Ask whether you can restore a clean bank, not just whether you can block an attack
Most bank security budgets go to keeping attackers out: EDR, email filtering, privileged access, segmentation. That spending is necessary. But it doesn't answer the question boards and regulators ask after every serious incident. If your production estate and your ordinary backups were encrypted tonight, how many hours until you could post transactions, settle payments and let customers log in? And how would you prove the restored systems are clean?
Ransomware protection for banks is really two disciplines. Prevention lowers the odds of an attack. Recovery decides how bad the damage gets. Ransomware crews routinely go after backup consoles, hypervisors and identity systems first, because a bank that can restore quickly has little reason to pay. This playbook covers recovery only, in 11 steps: define what has to come back, isolate and lock your copies, rebuild in the right order, reconcile money in flight, meet every notification deadline, and test the whole chain before you need it.
None of this replaces your legal counsel or your regulator's current rules. Notification regimes change, so treat the deadlines below as a starting map, not a final answer.
Looking to streamline vendor RFP selection?
Access pre-negotiated tier discounts through our independent institutional network.
Steps 1–2: Define your minimum viable bank and put numbers on it
Step 1: Name the minimum viable bank. This is the smallest set of services you must restore to meet customer and regulatory obligations. Usually that means account balances and the core ledger, inbound and outbound payments (high-value, instant and card authorization), cash access, and a way to talk to customers. Wealth platforms add order management and custody reconciliation. Lenders add servicing and collections. Everything else, including analytics, marketing, CRM enrichment and most of the data warehouse, can wait days.
Step 2: Set recovery time and recovery point objectives per service, then check them against reality. UK firms set impact tolerances under the operational resilience rules. EU firms map critical or important functions under DORA. Australian entities do something similar under APRA's CPS 230. Even where no rule requires it, the exercise is the same: write down a maximum tolerable outage and a maximum acceptable data loss for each service.
Then do the uncomfortable part. Compare those targets with what your backup tooling can deliver at full scale. Say a hypothetical mid-sized bank sets a 4-hour RTO for its core ledger, but its last full restore test of a 40 TB core database took 19 hours. The 4-hour target is fiction. Either fund faster recovery (more vault throughput, pre-staged clean compute) or have the board formally accept a longer tolerance. Either beats finding out during an attack.
Steps 3–4: Build isolated, immutable backups and a cyber recovery vault
Step 3: Make backups immutable. Immutable backups are copies that nobody can alter or delete before retention expires, including a domain administrator or a backup admin whose credentials have been stolen. In practice that means WORM storage, object lock in compliance mode (not governance mode, which privileged users can override), or appliance retention locks that need multi-person approval to change.
Step 4: Isolate a copy in a cyber recovery vault. Immutability alone isn't enough if attackers can reach the management plane. A vault keeps a further copy of critical data and system images in a separate environment, with its own identity store, no standing network connection (an operational air gap that opens on a schedule), and tooling that scans copies for tampering. For mainframe cores, snapshot features such as IBM's Safeguarded Copy fill a similar role. Institutions in the US can also look at Sheltered Harbor, an industry standard for vaulting customer account data in a common format.
| Approach | What it protects against | Typical 2026 cost range (indicative) | Main trade-off |
|---|---|---|---|
| Immutable object storage (cloud or on-premises) with object lock | Deletion and encryption of backup data | Roughly $20–$60 per TB per month, plus egress on restore | The management plane stays reachable if cloud or storage admin accounts are compromised |
| Hardened backup appliance with retention lock (offerings from Dell, Rubrik, Cohesity, Veeam, Commvault and others) | Deletion, tampering, many admin-credential attacks | Roughly $150,000–$1M+ upfront for mid-sized banks, plus annual support | Often shares a network or identity with production unless configured carefully |
| Fully isolated cyber recovery vault with clean-room compute | Full-estate compromise, including identity and backup infrastructure | Roughly $500,000–$3M+ to build; managed vault services often tens of thousands per month | Highest cost; needs disciplined operations and regular restore drills |
These are typical 2026 market ranges. They vary widely with data volume, region and contract terms, so confirm with vendors. For most banks with a few billion dollars in assets or more, the isolated vault is worth it for tier-0 and tier-1 data, with immutable storage covering everything else. Smaller institutions may get there through a managed service.
Red flags in vendor pitches: "air-gapped" that turns out to mean a separate VLAN; immutability that one admin can switch off; no documented restore throughput at your data volume; and anomaly detection sold as a replacement for restore testing.
Step 5: Prepare a clean room and protect the identity layer first
Restored data is useless if you load it into an environment the attacker still controls. Step 5 is to pre-build a clean recovery environment: isolated compute, network and a fresh identity tenant where you can rebuild systems, scan them and verify them before reconnecting anything.
Inside that environment, identity is the first thing to restore and the hardest. Attackers often hold domain admin privileges before they encrypt anything. Keep offline, immutable copies of Active Directory or your cloud directory configuration, privileged access vault contents, HSM key material backups, certificate authority data and DNS zones. Keep break-glass credentials in physical custody, split across at least two named officers. Banks that skip this can find perfectly good core backups sitting unusable because nobody can authenticate to the systems that hold the keys.
Looking to streamline vendor RFP selection?
Access pre-negotiated tier discounts through our independent institutional network.
Steps 6–7: Restore core, payments and digital channels in a deliberate order
Step 6: Write down the restore sequence and the dependencies behind it. The order below fits most retail and commercial banks. Adjust it for your architecture, but don't leave it to be improvised mid-incident.
