Cybersecurity
Zero Trust Architecture - how to introduce the NIST SP 800-207 model in IT and OT
Zero trust under NIST SP 800-207 step by step: seven tenets, PE, PA and PEP, four deployment models, the CISA maturity assessment and the limits of use in OT networks.
Józef SulwińskiPublished Updated 16 min read

Contents
In a network built around a perimeter, anyone who gets past the firewall gains broad access to the resources inside. A compromised employee account or an infected laptop gives an attacker the same permissions as its owner. Firewalls between zones do not change this, because they do not control traffic within a zone.
NIST and CISA publications show that:
- zero trust assumes the network is compromised and grants each access separately, with the least privileges needed,
- according to NIST, a large organisation may combine zero trust flows with a perimeter-based architecture for an indefinite period,
- according to NIST, some OT devices, such as controllers and HMI panels, do not support the technologies zero trust needs,
- the CISA maturity model does not address OT-specific challenges, so the control network needs separate decisions.
What a Zero Trust Architecture is
NIST defines zero trust as a collection of concepts that reduce uncertainty in every access decision. The decision is meant to be accurate and grant the least privileges needed, on the assumption that the network is compromised. A Zero Trust Architecture (ZTA) is the organisation’s security plan that describes components, workflows and access policies.
NIST calls the part of a network where everyone trusts everyone else an implicit trust zone. Zero trust moves access control as close to the resource as possible and shrinks this zone. Why: the smaller the zone, the fewer resources an intruder inside it can see.
NIST describes seven tenets that the organisation applies when designing the architecture.
| NIST tenet | What the organisation does |
|---|---|
| 1. All data sources and computing services are resources. | The organisation treats printers, sensors and personal devices with access to company data as resources too. |
| 2. All communication is secured regardless of network location. | The organisation applies the same requirements to requests from the corporate network as to requests from the internet. |
| 3. Access to a resource is granted per session. | The organisation evaluates trust before every session, and access to one resource does not open others. |
| 4. Access is determined by dynamic policy. | The organisation takes into account identity, device state, the application, and the behaviour and environment of the request. |
| 5. The organisation monitors the integrity and security posture of all its assets. | The organisation assesses the device state with every request and treats vulnerable or unmanaged devices differently. |
| 6. Authentication and authorisation are dynamic and enforced before access. | The organisation has an identity management system, uses MFA and re-checks access during the session. |
| 7. The organisation collects as much information as possible about assets, the network and communication. | The organisation uses this data to improve policies and to evaluate later requests. |
NIST notes that the tenets are a goal and that not every strategy implements them in their purest form. MFA (multi-factor authentication) is a login with at least two kinds of proof, for example a password and a hardware key.
Three components of the access decision
NIST separates the access decision from its enforcement. Control plane components make the decision, and data flows through the data plane.
| Component | What it does |
|---|---|
| Policy engine (PE) | The engine decides whether a subject gets access to a resource. It uses the organisation’s policy and external data, and it logs every decision. |
| Policy administrator (PA) | The administrator carries out the engine’s decision. It establishes or shuts down the connection and issues session credentials. |
| Policy enforcement point (PEP) | The enforcement point enables, monitors and terminates the connection between a subject and a resource. |
The policy engine uses data on device state, threat intelligence, activity logs and data access rules. NIST calls the way these data are combined the trust algorithm.
A criteria-based algorithm grants access when the request meets all set conditions. A score-based algorithm calculates a confidence level from weights and grants access above a threshold. A contextual algorithm takes the subject’s history into account, so it is better at detecting an account taken over by an intruder.
NIST considers the contextual algorithm ideal, although it is not always possible. The organisation chooses the variant according to what its components can do and records the criteria or weights for each resource.
