What Happened in Korea, and Why the Entry Point Matters
In early October 2026, seven South Korean financial institutions were reported breached in what looks like a single campaign. According to Insurance Business Asia (8 October 2026), personal data was exposed at Shinhan Bank, KB Kookmin Bank, Hana Bank, BNK Busan Bank, Welcome Savings Bank, Yegaram Savings Bank and Hyundai Capital. Woori Bank and NH NongHyup detected attempts but reported no breach.
The detail that matters most for Indian lenders is the entry point. The attackers did not go after core transaction networks. They targeted external-facing systems used by employees and loan agents, the portals that sit outside the hardened core so that intermediaries can submit applications, check status and see credit decisions. The same attacker IP address appeared at all seven firms.
The stolen data was the kind a loan-origination portal holds by design:
- names and phone numbers
- resident registration numbers, Korea's national identifier
- loan application details
- calculated credit limits
Yegaram Savings Bank lost about 40,000 records and Shinhan Bank about 25,000. None of this required touching a ledger. The portal itself was the asset.
Detection Took Up to 68 Hours
The report gives detection times that any Indian CISO should hold up against their own monitoring. Shinhan took about 15 hours to detect the intrusion, Hana about 42 hours and KB Kookmin about 68 hours. These are large, well-resourced banks with mature security operations, and the attacker still had between most of a day and nearly three days inside agent-facing systems before anyone noticed.
That gap drives cost. Every hour of undetected access widens the set of records touched, lengthens the forensic scope and enlarges the notification population. It also pushes against reporting clocks. In India, the CERT-In directions of April 2022 require specified cyber incidents to be reported within six hours of being noticed. A six-hour clock that starts after a 68-hour blind spot leaves a firm reporting an incident whose full extent it cannot yet describe.
The Korean response was fast once the pattern was visible. President Lee Jae Myung raised signs of AI use at a cabinet meeting on 6 October 2026. The Financial Supervisory Service alerted about 500 financial firms to roughly 30 malicious IPs across 12 countries and territories, and the Financial Services Commission ordered firms to block non-essential external access. That last order is the operational lesson: if regulators can tell an entire sector to switch off non-essential external access, those access points were never essential to begin with, or were never inventoried well enough to know.
India Runs the Same Architecture: DSA, Loan-Agent and POSP Portals
Indian credit and insurance distribution leans heavily on intermediaries. NBFCs and banks originate a large share of retail and MSME loans through direct selling agents (DSAs) and loan-service providers. Insurers distribute motor, health and term products through point of sales persons (POSPs) and corporate agents. Each of these channels runs on an external-facing portal or app.
Those portals typically hold exactly what the Korean attacker took:
- KYC identifiers such as PAN, Aadhaar-linked references and mobile numbers.
- Loan application data, income documents and bank statements.
- Bureau pulls and computed eligibility or credit limits.
- For insurers, proposal forms with health disclosures and nominee details.
The access model is also similar. Thousands of agent logins, often on personal devices, with credentials shared inside small DSA firms, and sessions that stay open for convenience. Portals are built for onboarding speed, not for the same scrutiny as the core banking system or policy administration system.
For background on how digital lenders structure their wider insurance programme, see our guide to insurance for digital lending and NBFC-fintech startups.
Sizing the Cyber Limit for a Portal Breach
A breach confined to agent-portal data does not shut down lending, so business interruption may be modest. The cost sits in first-party response and in regulatory and third-party exposure. A limit-sizing exercise for an Indian NBFC or insurer should build the scenario from records, not from turnover.
The first-party build
Start with the number of records reachable through a single compromised agent or employee account, and through the portal as a whole. Then cost each response line:
- Forensics across the portal, identity provider and any API that the portal calls, scoped for a multi-day dwell time rather than a single-day intrusion.
- Notification and call-centre support for every affected borrower or policyholder, plus credit-monitoring or fraud-alert support where identifiers were exposed.
