What Zero Trust Actually Means for a Bank and What It Takes to Get There
Supervisors don't ask whether your bank is "zero trust." They ask who can reach critical systems, from which devices, and how you'd know if data left. This piece covers the controls they check, the order to build them in, and indicative costs.
Supervisors ask about controls, not about zero trust
Almost every security vendor will say its product makes your bank "zero trust." No supervisor grades you on that label. Examiners working under the EU's Digital Operational Resilience Act (DORA), the Monetary Authority of Singapore's Technology Risk Management (TRM) Guidelines or the Reserve Bank of India's cyber and IT governance directions ask much narrower questions. Who has privileged access to core banking, and who approved it? Can someone get in with nothing more than a stolen password? If one laptop is compromised, can the attacker reach the payments switch from it? Would you notice if customer data left the building?
Those questions are the practical meaning of zero trust security for financial services. A device on the corporate network gets no automatic trust. Each request to a system or dataset is checked against who is asking, from what device, and whether the request makes sense, and the checks keep running after login. NIST SP 800-207 is the reference definition most vendors and supervisors use. CISA's Zero Trust Maturity Model splits the work into five pillars: identity, devices, networks, applications and workloads, and data.
For a buyer, the real questions are which controls to fund first, how long each takes and what it costs. The sections below take the four main areas in the order most banks should tackle them.
What DORA, MAS TRM and RBI guidance actually look for
None of these frameworks requires a specific architecture. They describe outcomes, and zero trust controls happen to produce those outcomes. Under DORA, the detail is in the regulatory technical standards on the ICT risk management framework, which cover access control, strong authentication, network segmentation, encryption and logging. MAS TRM covers similar ground. The MAS Notice on Cyber Hygiene turns some of it into mandatory baseline requirements, including multi-factor authentication for privileged and critical-system access. In India, the RBI's Cyber Security Framework for banks and its Master Direction on IT governance, risk, controls and assurance practices set expectations on privileged access, segmentation and data leakage.
Other jurisdictions ask for the same things in their own wording. The UK operational resilience regime, the Saudi Central Bank's cyber security framework, the New York DFS cybersecurity regulation and the Hong Kong Monetary Authority's Cyber Resilience Assessment Framework all probe some version of the areas in the table below. These documents are revised regularly, so confirm the current text with your compliance team before mapping controls to them.
| Control area | What examiners typically probe | Evidence they expect |
|---|---|---|
| Identity | Phishing-resistant or strong MFA, joiner-mover-leaver discipline, periodic access recertification | MFA coverage reports, access review sign-offs, orphaned account counts |
| Privileged access | Who holds admin rights, whether they're time-bound, whether sessions are recorded | PAM vault inventory, session logs, break-glass procedures |
| Devices | Only managed, patched devices reach sensitive systems | Endpoint coverage, patch SLAs, device posture policies |
| Network | Critical systems segmented, remote access controlled | Segmentation diagrams, firewall rule reviews, penetration test results |
| Data | Classification, encryption, leakage detection | Data inventory, DLP policies and incident records |
| Third parties | Vendor and outsourcer access is scoped and monitored | Third-party access register, contract clauses, access logs |
Start with identity and privileged access management
Identity comes first because stolen credentials and over-privileged accounts appear somewhere in the chain of most serious intrusions. Identity is also where examination findings tend to cluster.
The baseline is one identity provider for all workforce access, with MFA on every application, not only email and VPN. For administrators and payment operators, examiners increasingly expect phishing-resistant methods such as FIDO2 security keys or platform passkeys. SMS one-time codes are now more likely to be flagged as a weakness than accepted as adequate.
