Full Text
1. Introduction
Information systems have undergone a profound architectural
transformation. Organizations that once operated centralized data
centres and office-based networks now depend on cloud platforms,
distributed applications, remote access technologies, mobile endpoints,
application programming interfaces and third-party services. Business
processes increasingly cross organizational boundaries, while data is
replicated across multiple environments to support availability,
analytics and collaboration. These developments have improved
flexibility and scalability, but they have also expanded the attack
surface and complicated the task of determining which entities should be
permitted to access sensitive resources.
Traditional network security commonly relied on a perimeter model.
Firewalls, virtual private networks and network access controls were
used to separate a supposedly trusted internal environment from an
untrusted external environment. Once a user or device crossed the
perimeter, internal access was often governed by comparatively broad
network permissions. This approach can provide useful protection against
unsolicited external connections, but it does not adequately address
compromised credentials, malicious insiders, vulnerable endpoints,
lateral movement or the use of legitimate accounts by attackers. A
connection originating inside a corporate network may be malicious,
while a legitimate employee working remotely may connect from an
unfamiliar network. Network location therefore provides only limited
evidence about the legitimacy of a request.
The zero trust approach responds by changing the basis of security
decisions. Access is granted to a specific resource after evaluating
identity, device state, requested action, environmental conditions and
applicable organizational policy. The architecture seeks to minimize
implicit trust and restrict the consequences of compromise. The National
Institute of Standards and Technology (NIST) describes zero trust as a
collection of concepts and ideas designed to reduce uncertainty in
enforcing accurate, least-privilege, per-request access decisions in
information systems and services (Rose et al., 2020). This definition is
important because it treats zero trust as a coordinated architecture
rather than as a single network appliance or authentication
mechanism.
The central proposition is not that every interaction can be made
perfectly safe. Rather, an organization should assume that breaches may
occur and design its systems so that a compromised account, endpoint or
application does not automatically provide unrestricted access to the
wider environment. Authentication and authorization become
resource-centred, granular and informed by changing evidence. A user who
has successfully authenticated may still be denied access to a
particular database because the device is unmanaged, the requested
operation is unusually sensitive, or the user's role does not justify
the action.
This article provides a detailed examination of zero trust security
architecture for modern information systems. It addresses five related
questions. First, what conceptual and technical principles distinguish
zero trust from perimeter-centred security? Second, which architectural
components are needed to implement resource-level access decisions?
Third, how can identity, devices, applications, data and network
controls be integrated into a coherent model? Fourth, what
organizational, technical and ethical challenges affect implementation?
Finally, how can an organization adopt zero trust incrementally while
maintaining operational continuity? The article contributes an
integrated implementation framework that connects technical controls
with governance, risk management and organizational capability.
2.
Conceptual Foundations and Evolution of Security Architecture
2.1 The traditional perimeter
model
The perimeter model developed when many organizations operated
relatively stable internal networks, centrally managed computers and
applications hosted in controlled facilities. The model placed security
controls at network boundaries and treated the internal network as a
comparatively trusted zone. Firewalls filtered traffic according to
addresses, ports and protocols, while virtual private networks created
encrypted tunnels for remote users. Intrusion detection systems and
antivirus tools supplemented these controls. The approach was not
inherently unreasonable: a clearly defined boundary made it possible to
concentrate defensive resources and reduce exposure to external
networks.
Its limitations became increasingly visible as organizations adopted
mobile computing, cloud services and remote work. The distinction
between internal and external networks became less meaningful when
employees accessed enterprise applications from homes, hotels, partner
organizations and personal devices. Cloud applications could be reached
directly through the internet, bypassing traditional corporate gateways.
A compromised endpoint connected to the internal network could also
communicate with other systems if internal segmentation and
authorization were weak. As a result, attackers who obtained valid
credentials or established a foothold could exploit implicit trust to
move laterally.
