Guide Cybersecurity

Zero Trust Architecture: What It Is and How to Implement It

Zero Trust architecture removes the implicit trust a network used to give anyone connecting from inside it. Each access request is checked against the identity behind it, the state of the device and the context, and access is limited to what the task needs. This guide explains how NIST and CISA define the model and how to implement it with tools a hybrid organisation usually already has.

15 June 2026
Updated 18 November 2026
12 min read
Key takeaways
  • Zero Trust is an architecture and a set of principles, not a product. It removes implicit trust based on network location alone and decides access from identity, device state, context and risk, with least privilege and an assumption that a breach can happen.
  • NIST SP 800-207 (2020) sets out the tenets and the logical components that make access decisions. CISA's Zero Trust Maturity Model v2.0 (2023) organises the work into five pillars and three cross-cutting capabilities.
  • Many organisations start with identity because it is a practical first step that uses tools they already licence. Your own risk assessment should set the order.
  • Microsoft 365 Business Premium includes Conditional Access, Intune and Defender for Business. Privileged Identity Management and risk-based sign-in policies need Entra ID P2 or an equivalent add-on.
  • Zero Trust controls support NIS, NIS2, ISO 27001 and DORA requirements, but they do not show conformity on their own. You still need the policies, risk assessments and evidence each framework asks for.

What Zero Trust architecture is

Zero Trust architecture is a security model in which no user, device or workload is trusted just because of where it connects from. Each request for a resource is checked against policy using the identity behind it, the state of the device, and signals such as location, time and behaviour. Access is limited to what the task needs. John Kindervag described the model at Forrester Research in 2010, and NIST Special Publication 800-207 formalised it in 2020.

Why the perimeter stopped working

Traditional network security rested on one assumption: traffic inside the corporate network is trusted, traffic outside is not. A firewall held the line and remote workers tunnelled in through a VPN. That worked when most users sat in the same building, applications were run on infrastructure you owned, and data lived on servers you could physically touch.

Today users work from home and client offices, applications run in Microsoft 365, Salesforce, AWS and dozens of SaaS platforms, and partners connect from networks you have never seen. An attacker who compromises one laptop, one phished account or one VPN credential lands inside the trusted zone, where internal traffic often goes uninspected and lateral movement can be quick.

Zero Trust removes implicit trust based on network location alone. Location can still be one signal in a policy (a sign-in from a country where you have no staff is worth blocking), but it no longer grants access by itself. Depending on the architecture, decisions are made per session or per request and re-evaluated as policy, context and risk signals change.

48%
of breaches involved a third party such as a supplier or partner (Verizon DBIR 2026)
31%
of breaches started with exploitation of a software vulnerability (Verizon DBIR 2026)
5 + 3
pillars and cross-cutting capabilities in CISA's Zero Trust Maturity Model v2.0

How access decisions are made

Section 3 of NIST SP 800-207 describes three logical components. The policy engine (PE) decides whether to grant access, using policy and inputs such as identity, device posture and threat intelligence. The policy administrator (PA) acts on that decision by setting up or closing the session. The policy enforcement point (PEP) sits in the path between the user and the resource and allows, blocks or ends the connection. Together the PE and PA form the policy decision point.

In a Microsoft 365 environment, Entra ID Conditional Access plays the decision role for cloud sign-ins, and the application, proxy or ZTNA connector enforces it. Products differ in how they split these roles, so map your own tools to the components rather than assuming one product covers all three.

The five pillars

CISA's Zero Trust Maturity Model v2.0 groups the work into five pillars: Identity, Devices, Networks, Applications and Workloads, and Data. Three cross-cutting capabilities run across all of them: Visibility and Analytics, Automation and Orchestration, and Governance. The model describes maturity stages for each pillar, from traditional to optimal.

🔐
1. Identity
Verify users and non-human identities before granting access: MFA, Conditional Access policies that evaluate each sign-in, and time-limited activation of administrative roles. Service accounts, workload identities and API keys need the same discipline; NIST SP 800-207A extends the model to services in cloud-native applications. Our guide to machine identities covers the inventory and rotation work.
💻
2. Devices
Make sensitive resources reachable only from managed, compliant devices: operating system patched, disk encrypted, EDR running, enrolled in device management. Conditional Access checks compliance at sign-in and can block an unmanaged device or limit it to restricted access.
🖧
3. Networks
Segment the internal network so a compromised device in one zone cannot freely reach another, encrypt traffic, and give remote users access to specific applications rather than whole network ranges. Monitoring east-west traffic helps you spot lateral movement.
📦
4. Applications and Workloads
Enforce authentication at the application layer whether the user is in the office or at home, and remove applications that grant unauthenticated access from inside the network. A CASB can extend visibility and some controls to SaaS platforms you do not operate.
📂
5. Data
Classify data by sensitivity and apply protection that travels with it, such as sensitivity labels with encryption. DLP policies can detect or restrict certain transfers, for example uploads to personal cloud storage, but they reduce risk rather than guarantee that data stays in. Most organisations tackle this pillar later because it depends on the others.

