor had also posted. The samples were mostly loan related and contained KYC fields. That check could not establish whether the data came from SBI itself, a subsidiary, or a third-party service provider (The Hacker News; Free Press Journal).

SBI's filing does not name Siriusim, does not tie its analysis to the BreachForums listing, and does not say whether the 2,722 PDFs were examined (CRN Asia; The Hacker News). The 375,000 figure matches the seller's claim, which suggests SBI looked at the same set, but the bank has only confirmed what its own preliminary analysis found.

A seller's claim is marketing for a sale; a bank's filing is a careful legal statement. Neither tells you the root cause.

For what customers should watch for after the October exposure, see SBI Data Exposure October 2026: What Customers and Banks Should Know.

Timeline

Date (2026)Event
25 MarchRBI issues Advisory No. 3/2026 on Best Practices relating to Customer Data Protection, recommending that affected customers be notified (IFF)
27 JulyBank of Baroda files a stock exchange disclosure about a similar incident, citing "potential business email compromise" as the preliminary cause (The Hacker News)
16 OctoberCyberPulse56 flags the Siriusim claim on X. SBI receives alerts about suspected exposure of customer data on the dark web
19 OctoberIndia Today Tech reports SBI has not addressed the claim, as cited by IFF (IFF)
21 OctoberSBI files its Regulation 30 disclosure. IFF publishes its statement and sends a representation to SBI's Chairman, copied to the RBI Governor and CERT-In's Director General
23 OctoberSBI says it has contacted affected customers individually by email or WhatsApp and advises customers never to share an OTP, PIN or password (Madhyamam)
15 NovemberNo root cause has been published. The independent investigation is ongoing as far as public reporting shows

The first two rows are context from earlier in the year. Five days passed between the alert and the exchange filing, a point IFF raised directly (IFF).

What is not known

  • The source. SBI has not said which system held the records, whether that system belongs to the bank, a subsidiary or a service provider, or how the data left it.
  • The cause. No vulnerability, misconfiguration, credential theft, insider action or partner compromise has been named. Bank of Baroda's July filing named a preliminary cause; SBI's did not.
  • The PDFs. Whether the 2,722 documents the seller described are real, belong to SBI customers, and contain identity documents is unconfirmed. IFF has asked SBI to say whether Aadhaar, PAN or passport copies are involved (IFF).
  • The age of the data. No one has said when the records were created or last updated, or whether they come from an earlier incident.
  • The rest of the set. What the remaining rows of the 375,000 contain, and whether they relate to SBI at all, has not been said.

IFF has asked SBI to disclose, after the investigation, whether a vendor or direct selling agent was involved (IFF). That is a question, not a finding.

Reading the field list like an engineer

The field list is the most useful public fact for a security or platform team, because it describes the shape of a record rather than the path it took. Without suggesting anything about SBI's systems, here is where records with this shape usually live in an Indian retail lender's estate, and why each place deserves attention.

CIF identifier plus name, mobile, date of birth and address. This is the core customer master view. In most banks the authoritative copy sits in the core banking system, but derived copies spread widely: customer relationship management (CRM) platforms, analytics warehouses, marketing and campaign tools, contact centre applications, collections platforms, and the reporting extracts that feed all of them.

Lead IDs and loan lead metadata. Lead records are created long before an account exists. They come from website and app forms, branch walk ins, call centres, partner referrals, direct selling agents (DSAs), aggregators and co-lending or fintech partners. They move through lead management systems into loan origination systems (LOS), credit decisioning, document collection and disbursement. Every hand off is an API call, a batch file, an SFTP drop or a spreadsheet.

Loan applications and account opening forms (as claimed by the seller). If the PDF claim were accurate, it would point to document stores: the upload buckets behind an application journey, the document management system, a scanning and digitisation workflow, or a partner's case file. These are often object storage buckets with long retention and broad read access because many teams need to see documents during processing.

Two things follow for any lender, whether or not they ever apply to SBI.

First, the same record exists in many more places than the customer master. A lead created in a partner app may be copied into the partner's database, the bank's lead system, the LOS, the CRM, a data lake, a fraud screening service, a call centre dialler list and several analysts' laptops. Any one copy is enough to produce a dataset like the one described.