Four deployment models
NIST describes four models that differ in where the enforcement point sits. One organisation can use several models for different processes.
| NIST model | Where the enforcement point sits | When it fits | Limitation |
|---|---|---|---|
| Device agent and resource gateway | An agent runs on the user’s device, and a gateway sits directly in front of the resource. | The model fits organisations that manage their devices and do not allow personal devices. | The model requires an agent on every device. |
| Enclave | A gateway sits at the boundary of a group of resources, for example a data centre. | The model fits legacy systems that cannot have their own gateway in front of them. | The gateway protects the whole group and does not protect each resource individually, and users may see resources they have no privileges for. |
| Resource portal | A single portal accepts requests and brokers access. | The model fits personal devices and collaboration with other companies, because it needs no agent. | The organisation sees the device only at connection time, and the portal is visible to attackers and exposed to denial of service. |
| Application sandboxing | Approved applications run in a separate environment on the device. | The model fits where the device may be compromised and the application has to run independently of it. | The organisation has to maintain the approved sandboxed applications on every device. |
Give a resource portal spare capacity. Why: NIST warns that a portal visible on the network attracts denial-of-service attacks.
How to introduce Zero Trust step by step
NIST describes introducing zero trust as a journey that the organisation takes gradually, one process at a time. Steps 1-3 build an inventory without which the policy engine rejects valid requests. Steps 4-7 are repeated for each further process.
Step 1. List the subjects, meaning users and service accounts.
The policy engine must know every subject that requests access, including service accounts and automated processes. Check administrator and developer accounts more closely. Why: in older architectures such accounts often have access to all resources.
Step 2. List the assets and build the ability to detect new ones.
The inventory covers hardware, accounts, applications and certificates, as well as devices outside company management that are visible on its network. A complete census is often impossible, so the organisation must quickly identify and assess newly discovered devices. Why: according to NIST, incomplete knowledge of assets most often ends in denied access in a running process.
Step 3. Rank the business processes and choose the first one.
Choose a low-risk process first. Why: a disruption in such a process will not stop the whole organisation, and the team gains experience before the critical processes. According to NIST, processes based on cloud services and processes of remote workers are good candidates.
Step 4. Set the policies for the chosen process.
List the resources the process depends on, for example the identity system and databases. Also list the systems that depend on it, for example logging and monitoring. Then set the trust algorithm criteria or weights for each resource.
Step 5. Choose a solution using the NIST questions.
| Question | What to check |
|---|---|
| Does the solution require a component on the user’s device? | Check whether the process uses personal or partner devices on which an agent cannot be installed. |
| Does the solution work when the resources sit in the company network? | Check whether the solution also controls traffic between resources inside the company network, in addition to traffic to the cloud. |
| Does the solution log interactions for analysis? | Check whether the logs feed back into the policy engine as input for later decisions. |
| Does the solution support the protocols and applications needed? | Check the list of protocols, because some solutions only handle web traffic or email. |
| Does the solution change the way users work? | Check which extra steps users get and whether the process can bear them. |
Step 6. Run the policies first in observation mode.
In observation mode the solution grants most requests and logs all connections, and the team compares them with the policy. From the start, however, block requests that fail MFA and requests from known malicious addresses. Why: the first version of a policy is rarely complete.
Step 7. Expand the architecture and reassess it after changes.
Once the first process is stable, the team chooses the next process and repeats steps 4-7. New devices, major updates and organisational changes require the policies to be reassessed. Why: a policy set for the old state allows or blocks the wrong things.
Protecting the decision components themselves
The policy engine and administrator decide on all access, so their compromise or failure stops or opens the whole organisation. Log and audit every change to their configuration. Why: according to NIST, an administrator with access to the engine’s rules can make an unapproved change or a mistake that disrupts the business.
Replicate the engine, the administrator and the enforcement points in several locations or in a secured cloud. Why: NIST points out that a denial-of-service attack may cut off access for some users, and an outage at a cloud provider for the whole organisation.