Starting with identity

Identity is a common starting point because it covers every cloud application at once, uses tools many organisations already licence and needs no network changes. It is a practical choice, not a rule: if your main exposure is unpatched internet-facing systems or a flat OT network, start there.

Conditional Access policies

Conditional Access assesses each sign-in against signals such as user, device state, location, target application and, where licensed, risk scores from Entra ID Protection. The policy then allows, blocks or demands step-up verification. A baseline set requires MFA for all users, blocks legacy authentication, blocks sign-ins from countries where you have no presence and requires a compliant device for sensitive applications. Policies that react to user risk or sign-in risk need Entra ID P2.

Privileged Identity Management

Admin accounts are the highest-value target in any environment. PIM removes standing privilege: an administrator requests, justifies and activates a role for a limited time. Between activations the account holds no admin rights, so a stolen password alone does not give an attacker admin access. Someone who can also pass MFA and any approval step could still activate the role, so protect activation with phishing-resistant MFA and require approval for the most sensitive roles. PIM needs Entra ID P2 or Entra ID Governance.

What Microsoft 365 Business Premium includes

CapabilityIn Business Premium?Notes
Conditional Access (Entra ID P1)IncludedCovers MFA, device compliance and location policies
Intune device managementIncludedCompliance policies feed Conditional Access
Defender for Business (EDR)IncludedEndpoint detection and response for up to 300 users
Privileged Identity ManagementNot includedNeeds Entra ID P2 or Entra ID Governance
Risk-based policies (user and sign-in risk)Not includedNeeds Entra ID P2
Full Defender for Cloud Apps (CASB)Not includedSeparate licence or add-on

Microsoft changes its plans from time to time, so check the Business Premium service description before you buy.

Device trust and compliance

A valid MFA sign-in from a laptop running unpatched software or carrying an active infection is still a real access risk. Compliance policies in Intune or Jamf define the minimum bar: operating system version, encryption, EDR agent, screen lock and jailbreak detection on mobiles. The endpoint security guide covers each of these controls. Conditional Access checks compliance at sign-in and blocks devices that fall short.

If you allow personal devices, split the access model. Managed devices get full access. Unmanaged devices can be limited to browser-only sessions, using app enforced restrictions or Defender for Cloud Apps session controls where your applications and licences support them. These controls can block downloads, and in some apps copy and print, but they do not stop screenshots or a photo of the screen.

Network segmentation in practice

Start with your highest-value assets: finance systems, domain controllers, backup infrastructure and any operational technology (OT). Isolate them with deny-by-default inbound rules that permit only the flows each service needs. You do not have to segment the whole network to get most of the value.

Zero Trust Network Access (ZTNA) services such as Cloudflare Access, Microsoft Entra Private Access and Zscaler Private Access sit in front of internal applications and check identity and device before opening a connection. A traditional VPN usually gives network-level access once connected, limited only by whatever segmentation and filtering you have added. ZTNA brokers access per application, which limits what a stolen credential can reach. It still depends on well-written policies, and some ZTNA products can also be configured for broader network access.

Zero Trust and regulatory compliance

None of the main frameworks requires Zero Trust by name. Zero Trust contributes to the access control, authentication, least privilege and network security controls they ask for. Those frameworks also require risk management, incident handling, business continuity, supplier security, documentation and evidence that the controls work. Implementing Zero Trust supports those requirements; it does not show conformity on its own.

UK NIS Regulations 2018. Operators of essential services and relevant digital service providers are often assessed against the NCSC's Cyber Assessment Framework, whose principle B2 (identity and access control) maps closely to the identity and device pillars. The Cyber Security and Resilience Bill, still before Parliament, would extend the regime to more organisations, including managed service providers. The NCSC also publishes its own zero trust architecture design principles.

NIS2. If you provide services in the EU, NIS2 may apply to that activity. Article 21(2)(i) covers access control policies and Article 21(2)(j) covers multi-factor or continuous authentication.

DORA. The Digital Operational Resilience Act (Regulation (EU) 2022/2554) has applied to EU financial entities since 17 January 2025. Its ICT risk management rules, detailed in Delegated Regulation (EU) 2024/1774, include network segmentation, identity management, access control and strong authentication. UK firms feel it mainly through EU subsidiaries or as ICT providers to EU financial entities. See our DORA gap analysis guide.

ISO 27001:2022 Annex A. The controls below map most directly to a Zero Trust programme:

ControlZero Trust practiceEvidence an auditor may ask for
A.5.15 Access controlPolicy-based access per applicationAccess control policy, Conditional Access export
A.5.16 Identity managementLifecycle for user and non-human identitiesJoiner, mover and leaver records
A.5.17 Authentication informationMFA, passwordless, secret rotationMFA registration report, secrets policy
A.5.18 Access rightsLeast privilege, periodic reviewsAccess review results
A.8.2 Privileged access rightsJust-in-time admin roles (PIM)Role activation logs
A.8.3 Information access restrictionPer-resource authorisation, labelsPermission reports, label policy
A.8.5 Secure authenticationPhishing-resistant MFA, legacy auth blockedSign-in logs, policy settings
A.8.20 Networks securityEncrypted traffic, deny by defaultFirewall and ZTNA rule sets
A.8.21 Security of network servicesZTNA and DNS filtering requirementsService agreements, configuration
A.8.22 Segregation of networksSegmented zones for critical assetsNetwork diagrams, segmentation tests

You still need a risk assessment, a Statement of Applicability and records that the controls operate. See the ISO 27001 guide for SMEs.