Second, lead and loan data often sits outside the controls designed for account data. Core banking gets the hardest access control and closest monitoring. Lead pipelines are built faster, by more teams, with more partners, yet carry nearly the same identity fields.

Why third parties and data pipelines are the exposure to plan for

Indian retail lending runs on partnerships: DSAs, digital lending and co-lending partners, outsourced contact centres and collections, fintech onboarding platforms, and IT service providers that build and run much of the stack. Each partnership is a data flow, and the common failure patterns need no sophisticated attacker:

  1. Over-sharing in feeds. A partner integration sends the full customer record when the partner needs a lead ID and a phone number.
  2. Long-lived exports. Monthly or ad hoc extracts for reconciliation, campaigns or audits land in shared drives, email attachments or buckets and are never deleted.
  3. Standing vendor access. Support engineers, implementation partners and outsourced operations staff keep accounts long after a project ends, often without phishing-resistant MFA.
  4. Shared credentials for machine access. One API key or SFTP account shared across environments and people, rarely rotated.
  5. Unpatched partner facing applications. Partner portals and lead capture applications built on frameworks and libraries with known vulnerabilities, exposed to the internet.
  6. Non-production copies of production data. Test and UAT environments loaded with real customer records, protected like test environments.
  7. Weak query monitoring. Bulk reads of customer tables or buckets that no one would notice until the data appears for sale.

The legal position in India makes it clear who carries the risk in each case. Section 8(1) of the DPDP Act says a Data Fiduciary is responsible for complying with the Act "irrespective of any agreement to the contrary" in respect of processing undertaken by it "or on its behalf by a Data Processor." Section 8(5) requires it to protect personal data in its possession or under its control, including data processed on its behalf, "by taking reasonable security safeguards to prevent personal data breach" (DPDP Act, 2023). The RBI's outsourcing rules say the same thing in banking terms: the regulated entity remains responsible for its customers, whoever runs the system (Legal 500 summary of the RBI IT outsourcing direction).

The reporting clocks an Indian bank works against

A bank that learns its customer data may be on a forum faces several clocks at once. They run to different recipients and were written for different purposes.

ObligationWho receives itDeadlineStatus in November 2026Source
CERT-In Directions under Section 70B(6) of the IT Act, 28 April 2022CERT-InWithin 6 hours of noticing an incident or having it brought to noticeIn forceCERT-In Directions
RBI cyber security framework for banksRBIInitial report within 2 to 6 hours, as summarised by IFFIn forceIFF
RBI Outsourcing of IT Services directionService provider to bank, so the bank can report to RBI within 6 hours of detectionWithout undue delayIn force since 1 October 2023Legal 500
SEBI LODR Regulation 30Stock exchanges and investorsAs prescribed for material eventsIn forcePSU Connect
RBI Advisory No. 3/2026Affected customersRecommends notice through multiple channels with guidance and helpline supportAdvisory, not bindingIFF
DPDP Act Section 8(6) with DPDP Rules 2025, Rule 7Data Protection Board and each affected Data PrincipalRule 7 sets intimation to the Board and each affected Data Principal, and a detailed report to the Board within 72 hours of becoming awareComes into force 13 May 2027IFF, DSCI brief on the DPDP Rules

CERT-In's six hours, read precisely

The CERT-In Directions are short, and the parts that matter for an incident like this are worth reading word for word.

  • The clock starts early. Direction (ii) requires covered entities, including any "body corporate", to report incidents listed in Annexure I "within 6 hours of noticing such incidents or being brought to notice about such incidents." An external alert counts. A threat intelligence feed telling you your data is on a forum starts the clock just as surely as your own detection does (CERT-In Directions).
  • Data leaks are explicitly in scope. Annexure I lists "Data Breach" and "Data Leak" among reportable incident types. You do not need to know whether you were breached to know that you have a data leak to report.
  • Logs must already exist. Direction (iv) requires entities to "enable logs of all their ICT systems and maintain them securely for a rolling period of 180 days," within Indian jurisdiction, and to provide them to CERT-In with an incident report or on direction.
  • Clocks must agree. Direction (i) requires ICT system clocks to synchronise with the NTP servers of the National Informatics Centre or National Physical Laboratory, or with servers traceable to them. When you reconstruct a data flow across a bank, a partner and a cloud provider, mismatched timestamps can cost you days.
  • Someone must be reachable. Direction (iii) requires a designated Point of Contact for CERT-In, and entities must provide information in the format and timeframe CERT-In specifies.