Maturity assessment according to CISA
CISA prepared the Zero Trust Maturity Model (ZTMM) for US federal agencies and recommends that other organisations consider using it. The model has five pillars: identity, devices, networks, applications and workloads, and data. Three cross-cutting capabilities, namely visibility and analytics, automation and orchestration, and governance, apply to every pillar.
| CISA stage | What the organisation looks like at this stage |
|---|---|
| Traditional | The organisation configures permissions manually, and policies are static and cover one pillar at a time. |
| Initial | The organisation starts automating attribute assignment and access decisions and links the first pillars. |
| Advanced | The organisation automates controls where possible, has central visibility of identities and changes permissions based on risk assessment. |
| Optimal | The organisation assigns attributes and permissions automatically, just in time, and continuously monitors all pillars. |
Assess the stage separately for each pillar and record it. Why: according to CISA, each pillar progresses at its own pace, and a single assessment for the whole organisation hides the weakest pillar.
Treat the model as a guide. CISA notes that the ZTMM is not a strict set of requirements. For network segmentation the model describes a path from large segments with almost free movement inside, through isolation of critical workloads, to fully distributed micro-perimeters.
Implementation in Microsoft 365 services is described in the article on Zero Trust with Microsoft 365.
Zero Trust in OT networks
The control network needs separate decisions, because the ZTMM does not address OT-specific challenges. According to NIST, some controllers and HMI panels do not support the technologies zero trust needs.
1. Apply zero trust to devices that support it, usually at the higher levels of the OT network.
NIST recommends applying zero trust to compatible devices at levels 3, 4 and 5 of the Purdue model and in the DMZ of the OT network. The Purdue model is a hierarchy of levels from sensors (level 0) to business systems. Why: NIST names these levels as the place of zero-trust-compatible devices, while controllers and HMI panels lower down often do not support the protocols needed.
2. Protect legacy devices with a gateway at the boundary of their group.
Cover controllers without authentication with the enclave model, in which a single enforcement point sits at the boundary of the whole group of devices. For OT, CISA recommends network-based segmentation and limiting traffic to and from such devices to the necessary minimum. In IEC 62443 terms the boundary of the group is a zone boundary, and its target level (SL-T) follows from a risk assessment. Zone design is described in the article on defense in depth in DCS.
3. Assess the impact of latency and component failures on the process before deployment.
NIST recommends checking whether the zero trust solution increases response time and what happens when one of its components stops working. Based on this analysis, minimise latency and provide redundant components. Why: latency or an unavailable policy engine can stop communication the process depends on.
4. Decide in the risk assessment how each flow behaves when the decision components fail.
For every flow through an enforcement point, record what happens when the policy engine is unavailable. The decision follows from the risk assessment and the consequences for the process. IEC 62443-3-3 sets limits for it:
- security measures must not adversely affect functions essential to health, safety, the environment and the availability of the installation, unless a risk assessment supports this,
- security measures must not lead to a loss of protection that endangers health, safety or the environment,
- accounts used for essential functions must not be locked out, even temporarily,
- the failure of a certificate authority in a high-availability system must not interrupt essential functions,
- blocking traffic at the boundary when a boundary device fails (SL 3) must be designed so that it does not interfere with the SIS and other safety functions.
NIST also recommends defining the thresholds at which OT components can operate with reduced connectivity. Why: loss of connectivity can trigger the installation’s move to a safe state, and the team must know in advance when that happens.
5. Keep safety systems in separate zones.
An SIS (safety instrumented system) automatically brings the process to a safe state. According to NIST, an SIS is designed to IEC 61511 and often operates independently of the control system. IEC 62443-3-2 requires safety systems to sit in zones separated from the others. Assess every change to the SIS communication path in the management of change process before it reaches the installation.
6. Replace shared accounts with personal accounts where possible.
NIST notes that OT uses shared credentials, which makes zero trust harder to implement. Record shared accounts that have to stay, for example at operator stations, in the risk assessment. Why: the policy engine cannot assess the history of a subject it cannot tell apart.
The principles of remote access to the OT network are described in the article on secure remote access to ICS.
Checklist
- All subjects, including service accounts, are listed, and privileged accounts get additional scrutiny.
- All assets are listed, and the organisation detects and assesses new devices on the network.