Privileged access management (PAM) gets the closest scrutiny. At a bank, a workable PAM program covers:
- Vaulting and rotating credentials for domain admins, database administrators, core banking super-users and service accounts
- Just-in-time elevation, so admin rights exist only for an approved window rather than permanently
- Session recording on tier-zero systems such as Active Directory, the payments switch, SWIFT or local RTGS connectivity, and HSM management
- Routing vendor and outsourcer support access through the same broker, with no shared accounts or standing VPN tunnels
In international bank shortlists, CyberArk, BeyondTrust and Delinea come up most for PAM, and Saviynt appears where banks want identity governance and privilege in one platform. Microsoft Entra, Okta and Ping cover workforce identity. Brand matters less than coverage. If a PAM tool vaults only 40% of privileged accounts, the examiner sees a finding. Before signing, ask each vendor how it discovers unmanaged service accounts and what onboarding costs for mainframe or IBM i (AS/400) environments if you still run them.
Devices and the network: retiring the flat network and the always-on VPN
On a flat bank network, one phished branch PC can scan its way to the core ledger. Zero trust deals with that from two directions.
Device trust. Run endpoint detection and response (EDR) on every managed laptop, server and virtual desktop. Feed posture signals such as patch level, disk encryption and EDR health into each access decision. An unmanaged or out-of-date device should get a restricted session or no access. Banks with large contractor workforces or offshore operations centers need this most.
Network access. Zero trust network access (ZTNA) replaces broad VPN access with per-application connections. It's usually bought as part of a security service edge (SSE) platform from vendors such as Zscaler, Netskope, Palo Alto Networks, Cloudflare or Microsoft. A user can reach the loan origination system without having any access to the subnet it runs on.
Inside the data center, segmentation does the same job for east-west traffic. You don't have to microsegment everything. Ring-fence the crown jewels first: core banking, payments, card processing, identity infrastructure and backups. Backups are the item most often left out, and they're among the first things ransomware operators go after.
ZTNA has a real limitation. It handles web and modern applications well, but older thick-client and terminal-emulation apps often don't work cleanly behind it. Test those apps early, and expect to keep a small, tightly controlled VPN for a handful of legacy cases for a year or more.
Data controls: classification before data loss prevention
Banks often buy data loss prevention for banks before they're ready for it. With no classification scheme in place, a DLP tool either floods analysts with false positives or gets tuned so loosely that it misses real leaks. These projects usually fail because steps were done in the wrong order, not because the product was weak.
Follow this sequence:
- Define three or four classification levels, each tied to a real obligation: customer personal data, card data under PCI DSS, market-sensitive information and internal material.
- Run automated discovery to find where that data actually sits, including file shares, SharePoint or Google Drive, email archives and analytics sandboxes.
- Apply labels and encryption to the highest tiers first.
- Run DLP in monitor-only mode for several weeks, typically six to ten, and tune it. Then switch to blocking for clear-cut cases such as bulk customer exports to personal email or unsanctioned cloud storage.
Data residency rules change how you deploy all this. The GDPR in Europe, India's Digital Personal Data Protection Act, Saudi Arabia's PDPL, Indonesia's PDP Law and several Latin American privacy laws can limit where DLP inspection logs and content samples are stored and processed. If your SSE or DLP vendor inspects traffic in its own cloud, confirm which regional points of presence handle your traffic and whether you can pin processing to a specific jurisdiction. Privacy counsel should sign off before rollout, because the answer depends on your jurisdictions and your contracts.
Realistic phasing and budget ranges
For a mid-sized bank this is a two-to-three-year program. The phasing below suits most institutions. If a recent audit finding forces one area to the front, move that area up.