Six hours is not enough to investigate. It is enough to report what you know, if the facts are easy to pull. Most of the engineering work below is about making them easy to pull.

What the RBI expects of a bank's technology estate

The RBI's Master Direction on Information Technology Governance, Risk, Controls and Assurance Practices was issued on 7 November 2023 and took effect on 1 April 2024. Several of its paragraphs map directly onto the questions an incident like this raises (RBI Master Direction):

  • Classification. Paragraph 9(c) calls for identification and mapping of the security classification of information assets in terms of confidentiality, integrity and availability.
  • Vendor risk. Paragraph 10 requires vendor risk assessment and controls proportionate to risk, including concentration risk and supply chain risk.
  • Patching. Paragraph 13 requires documented change and patch management, with patches applied and reviewed "in a secure and timely manner."
  • Audit trails. Paragraph 15 requires audit and system logging for applications that access or affect critical or sensitive information, and monitoring of those logs to detect unauthorised activity.
  • Access. Paragraph 19 limits access to a valid business need and requires multi-factor authentication for privileged users of critical systems.
  • Testing. Paragraph 26(a) requires vulnerability assessment at least once every six months and penetration testing at least once every 12 months for critical systems and systems in the DMZ with a customer interface, with vulnerabilities fixed within a set timeframe.
  • Incident notice. Paragraph 27(d) requires entities to notify CERT-In and the RBI of incidents as per regulatory requirements and to escalate to the Board and senior management.

The Master Direction on Outsourcing of IT Services, issued 10 April 2023 and effective 1 October 2023, adds the third-party half. It applies to material outsourcing, including services whose compromise could materially affect customer information. It requires a Board approved outsourcing policy, contracts that give the bank rights to access information and to audit, data that the service provider keeps separable, and a service provider duty to report cyber incidents "without undue delay, such that the incident is reported by the RE to the RBI within 6 (six) hours of detection." The RBI may inspect service providers and their subcontractors (Legal 500).

Read together, the two directions describe the evidence a bank should be able to produce within hours: which assets hold sensitive customer data, which vendors can reach them, when they were last tested and patched, and what the logs show.

The DPDP Act: what applies now and what applies from May 2027

The DPDP Act was enacted on 11 August 2023 and the DPDP Rules were notified in November 2025 with a staggered commencement. The breach notification duty in Section 8(6), and Rule 7 that sets out how intimation is made, come into force on 13 May 2027 (IFF; DSCI). IFF's position is that no law in force today clearly requires a bank to tell affected customers, and that the duties currently in force run to regulators (IFF). IFF also reported that MeitY floated a proposal in January 2026 to shorten the commencement window, with no amendment notified.

When the duties do apply, the stakes are clear in the Act's Schedule. A breach of the Section 8(5) obligation to take reasonable security safeguards may attract a penalty of up to ₹250 crore. A breach of the Section 8(6) obligation to notify the Board or affected Data Principals may attract up to ₹200 crore (DPDP Act, 2023, The Schedule).

Rule 7, as summarised in DSCI's brief, sets a two step structure: intimation to the Board and to each affected Data Principal, and a detailed report to the Board within 72 hours of becoming aware of the breach. Rule 6 sets out reasonable security safeguards that include measures such as masking or obfuscation of personal data, access control, and continuity measures such as backups (DSCI).

For a lender in late 2026, May 2027 is six months away, and the inventory, contracts, logging and notice templates the Act requires take longer than that to build. RBI's March 2026 advisory already recommends customer notice. Building to the DPDP standard now is the low regret choice.