Perimeter controls remain valuable. Firewalls, secure gateways and
network segmentation can reduce exposure and enforce traffic
restrictions. The limitation is treating the perimeter as the primary or
sufficient source of trust. Zero trust retains useful network controls
but places them within a broader framework in which every access request
is evaluated in relation to the resource and current context.
2.2 The principle of least
privilege
Least privilege requires that a subject receive only the permissions
necessary to perform an authorized task, for only as long as those
permissions are needed. Saltzer and Schroeder (1975) identified least
privilege as a foundational design principle for protecting information
systems. The principle reduces the damage that can result from
accidental misuse, compromised credentials and malicious activity. If an
account can read only a limited set of records, compromise of that
account should not automatically permit modification of the entire
database or access to unrelated systems.
In practice, least privilege is difficult to maintain. Organizations
accumulate permissions as employees change roles, projects evolve and
applications are integrated. Temporary access may become permanent,
service accounts may retain excessive privileges, and administrators may
use powerful accounts for routine activities. Zero trust encourages more
frequent evaluation of permissions, stronger separation of
administrative functions and access policies that reflect the
sensitivity of the requested resource. It also supports just-in-time
access, in which elevated permissions are granted for a limited period
and removed when the task is complete.
2.3 The principle of
explicit verification
Explicit verification means that access decisions should be based on
relevant evidence rather than on assumptions. Evidence may include the
authenticated identity, authentication strength, device registration,
endpoint health, location, time, network characteristics, requested
resource, sensitivity of the data and recent threat intelligence. No
single signal is necessarily conclusive. A familiar location does not
prove that a request is legitimate, and a valid password does not prove
that the endpoint is uncompromised.
The quality of verification depends on the reliability and freshness
of evidence. Identity information must be accurate, device inventories
must be maintained, and telemetry must be collected and interpreted
consistently. An organization that bases decisions on stale group
membership or incomplete endpoint records may create a zero trust policy
that appears sophisticated but produces weak outcomes. Verification
should therefore be supported by lifecycle processes, data quality
controls and clearly assigned ownership.
2.4 Assume breach and
limit blast radius
Assume breach is a planning principle that recognizes the possibility
that an attacker may already have gained access. It does not mean that
organizations should accept compromise as inevitable or abandon
prevention. Instead, it requires defensive design that limits the reach
and duration of an intrusion. Segmentation, restricted service
identities, application-level authorization, protected administrative
pathways and detailed audit trails can make it more difficult for an
attacker to move from one compromised component to another.
The concept of blast radius refers to the extent of damage that can
occur after a security failure. A flat network with shared administrator
credentials may allow a single compromised endpoint to affect many
systems. A segmented environment with distinct identities, narrowly
scoped permissions and monitored service-to-service communication can
constrain the same incident. Zero trust is therefore closely related to
resilience: it seeks not only to prevent unauthorized access but also to
reduce the operational consequences of failures that cannot be
prevented.
Research on zero trust draws on several established areas of computer
security. Access control research provides models for deciding which
subjects may perform which operations on which objects. Sandhu et al.
(1996) formalized role-based access control, in which permissions are
associated with organizational roles and users are assigned to those
roles. Role-based access control can simplify administration, but static
roles may not capture all contextual factors relevant to modern access
decisions. Attribute-based access control extends the decision process
by considering attributes of subjects, objects, actions and
environmental conditions. NIST's guide to ABAC explains how policies can
use these attributes to support fine-grained authorization (Hu et al.,
2014).
Identity assurance is another essential foundation. NIST Special
Publication 800-63-3 describes digital identity guidelines covering
identity proofing, authentication and federation (Grassi et al., 2017).
The guidance is relevant because zero trust depends on knowing which
user or service is requesting access and understanding the strength of
the authentication process. Bonneau et al. (2012) demonstrate that
authentication schemes should be evaluated across multiple dimensions,
including usability, deployability and security, rather than by password
complexity alone. Strong authentication is therefore necessary, but it
must be implemented in a way that users can apply consistently and
securely.