What Zero Trust does not solve on its own

Zero Trust is not a certification, a product or a substitute for a compliance programme. A first deployment usually covers staff identities, devices and remote access. Plan for these gaps as well:

  • Workload identities. Service principals, managed identities, API keys and other secrets need an owner, narrowly scoped permissions and rotation, just like user accounts.
  • Phishing-resistant MFA. Push approvals and SMS codes can be phished or abused through MFA fatigue. Move administrators and other high-risk access to FIDO2 security keys, passkeys or Windows Hello for Business first.
  • Legacy applications. An application without modern authentication needs an application proxy, a ZTNA connector or its own isolated segment until it is replaced.
  • Policy lifecycle. Test new Conditional Access policies in report-only mode, record every exception with an owner and an end date, and review policies when staff, applications or licences change.
  • Identity provider outages. Keep at least two cloud-only emergency access (break-glass) accounts that cannot be locked out by your normal policies. Protect them with strong authentication, alert on every sign-in and test them on a schedule.
  • Vulnerabilities and detection. Identity controls do not stop the exploitation of an unpatched internet-facing system. You still need vulnerability management, logging that covers the systems that matter and detections that are tested.

How to implement Zero Trust in practice

Work area by area, starting where your risk assessment points. Each row pairs a first control with a way to check it works.

AreaInitial controlHow to verify
IdentityAudit all user and service accounts, enforce MFA, block legacy authenticationMFA coverage report; no legacy sign-ins in the last 30 days
PrivilegesRemove standing admin rights, use time-limited activationCount of permanent admin assignments; activation logs
DevicesManage corporate devices and require compliance for sensitive appsCompliance dashboard; blocked sign-ins from non-compliant devices
NetworkSegment critical assets and replace broad VPN access where possibleRule review; test that a user subnet cannot reach backup or domain controllers
ApplicationsPut internal apps behind authenticated access, remove anonymous accessApplication inventory with authentication method for each
DataLabel sensitive data and add DLP for the riskiest transfersLabel coverage; DLP matches reviewed each month
MonitoringSend identity and access logs to a SIEM with alerts on risky eventsAlerts for impossible travel, new device sign-ins and out-of-hours role activation fire in testing

Zero Trust assumes a breach can happen. The design question is how much an attacker can reach from their first foothold.

Zero Trust architecture is one workstream inside a security programme, not the whole programme. A cybersecurity roadmap places it alongside identity governance, endpoint management and incident response so the controls build on each other.

Ryland Deakin
About the author
Lead Consultant, Cyvra · CISM · CompTIA Security+ · MCP

Ryland has delivered cybersecurity, compliance, and IT management programmes for regulated organisations across the UK and the Netherlands for over 20 years. View full profile

Common questions

Is Zero Trust a product you can buy?

No. Zero Trust is an architecture and a set of principles. Vendors sell products labelled as Zero Trust, and Entra ID, Okta, Intune, Jamf and ZTNA platforms all support parts of it, but none covers every pillar. You also have to change how you manage identities, devices, network access and data.

Can a small business implement Zero Trust?

Yes. The usual starting point is MFA for all accounts, a baseline set of Conditional Access policies and device compliance checks. Microsoft 365 Business Premium includes Conditional Access and Intune; Privileged Identity Management and risk-based policies need Entra ID P2. A small organisation can usually put the identity and device basics in place within a few months. Reaching higher maturity across all five pillars is a longer programme.

What is the difference between ZTNA and a VPN?

A traditional VPN usually gives network-level access once connected, limited only by whatever segmentation and filtering you have added. ZTNA brokers access to specific applications and checks identity and device before each session, and in some products during it. A stolen credential therefore reaches fewer systems, but only if the access policies are well written. Some ZTNA products can also be set up for broader network access, which narrows the difference.

How does Zero Trust relate to NIS2 and ISO 27001?

Neither requires Zero Trust by name. Both ask for access control, authentication, least privilege and network security, which Zero Trust controls deliver. NIS2 Article 21(2)(i) and (j) cover access control and multi-factor authentication, and ISO 27001 Annex A controls A.5.15 to A.5.18, A.8.2, A.8.3, A.8.5 and A.8.20 to A.8.22 map closely to the model. Implementing Zero Trust supports these requirements, but you still need the risk assessment, policies and evidence each framework expects.

Zero Trust Assessment

Find out where your identity controls stand

We audit your Conditional Access policies, privileged access posture, and network segmentation against the Zero Trust framework and give you a prioritised remediation plan.

Disclaimer: This article is for general informational purposes only and does not constitute legal, regulatory, or professional advice. Cyvra makes no warranty as to the accuracy or completeness of this content, which may not reflect the most current regulatory developments. Readers should seek independent legal and regulatory advice appropriate to their specific circumstances. Cyvra accepts no liability for any loss arising from reliance on this content.