How attackers use this kind of data

Whatever its origin, a set of names, mobile numbers, dates of birth, addresses, branch details and loan context is ideal raw material for impersonation. IFF made exactly this point: callers can pose as bank officials and quote real details to win trust (IFF). The Hacker News described impersonation as the main risk in the listed categories (The Hacker News).

Common follow-on patterns include:

  • Loan themed calls and messages. "Your loan application needs a final KYC step" or "your pre-approved offer expires today", with a link to a fake form that harvests credentials or OTPs.
  • Fee scams. A demand for a processing or insurance fee to release a loan the customer did apply for, made credible by knowing the branch and the application.
  • Account takeover attempts. Using date of birth and address to pass weak identity checks at a contact centre or to support a SIM swap request.

SBI's response on 23 October included the standard advice never to share an OTP, PIN or password (Madhyamam). IFF's statement repeated that advice and pointed customers to SBI's helpline at 1800 1111 09 and the national cyber crime helpline at 1930 or cybercrime.gov.in for unauthorised transactions (IFF).

For a bank, a leak of identity and context data is a fraud operations event as much as a security event. Call scripts, call back rules and monitoring thresholds should change the day the data is confirmed.

What an engineering team can do now

None of this depends on knowing how SBI's data left its source. Each step lowers the chance of producing a dataset like this one and creates evidence regulators will ask for.

1. Build a data inventory keyed to the leaked field set

Start with the fields, not the systems. Take the field list from this incident (name, mobile number, date of birth, gender, email, address, branch, CIF, lead ID, loan or lead metadata, and KYC documents) and find every place in your estate that stores, processes or receives that combination.

A useful inventory record is small and specific. One entry per data store or data flow, owned by a named team:

# data-inventory/lead-management.yaml
asset: lead-management-db
owner: retail-lending-platform
environment: production
hosting: bank-managed-cloud-account
fields:
  - name
  - mobile_number
  - date_of_birth
  - residential_address
  - cif_id
  - lead_id
  - loan_product
classification: confidential-customer-pii
inbound_flows:
  - source: dsa-partner-portal
    method: https-api
  - source: website-loan-form
    method: https-api
outbound_flows:
  - target: loan-origination-system
    method: message-queue
  - target: analytics-warehouse
    method: nightly-batch
  - target: contact-centre-dialler
    method: sftp-csv-daily
vendor_access:
  - vendor: implementation-partner-a
    access: read-only-db-role
    mfa: phishing-resistant
    review_due: 2026-12-31
retention: 180-days-after-lead-closure
logging: query-audit-enabled

Three rules make the inventory useful in an incident:

  • Include flows, not only stores. Most exposure happens in transit or in copies. A nightly CSV to a dialler vendor is an asset.
  • Include non-production. If UAT holds masked data, write that down with the masking method. If it holds real data, write that down too, and fix it.
  • Map it to RBI classification. Paragraph 9(c) of the IT governance direction asks for security classification by confidentiality, integrity and availability. Use the same labels so the inventory doubles as compliance evidence.

2. Produce an SBOM for every system that holds or moves the data

Once you know which systems hold the field set, you need to know what software they are made of. A software bill of materials (SBOM) lists every component, direct and transitive, with versions. It turns "are we exposed to this vulnerability?" from a week of emails into a query.

Generate SBOMs from source and from built images, because they differ:

# From source, for a Java lead management service
cdxgen -t java -o sbom/lead-management.src.cdx.json ./lead-management

# From the container image actually deployed
syft registry.example.internal/lending/lead-management:2026.11.03 \
  -o cyclonedx-json=sbom/lead-management.image.cdx.json

# Match components against known vulnerabilities, fixed versions only
grype sbom:sbom/lead-management.image.cdx.json --only-fixed -o table

Then prioritise against what attackers are actually using. The CISA Known Exploited Vulnerabilities catalog is published as JSON and is easy to join against scanner output:

curl -s https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json \
  | jq -r '.vulnerabilities[].cveID' | sort -u > kev.txt

grype sbom:sbom/lead-management.image.cdx.json -o json \
  | jq -r '.matches[].vulnerability.id' | sort -u > found.txt

comm -12 kev.txt found.txt

Every line that comes out of that last command is a known exploited vulnerability in a system that holds customer identity data. Those go to the top of the queue.