The NIST Zero Trust Architecture publication provides a widely used
reference model for translating these principles into an enterprise
architecture (Rose et al., 2020). It identifies components such as the
policy engine, policy administrator and policy enforcement point. The
policy engine makes access decisions based on policy and information
from supporting systems. The policy administrator establishes or
terminates the communication path, while the enforcement point enables,
monitors and ultimately terminates access. The model also recognizes the
role of data sources that provide identity, device, threat intelligence
and activity information.
Google's BeyondCorp work is an influential practical example of
moving access decisions away from traditional network location and
toward user and device context. Ward and Beyer (2014) describe the shift
toward a model in which access to enterprise applications is based on
information about the user and device rather than on whether the user is
connected to a corporate network. Later BeyondCorp publications describe
implementation experience and the operational systems required to
support the approach (Osborn et al., 2016). The work illustrates that a
zero trust transformation requires changes to identity systems, device
management, application access and operational processes.
NIST publications on security controls, risk assessment, incident
response and remote access provide complementary guidance. SP 800-53
Rev. 5 organizes security and privacy controls for information systems
and organizations (Joint Task Force, 2020). SP 800-30 Rev. 1 describes
risk assessment methods (Joint Task Force Transformation Initiative,
2012), while SP 800-61 Rev. 2 addresses computer security incident
handling (Cichonski et al., 2012). SP 800-46 Rev. 2 discusses enterprise
telework and remote access security (Souppaya & Scarfone, 2016).
These publications are not zero trust architectures in themselves, but
they help organizations implement the governance and operational
capabilities on which zero trust depends.
Industry guidance has also contributed to the evolution of zero
trust. The Cybersecurity and Infrastructure Security Agency's Zero Trust
Maturity Model provides a maturity-oriented view of capabilities across
identity, devices, networks, applications and workloads, and data (CISA,
2021). It encourages organizations to progress from traditional
approaches toward more integrated and automated capabilities. Such
maturity models are useful because organizations rarely move directly
from a perimeter-based environment to a fully integrated zero trust
architecture. They need a sequence of achievable improvements,
measurable outcomes and feedback mechanisms.
4. Architecture and Core
Components
4.1 Resource-centred
protection
The protected resource is the central object of a zero trust
architecture. A resource may be a database, application, virtual
machine, file, API endpoint, cloud service, administrative interface or
operational technology component. The architecture should identify the
resource, its owner, its business purpose, its sensitivity and the
conditions under which access is permitted. A policy that merely states
that an employee belongs to an internal network is too broad to provide
meaningful resource-level control.
Resource classification is a practical prerequisite. Organizations
should distinguish public information from internal business
information, confidential records, regulated personal data, financial
information, intellectual property and critical operational assets.
Classification does not have to begin with an elaborate taxonomy. A
manageable initial scheme can identify a small number of sensitivity
levels and map each level to authentication, authorization, monitoring,
encryption and retention requirements. As the asset inventory improves,
policies can become more precise.
4.2 Policy decision and
enforcement
The policy decision process evaluates whether a particular request
should be allowed. In the NIST model, the policy engine makes the
decision using enterprise policy and information from relevant data
sources. The policy administrator translates the decision into commands
that establish or terminate the required communication path. The policy
enforcement point mediates the connection between the subject and the
resource (Rose et al., 2020). Separating decision logic from enforcement
makes it possible to apply consistent policies across different
applications and network environments.
A policy decision may consider several dimensions. The subject may be
an employee, contractor, service account or workload. The device may be
managed, compliant, encrypted, patched or potentially compromised. The
action may involve reading, editing, deleting, exporting or
administering information. The resource may have a high or low
sensitivity classification. Environmental context may include time,
location, network risk and current threat indicators. Policies should
specify which signals are mandatory and how conflicting signals are
resolved.
4.3 Identity and
authentication
Identity is the foundation for attributing actions to users and
services. Human identities should be created through controlled
onboarding processes, linked to authoritative personnel records where
appropriate, and disabled promptly when access is no longer justified.