| Phase | Typical timing | Main deliverables |
|---|---|---|
| Foundations | Months 0-4 | Asset and identity inventory, crown-jewel mapping, single identity provider, MFA on all remote and admin access |
| Privilege and endpoints | Months 3-12 | PAM for tier-zero and vendor access, EDR coverage above 95%, device posture policies |
| Network | Months 9-20 | ZTNA for most applications, VPN reduced to exceptions, segmentation of core, payments and backups |
| Data and continuous verification | Months 15-36 | Classification and labeling, DLP in blocking mode, risk-based access policies, third-party access monitoring |
The licensing figures below are typical 2026 market ranges. They vary widely with contract size, region and bundling, so confirm them with vendors:
- Workforce identity and MFA: roughly $4-15 per user per month
- PAM: roughly $800-2,500 per privileged identity per year, and minimum contracts often start in the tens of thousands of dollars
- ZTNA/SSE: roughly $6-20 per user per month
- EDR: roughly $5-15 per endpoint per month
- Data security and DLP: roughly $3-12 per user per month
- Workload segmentation: roughly $150-500 per server workload per year
Hypothetical worked example: a bank has 1,500 staff, 2,000 endpoints, 120 privileged identities and 600 server workloads. At mid-range prices, annual licensing comes to about $144,000 for identity, $180,000 for ZTNA, $180,000 for PAM, $190,000 for EDR, $110,000 for DLP and $180,000 for segmentation. That's roughly $1 million a year. In year one, implementation partners typically add another 50-120% of first-year licensing, which puts external spend at about $1.5-2.2 million. On top of that you need two to four internal engineers, and their cost depends heavily on the country. From year two, expect external costs of around $1-1.3 million a year. Prices in India, Southeast Asia, Africa and Latin America are often lower, and regional resellers frequently set the final figure, so get quotes in your own market.
Three things drive the total more than anything else: how many legacy applications you run, how many privileged and service accounts discovery uncovers, and whether you consolidate on one platform or buy best-of-breed tools.
Red flags in zero trust pitches and a decision rule
Treat any of these in a vendor meeting as a warning sign:
- "One product makes you zero trust." No single product covers identity, privilege, devices, network and data to the depth examiners expect.
- No regulatory mapping. A serious vendor can show how its logs and reports become evidence for DORA, MAS TRM or RBI reviews without a separate consulting engagement.
- Vague data residency answers. "We're compliant everywhere" doesn't tell you anything. Ask for the list of processing regions.
- Exit terms treated as an afterthought. DORA expects exit strategies for ICT services that support critical or important functions, and outsourcing rules in many other jurisdictions say much the same. Ask how you would move your policies and logs out.
- Licensing priced on the full future estate. Pay for what you'll deploy in the next 12-18 months, and get price protection for expansion.
When choosing between finalists, pick the platform that closes your open audit or examination findings fastest and produces evidence you can hand to a supervisor. If you have no open findings, fund identity and privileged access first. This quarter, ask your security team for three numbers: MFA coverage, the share of privileged accounts that are vaulted, and EDR coverage. Whichever is lowest is where the first budget should go.
Frequently asked questions
Does DORA require banks to adopt zero trust?
No. DORA doesn't mention zero trust or require any particular architecture. Its ICT risk management requirements and supporting technical standards do expect strong authentication, least-privilege access, network segmentation, encryption and logging, and zero trust controls provide all of these. Work with your compliance team to map your controls to the current regulatory text.
How long does it take a mid-sized bank to implement zero trust?
Plan on two to three years to reach a mature state across identity, devices, network and data. MFA coverage and privileged access controls can usually be in place within the first 6-12 months, and examiners check those first. Data classification and blocking-mode DLP usually come last, because you first need to know where sensitive data lives.
Should we buy a single zero trust platform or best-of-breed tools?
A consolidated platform costs less to integrate and is simpler to run, which helps banks with small security teams. Best-of-breed tools, especially for PAM, often handle legacy systems and tier-zero assets in more depth. Many banks settle on a hybrid: one SSE platform for network and data controls plus a specialist PAM vendor.
Do zero trust controls change for offshore operations centers or outsourced support?
The controls stay the same, but they matter more in these setups. Route outsourcer access through the PAM broker and ZTNA rather than site-to-site VPNs, require managed or posture-checked devices, and record sessions on tier-zero systems. Outsourcing and data-transfer rules differ by country, so have compliance and privacy counsel confirm what each regulator expects.
Written by the Adviser Hub Cloud editorial team. Prices, features and rules change often - confirm details with the vendor or a licensed professional before you decide.