Ask service providers for SBOMs of the systems they run for you. The RBI outsourcing direction already gives you information and audit rights; an SBOM is a precise way to use them.

3. Check reachability in customer and partner facing applications

A long CVE list is not a plan. Reachability analysis traces whether your code actually calls the vulnerable function on a path that untrusted input can reach.

Apply it first to applications that face customers or partners and touch the field set:

  • loan application and lead capture journeys on the web and in mobile app back ends,
  • DSA and partner portals,
  • document upload and retrieval services,
  • APIs that return customer or lead records by ID,
  • internal tools exposed through VPN or zero trust gateways to vendor staff.

Pair static reachability with dynamic testing of the running applications you own and are authorised to test. The RBI direction already requires vulnerability assessment every six months and penetration testing every year for critical and DMZ systems with a customer interface. Continuous dynamic scanning between those cycles catches regressions that a yearly test cannot.

Look especially for broken object level authorisation in APIs that take a lead ID, application ID or CIF as a parameter. If changing one ID in a request returns another customer's record, an attacker can enumerate the whole table without ever "breaching" anything in the traditional sense. This is a pattern worth testing for in any lending API, whatever the cause turns out to be in this case.

4. Review every vendor and partner path to the data

Use the inventory to list every external party with access to the field set, then answer four questions for each:

  1. What exactly can they reach? Database roles, API scopes, bucket policies, file shares, SFTP folders, and admin consoles.
  2. How do they authenticate? Named accounts with phishing-resistant MFA, or shared credentials and static keys?
  3. When was access last used and last reviewed? Accounts unused for 90 days should be disabled pending a reason.
  4. What would they tell you, and how fast? Check the contract for the incident reporting clause the RBI outsourcing direction requires, and test it with a tabletop exercise.

For machine access, look for long-lived credentials and rotate them. In most cloud environments, a simple listing of access keys and their age is a good start:

# AWS example: list IAM users with an active first access key and its last rotation date
aws iam generate-credential-report >/dev/null
aws iam get-credential-report --query Content --output text | base64 --decode \
  | awk -F, 'NR>1 && $9=="true" { print $1, $10 }'

For DSAs and lending partners, reduce what they receive. A partner that needs to call a lead needs a lead ID and a phone number for a defined period, not a date of birth, an address and a CIF. Change the integration mapping, not just the policy document.

5. Shrink and mark the data in your pipelines

Two engineering patterns cut both the chance and the impact of a leak like this.

Minimise and tokenise. Replace CIF identifiers with per-flow tokens in feeds that leave the core. Truncate or drop dates of birth where an age band will do. Mask addresses to PIN code for analytics. Rule 6 of the DPDP Rules names masking and obfuscation among reasonable safeguards (DSCI).

Seed canary records per flow. Insert a few synthetic records, unique to each outbound feed and partner, with phone numbers and email addresses you control. If one appears in a forum dataset or receives a call, you know which flow the data passed through. That turns "we do not know the source" into a narrow question within hours.

6. Make the six hour report possible

A six hour report needs facts that are already collected and easy to query.

  • Keep 180 days of logs in India for every system in the inventory, as Direction (iv) requires, including database query audit logs and object storage access logs, not just application logs.
  • Synchronise clocks to NIC or NPL traceable NTP sources, per Direction (i).
  • Alert on bulk reads. A query audit table makes this straightforward:
/* PostgreSQL with pgaudit or an equivalent audit table */
SELECT actor, client_addr, date_trunc('hour', logged_at) AS hour,
       count(*) AS select_statements,
       sum(rows_returned) AS rows_returned
FROM   audit.customer_data_access
WHERE  object_name IN ('customer_master', 'lead', 'loan_application')
AND    logged_at > now() - interval '24 hours'
GROUP  BY actor, client_addr, hour
HAVING sum(rows_returned) > 5000
ORDER  BY rows_returned DESC;
  • Watch object storage. Alert on large numbers of GET requests against document buckets from a single principal, and on any bucket policy change that grants public or cross account read.
  • Pre-write the report. Keep a template with the fields CERT-In and the RBI ask for, the Point of Contact details, and a decision log. Fill it in during a tabletop exercise each quarter.