Non-human identities require equal attention. Applications, automation
scripts, containers and cloud workloads often use service accounts or
tokens to communicate with other systems. These identities should have
owners, defined purposes, limited permissions, managed credentials and
review dates.
Multi-factor authentication reduces dependence on passwords by
requiring additional evidence. Factors may include possession of a
registered device, a cryptographic authenticator or a biometric
characteristic. Authentication strength should be proportionate to risk.
Access to a low-sensitivity resource may require a standard enterprise
authentication session, while privileged access to a production database
may require phishing-resistant authentication, device compliance and
additional approval. Reauthentication can be triggered by changes in
risk rather than by arbitrary repeated prompts.
Identity federation can support consistent access across multiple
applications and organizations. Standards-based federation reduces the
need for each application to maintain separate passwords and identity
stores. However, federation creates dependencies on identity providers,
token validation, signing keys, session management and account lifecycle
processes. A compromised identity provider or poorly protected
federation configuration can have wide consequences. Organizations
should therefore protect identity infrastructure as a critical security
service and monitor administrative changes to it.
4.4 Device posture and
endpoint trust
A user identity alone does not establish that the device is safe.
Endpoint posture may include operating system version, security patch
level, disk encryption, endpoint detection status, configuration
compliance, device ownership and evidence of compromise. A managed
device can be evaluated using endpoint management and security tools.
Unmanaged devices may receive restricted access through browser
isolation, virtual desktop infrastructure or application-specific
controls. The appropriate approach depends on the sensitivity of the
resource and the organization's ability to assess the endpoint.
Device posture is dynamic. A device that was compliant yesterday may
become vulnerable after a missed update or a security agent failure.
Zero trust policies should therefore support periodic reassessment and
event-driven reassessment. If an endpoint is reported as compromised,
the architecture may revoke sessions, restrict network access, require
remediation or deny access to sensitive resources. Such responses should
be designed carefully to avoid unnecessary disruption and to preserve
access to recovery services.
4.5 Network
segmentation and secure communication
Zero trust does not eliminate network security. It changes the role
of network controls. Segmentation divides environments into smaller
zones or workloads and limits communication to explicitly required
paths. Micro-segmentation applies granular controls close to workloads
or applications, making it harder for an attacker to move laterally.
Segmentation can be implemented through network firewalls,
software-defined networking, host controls, cloud security groups or
service mesh policies. The choice depends on the technical environment
and operational capacity.
Encryption protects information in transit and, where supported, at
rest. Secure communication protocols reduce the risk of interception and
manipulation, but encryption does not determine whether a user should
have access to a resource. Authorization must still be enforced at the
application or service level. Mutual authentication between services can
help establish the identity of both communication endpoints, while
short-lived credentials and certificate lifecycle management reduce the
exposure associated with long-lived secrets.
4.6 Application and workload
security
Modern applications are composed of services that communicate through
APIs, message queues and internal protocols. Each service should
authenticate the identity of the calling workload and authorize the
requested operation. Trust should not be inherited merely because two
services run in the same cloud account or cluster. Application gateways,
API management systems and service meshes can enforce policy, but their
configuration must be consistent with application-level authorization
and secure software development practices.
Workload identity is particularly important in containerized and
automated environments. Static secrets embedded in source code or
deployment files can be exposed through repositories, logs and build
systems. More secure designs use managed identity services, short-lived
credentials, secret stores and automated rotation. Deployment pipelines
should be protected because an attacker who compromises the build
process may introduce malicious code or obtain credentials that are
trusted by production systems.
4.7 Data protection and
governance
Data protection is the ultimate purpose of many zero trust controls.
Access policies should consider not only who is requesting access but
also what data is involved and what operation is being performed. Data
loss prevention, encryption, rights management, database activity
monitoring and audit logging can supplement access decisions. Sensitive
data should be discoverable and classified, and data owners should
approve access based on business need. Data retention and deletion
policies should also be enforced because unnecessary retention increases
exposure.