- Legal counsel and regulatory response, including engagement with RBI or IRDAI and CERT-In.
- Crisis communications, which matter more for a lender whose agents face customers daily.
The regulatory and liability layer
The Digital Personal Data Protection Act, 2023 allows the Data Protection Board to impose penalties of up to INR 250 crore for failure to take reasonable security safeguards and up to INR 200 crore for failure to notify a breach. Whether, and how far, such penalties are insurable under an Indian cyber policy depends on the wording and on public-policy limits, so the limit should not assume full recovery of fines. Borrower claims and regulatory defence costs, however, are typically within scope and grow with the record count.
A Korean-style scenario of 25,000 to 40,000 records per institution is a reasonable starting point for a mid-sized Indian NBFC with an active DSA channel. Our analysis of India's INR 25.5 crore average breach cost and the method in cyber risk quantification for limit sizing give two ways to stress the result.
Statutory Minimums Are Not Adequate Limits
The Insurance Business Asia report makes a point that translates directly to India: Korea's statutory insurance or reserve minimums for electronic-financial and data-breach incidents may not equal adequate limits. A minimum is set to guarantee that some money exists for affected customers. It is not calibrated to forensic scope, multi-institution campaigns or regulatory defence.
Indian buyers fall into the same trap through a different route. A board approves a cyber limit because it matches a peer, a lender covenant or an earlier renewal, and the number then hardens into a floor that nobody revisits. Three tests help:
- Record test. Does the limit cover first-party response for every record reachable from the largest external portal, at current per-record costs?
- Dwell-time test. Is the forensic budget built for 48 to 72 hours of undetected access, given that two of the three Korean banks reported took 42 hours or more?
- Sublimit test. Are regulatory defence, notification and forensic costs carved into sublimits that are smaller than the headline limit suggests?
AI-Misuse Exclusions: What to Ask at Renewal
President Lee Jae Myung has pointed to signs of AI use in the campaign. Whether or not that is confirmed, the insurance question is already live. The same report cites a November 2025 Delinea survey in which 42% of over 750 security leaders said their cyber policies specifically exclude AI misuse or liability.
For an Indian NBFC or insurer, an AI exclusion can bite in two directions. It may be drafted to exclude losses caused by attackers using AI tools, which would put campaigns like the Korean one at risk of dispute. Or it may exclude liability arising from the insured's own use of AI, such as automated credit-limit calculation or chatbot-driven agent support. Ask the broker and insurer for written answers to these questions:
- Does any clause exclude loss where an attacker used AI, machine learning or automated tooling?
- Does any clause exclude liability arising from the insured's own models, including credit scoring or limit calculation?
- Is the exclusion absolute, or is there a carve-back for standard network security and privacy events?
- How will the insurer establish that AI was or was not involved, and who bears that burden of proof?
Our piece on AI-powered cyberattacks and coverage implications sets out how underwriters are responding. The broader rule is simple: read the exclusions before agreeing the limit, because an excluded loss makes the limit irrelevant.
Controls That Change the Underwriting Conversation
The FSC's order to block non-essential external access gives Indian lenders a ready checklist, and these same controls are what cyber underwriters probe on a submission:
- An inventory of every external-facing portal used by DSAs, loan agents, POSPs and employees, with an owner for each.
- Phishing-resistant multi-factor authentication for every agent login, not only for staff.
- Geo and IP restrictions, so that a portal used by agents in India does not accept sessions from unrelated countries.
- Rate limits and anomaly alerts on bulk record views and exports, which is how 40,000 records leave a system.
- Field-level masking so that agents see only the identifiers they need for the application in hand.
- A tested incident plan that can meet the CERT-In six-hour reporting window.
Firms that can evidence these controls are better placed to negotiate limit, retention and sublimits, and to push back on broad AI exclusions. Firms that cannot will find underwriters pricing the portal as the weakest point, which after Korea is a fair reading. For the cover itself, see our cyber insurance page.