Here is a realistic first six hours for a bank that receives an external alert:

Elapsed timeWork
0 to 1 hourLog the alert time. Open an incident. Pull the sample records and match fields against the data inventory to shortlist candidate systems and flows
1 to 2 hoursCheck canary records. Query bulk read alerts and access logs for shortlisted systems for the last 180 days. Notify the CISO and legal
2 to 4 hoursContain where cheap and safe: rotate partner credentials on shortlisted flows, suspend unused vendor accounts, tighten bucket policies. Ask shortlisted service providers to confirm or rule out within their contractual window
4 to 6 hoursFile the initial report to CERT-In and the RBI with what is known and what is not. Brief fraud operations and the contact centre with the field list so call scripts and monitoring change today

7. Collect evidence auditors and regulators will ask for

When the investigation ends, the questions turn to whether controls were in place. Collect the evidence now, as part of normal operations:

  • the data inventory with owners, classification and review dates,
  • SBOMs for each system that holds customer identity data, with dates,
  • vulnerability assessment and penetration test reports and the fix timelines that followed,
  • reachability and remediation records for known exploited vulnerabilities,
  • vendor access reviews, MFA status and credential rotation records,
  • outsourcing contracts showing audit rights and incident reporting clauses,
  • log retention configuration and an example query showing 180 days of history,
  • tabletop exercise records for the six hour report and for customer notification.

Each maps to an RBI paragraph or a CERT-In direction, and most will map to DPDP duties from May 2027.

Fix it without waiting for upstream

You cannot speed up SBI's investigation, and you do not need its findings to act. You can also act without waiting for your vendors or for upstream open source maintainers. Every step below is in your control today.

  1. Run the field set search this week. List every store and flow in your estate that holds name, mobile, date of birth, address and a customer or lead identifier together. Assign an owner to each.
  2. Generate SBOMs for those systems from both source and deployed images, and join the results against the CISA KEV catalog. Patch or upgrade reachable known exploited components first.
  3. Where an upstream library has no fixed release yet, ship your own fix. Pin to a patched fork, backport the upstream commit into a vendored copy, or swap in a hardened build of the component. Record the change so you can return to the upstream release when it ships.
  4. Test every lending API that takes an ID for broken object level authorisation. Change the ID, see what comes back, and fix the access check in your own code.
  5. Cut the fields in partner feeds. Edit integration mappings so DSAs, dialler vendors and analytics partners receive only what they need, for as long as they need it.
  6. Rotate partner credentials and disable unused vendor accounts. Require phishing-resistant MFA on every console that reaches customer data.
  7. Purge stale exports. Search shared drives, buckets and mailboxes for customer and lead extracts older than their purpose, and delete them.
  8. Seed canary records in each outbound flow, so the next leak names its path.
  9. Turn on query and bucket access logging with 180 day retention in India, and add a bulk read alert.
  10. Add a release gate that blocks known exploited components and critical findings in any service that handles customer identity data.
  11. Rehearse the six hour report and the customer notice. Draft the CERT-In report, the RBI report, and an SMS, email and letter in English, Hindi and the regional languages your customers read.
  12. Brief fraud operations and the contact centre on the call patterns described above and require call backs on registered numbers for any change to contact details or loan disbursement instructions.

Each of these steps is worth doing whether or not you ever learn what happened at SBI.


How Safeguard helps with this class of problem

SBI received alerts about this incident from its threat intelligence provider and CERT-In, among others, and it disclosed the incident itself. Our part is helping banks, NBFCs and fintech lenders manage the software and third-party risk around systems that hold customer data. These are the generally available capabilities that map to the work above.

SBOMs for the systems that hold customer data. Safeguard generates and analyses SBOMs in SPDX and CycloneDX across source, containers and pipelines, with full transitive dependency inventory, and accepts CSAF and VEX documents from suppliers. When a new known exploited vulnerability lands, "which of our customer data systems run this?" has an answer in minutes.