- Foundations: clean network segments, DNS, time sources, identity, privileged access, HSMs and security monitoring, so you can see whether the attacker returns.
- Core ledger and customer master data: restored to the last verified clean point, then checked against the most recent end-of-day balances held outside the core.
- Payment gateways and messaging: RTGS and SWIFT interfaces, instant payment connections, ACH or batch clearing, and the card authorization switch. Card authorizations often run on stand-in processing at the scheme or processor during an outage, so agree those limits in advance.
- Branch and contact-center tools, so staff can serve customers with read access before full self-service returns.
- Digital channels: mobile and online banking, first read-only (balances, recent transactions), then payments with lower limits, then full functionality.
- Everything else: lending origination, wealth reporting, data platforms, internal apps.
Step 7: Plan for degraded operations. Decide in advance what you'll run manually: high-value payments through an authenticated alternative channel with dual sign-off, temporary withdrawal caps, paused new account opening. Pre-draft customer messages for each stage. Reopening digital channels early in read-only mode is usually the best way to cut contact-center load while the harder restores continue.
Step 8: Reconcile the money that was in flight
The hardest problems after a restore are accounting problems, not technical ones. If your clean recovery point is 02:00 and encryption hit at 09:30, there are seven and a half hours of payments, card authorizations and trades the restored core has no record of.
Step 8 is a documented reconciliation procedure. Rebuild the gap from external sources of truth: SWIFT and RTGS message logs, card scheme and processor settlement files, clearing house reports, correspondent bank statements and custodian records. Assign an owner per payment rail, define how duplicate or missing items are flagged and corrected, and set rules for customer-facing adjustments. Ask your processors and correspondents now how quickly they can deliver transaction files for a given window, and get the answer written into contracts.
Step 9: Map every incident reporting clock before you need it
A single ransomware event at a multinational bank can trigger a dozen notification duties, each with its own clock, trigger definition and template. Well-run recoveries still pick up regulatory findings here, usually because nobody owned the clock. Step 9: build a jurisdiction-by-jurisdiction matrix into your incident response plan, with the legal trigger, the deadline, the recipient, a pre-approved template and a named owner.
| Regime (examples) | Typical deadline | Notes |
|---|---|---|
| Singapore, MAS technology risk notice | Within 1 hour of discovery for relevant incidents | A root cause and impact report follows, generally within 14 days |
| EU, DORA major ICT-related incident | Initial report within 4 hours of classifying the incident as major, and no later than 24 hours after becoming aware | Intermediate and final reports follow on set timelines; confirm current technical standards |
| India, CERT-In directions | Within 6 hours of noticing the incident | RBI and other sector regulators have their own expectations as well |
| US, federal bank Computer-Security Incident Notification Rule | As soon as possible, and no later than 36 hours after determining a notification incident occurred | Bank service providers must notify client banks of significant disruptions; state rules, NYDFS and SEC disclosure can apply too |
| Australia, APRA CPS 234 | Within 72 hours of becoming aware of a material incident | Ransomware payments may also need reporting to government within 72 hours for in-scope businesses |
| Personal data breach, e.g., GDPR / UK GDPR | Within 72 hours to the supervisory authority, where required | Runs in parallel with financial regulator reporting |
UK firms must tell the PRA and FCA about material incidents promptly, and more structured operational incident reporting is being phased in, so check current status. Regulators across the Middle East, Africa and Latin America set their own timelines, commonly ranging from a few hours to a few days. Have local counsel confirm each entry once a year. Several clocks start at "awareness" or "discovery," not when your investigation ends.
Step 10: Settle the ransom, insurance and vendor questions in advance
Step 10: Get board-level decisions on the hard questions before an attack. Will the bank ever consider paying? Many regulators discourage payment, and paying a sanctioned group can breach sanctions law in the US, UK, EU and elsewhere. Some jurisdictions require ransom payments to be reported. Decide who has authority, which outside counsel and incident response firms are on retainer, and what your cyber insurance actually covers. Many policies require insurer-approved responders and prompt notice, and some exclude losses tied to known unpatched vulnerabilities. Have a licensed broker and counsel review the policy wording against this playbook.
Then extend the playbook to critical vendors. Core banking providers, cloud hosts and payment processors should commit contractually to recovery times, notification windows (ideally matching your shortest regulatory clock) and participation in your exercises. DORA makes this explicit for EU firms. Elsewhere, it's simply good practice.
Step 11: Prove it with restore tests, not slide decks
Step 11: Test the full chain on a fixed schedule. A workable cadence for most banks:
- Monthly: automated restore and integrity checks of sampled backups from the vault.
- Quarterly: full restore of one tier-1 system into the clean room, timed against its RTO.
- Twice a year: executive tabletop exercise covering the notification matrix, the ransom decision and customer communications.
- Annually: an end-to-end exercise restoring the minimum viable bank, including payment reconciliation with at least one external counterparty. EU firms in scope should align this with DORA's threat-led penetration testing cycle.
Measure actual restore times, data loss, missed notification steps and manual workarounds that failed. Feed every gap into the next budget cycle. The decision rule: if your last full-scale restore of the core ledger missed its RTO, your next security dollar should go to recovery capacity, not another prevention tool. Start this quarter by timing a restore of your core database from your most isolated copy. That one number will show how much of the plan is real.