- Business processes are ranked, and the first chosen process is low-risk.
- Trust algorithm criteria or weights are set for every resource in the process.
- The solution went through the NIST questions on agents, resource location, logging, protocols and changes to work.
- Policies ran in observation mode and from the start blocked requests without MFA and from malicious addresses.
- Configuration changes to the policy engine and administrator are logged and audited.
- The engine, the administrator and the enforcement points are redundant across several locations.
- The CISA maturity stage is assessed separately for each pillar.
- In the OT network zero trust covers compatible devices at levels 3-5 and in the DMZ, and a gateway at the zone boundary protects legacy devices.
- The behaviour when the policy engine is unavailable is set in the risk assessment for every OT flow.
- Safety systems sit in separate zones, and changes to their communication go through management of change.
Where we help
A security architect builds the inventory of subjects, assets and processes and designs a Zero Trust Architecture for selected processes, in IT and at the higher OT levels. In industrial networks, OT network segmentation defines zones and gateways for devices that do not support zero trust. An IEC 62443 audit checks whether the failure behaviour of security measures follows from a risk assessment and does not disrupt essential functions. You can reach us through the contact form.
Technical basis
| Step / topic | Where it is |
|---|---|
| Definition of zero trust and ZTA, implicit trust zone | NIST SP 800-207, section 2 |
| Seven tenets | NIST SP 800-207, 2.1 |
| PE, PA, PEP and data sources | NIST SP 800-207, chapter 3 |
| Four deployment models | NIST SP 800-207, 3.2.1-3.2.4 |
| Trust algorithm | NIST SP 800-207, 3.3.1 |
| Threats to the PE and PA, denial of service | NIST SP 800-207, 5.1 and 5.2 |
| Hybrid architecture | NIST SP 800-207, 7.2 |
| Steps 1-7 | NIST SP 800-207, 7.3 and 7.3.1-7.3.7 (questions: 7.3.5) |
| Pillars, cross-cutting capabilities, maturity stages | CISA ZTMM v2.0, section 5 |
| Scope of the ZTMM, OT challenges not addressed | CISA ZTMM v2.0, section 5, introduction (p. 12) |
| Network segmentation in the Networks pillar | CISA ZTMM v2.0, Networks pillar table |
| Perimeter, OT segmentation without agents | CISA, Microsegmentation Part One, 1.1 and 4.7 |
| Traffic within a zone, ZTA at levels 3-5 and in the DMZ, latency, redundancy, shared credentials | NIST SP 800-82r3, 5.2.3.4 |
| Safe state and operating thresholds when connectivity is lost | NIST SP 800-82r3, 5.3.1 |
| SIS and IEC 61511 | NIST SP 800-82r3, 2.3.7 |
| Essential functions | IEC 62443-3-3, 4.2 |
| Blocking traffic on boundary failure without interfering with the SIS | IEC 62443-3-3, SR 5.2 RE 3 (SL 3+) |
| Separation of safety systems | IEC 62443-3-2:2020, ZCR 3.3 (requirement) |
| SL-T from risk assessment | IEC 62443-3-2:2020, ZCR 5.6 |
Related articles
- Zero Trust with Microsoft 365 - implementing the CISA pillars in Microsoft 365 services
- Defense in Depth in DCS - zones, conduits and firewall rules in the control network
- Secure remote access to ICS - principles of vendor and service technician access to the OT network
- OT network segmentation - segmentation methods and the flow matrix
Sources
- NIST SP 800-207: Zero Trust Architecture (August 2020) - NIST (accessed: )
- NIST SP 800-82 Rev. 3: Guide to Operational Technology (OT) Security (September 2023) - NIST (accessed: )
- CISA - Zero Trust Maturity Model, Version 2.0 (April 2023) - CISA (accessed: )
- CISA - Journey to Zero Trust: Microsegmentation in Zero Trust, Part One (29 July 2025) - CISA (accessed: )
Need help in this area?
Our experts will help you assess the risk and plan next steps.