Reachability and prioritisation. Vulnerability findings are enriched with EPSS and CISA KEV data, and reachability analysis shows which vulnerable functions your code actually calls. Teams use it to focus on the reachable, known exploited issues in loan journeys, partner portals and document services first.

Application security across your own estate. Software composition analysis, static analysis of first-party code, and dynamic testing of verified running targets you own run in one platform, so issues such as broken authorisation in a lending API surface alongside dependency risk. That supports the six monthly and annual testing cadence in the RBI direction with continuous coverage in between.

Remediation without waiting. Griffin, our remediation model, opens policy-gated fix pull requests for vulnerable dependencies, one at a time or in bulk, and your own CI runs the tests before anyone merges. Gold Registry provides hardened, signed drop-in artifacts for open source components, which helps when upstream has not shipped a fix.

Third-Party Risk Manager (TPRM). Keep a vendor directory with contracts, risk tiers and owners. Request SBOMs and other evidence from service providers individually or in bulk, score supplier risk across vulnerability, compliance and component risk, and monitor continuously for new critical vulnerabilities and KEV additions in the components your vendors ship to you.

Policies and gates. Write organisation-wide policies, including from plain English prompts, and enforce them at build, CI and release. One example: no release of a service that handles customer identity data ships with a known exploited component.

Compliance evidence in one place. The compliance suite brings frameworks, controls, evidence, questionnaires, risks and vendors together, so the controls behind your CERT-In, RBI and DPDP obligations carry current evidence rather than last year's screenshots. Safeguard itself is SOC 2 Type I and Type II audited.

Deployment that fits a bank. Safeguard runs as SaaS, in your own cloud account, or on premises. The local runner CLI can scan on your own machines and send only encrypted results. Zero AI mode runs the platform with no language model in the loop, and teams that want AI assistance can bring their own model.

You can see the platform at https://safeguard.sh.


FAQ

What did SBI disclose about the dark web data exposure?

On 21 October 2026, SBI told the stock exchanges that it had experienced an incident involving the loss of certain data. A preliminary analysis of about 375,000 records found on the dark web identified 3,897 customer records with personal data, including names, mobile numbers, dates of birth, addresses, CIF identifiers, lead IDs and loan and lead metadata.

Who is Siriusim?

Siriusim is the handle of an actor who claimed on BreachForums to have released about 375,000 CSV rows and 2,722 PDF documents linked to SBI. Little is known about the actor, and SBI's filing does not name it or confirm its claims.

Were passwords, OTPs, card numbers or account balances exposed?

SBI's filing does not list account numbers, balances, passwords, OTPs, card details, PAN or Aadhaar numbers among the data identified. The seller claimed KYC documents were included. That claim is unverified.

How did the data leak?

The cause has not been made public. SBI engaged an independent CERT-In empanelled firm to establish the source and extent of the exposure. No system, vendor or vulnerability has been named.

Did SBI notify affected customers?

IFF said on 21 October that it had found no customer notice. SBI said on 23 October that it had contacted affected customers individually by email or WhatsApp.

Does Indian law require banks to notify customers of a data breach?

The DPDP Act's Section 8(6) requires notice to the Data Protection Board and each affected person, with Rule 7 of the DPDP Rules setting the process, but those provisions come into force on 13 May 2027. Today, CERT-In and the RBI require reporting to regulators within six hours, and RBI Advisory No. 3/2026 recommends, without requiring, customer notification.

How does Safeguard help?

Safeguard helps banks and lenders manage software and third-party risk with SBOM generation, reachability based prioritisation, SCA, SAST and DAST, automated fix pull requests, Gold Registry hardened components, third-party risk management, policies and gates, and compliance evidence.


Sources

All accessed 15 November 2026.

Incident reporting

Civil society

Regulation and law

Vulnerability data

Never miss an update

Weekly insights on software supply chain security, delivered to your inbox.

We email you a link to confirm, and only then add you to the list. Unsubscribe in one click. Privacy

Self-healing security runs on Safeguard.

Your first fix PR is minutes away.

No sales call required, even your agent can complete the purchase over MCP.