A resource-level model makes it possible to distinguish access to an
application from access to particular records or fields within that
application. For example, a customer service employee may need to view a
customer's contact details but not payment credentials or health
information. A financial analyst may need aggregate transaction data but
not personal identifiers. These distinctions require application and
data-layer controls; a network gateway alone cannot enforce them
reliably.
5. Access Control
Models and Policy Design
Role-based access control remains useful where job functions are
stable and permissions can be grouped into well-defined roles. It
supports administrative efficiency and separation of duties. However,
roles can become excessively broad or proliferate when every exception
is represented as a new role. Sandhu et al. (1996) show how role
hierarchies and constraints can structure authorization, while
subsequent access control guidance recognizes the value of combining
roles with additional attributes.
Attribute-based access control allows decisions to depend on subject,
resource, action and environmental attributes. Hu et al. (2014) describe
ABAC as a flexible approach for expressing policies that can adapt to
changing context. A policy might permit a member of the finance
department to approve an invoice only when the invoice value is below a
defined threshold, the user is accessing from a compliant device, and
the transaction is within the user's assigned business unit. A
higher-value transaction could require a second approver.
A practical zero trust policy model often combines role and
attribute-based approaches. Roles provide a manageable baseline of
organizational permissions, while attributes refine decisions according
to resource sensitivity, device posture, risk and transaction context.
Policy design should avoid unnecessary complexity. Policies that are too
broad create excessive access, while policies that are too intricate can
become difficult to test, explain and maintain. A policy catalogue
should record the business owner, purpose, required evidence,
exceptions, review date and expected audit events for each important
rule.
Authorization should distinguish between authentication and
permission. Authentication establishes evidence about the identity of a
subject. Authorization determines what that subject may do. A successful
login should not automatically grant access to all functions within an
application. Applications should enforce authorization at meaningful
boundaries, including API operations, records and privileged functions.
Deny-by-default is a useful design principle: access is denied unless a
valid policy explicitly permits it. Exceptions should be documented,
time-limited and reviewed.
Policy evaluation also needs to account for emergency access.
Critical services may require break-glass procedures when identity or
policy infrastructure is unavailable. Such procedures should be narrowly
scoped, strongly protected, logged and reviewed after use. An emergency
account that remains permanently available with broad permissions
creates a significant weakness. Organizations should test emergency
access regularly and ensure that it does not become a routine bypass of
standard controls.
6. Continuous
Monitoring, Risk and Threat Response
Zero trust requires ongoing observation because the context of an
access request can change during a session. Continuous monitoring
involves collecting and analysing events from identity providers,
endpoint management systems, network devices, applications, cloud
platforms and data repositories. The purpose is not simply to accumulate
logs. Telemetry must support decisions, investigations and improvements.
NIST guidance on log management emphasizes the importance of
establishing processes for generating, transmitting, storing, accessing
and disposing of log data (Kent & Souppaya, 2006).
Risk-based access decisions should be proportionate to the
sensitivity of the resource and the potential impact of misuse. A risk
assessment considers threats, vulnerabilities, likelihood and
consequences. NIST SP 800-30 Rev. 1 provides a structured approach to
assessing information security risk (Joint Task Force Transformation
Initiative, 2012). In a zero trust environment, risk assessment informs
the policy model, while live telemetry can indicate when a particular
request requires additional verification or restriction.
Security information and event management systems can correlate
events across the enterprise. User and entity behaviour analytics may
identify unusual patterns, such as an employee downloading an unusually
large volume of data, a service account accessing a new database, or a
privileged login occurring at an unexpected time. Automated response can
revoke sessions, quarantine endpoints or block suspicious activity.
Automation should be calibrated to the confidence of the signal and the
operational impact of the response. False positives can interrupt
legitimate work, while false negatives can allow attacks to
continue.
Incident response remains necessary even when zero trust controls are
mature. NIST SP 800-61 Rev. 2 organizes incident handling around
preparation, detection and analysis, containment, eradication, recovery
and post-incident activity (Cichonski et al., 2012). Zero trust can
improve incident response by providing more granular access records,
clearer identity attribution and smaller network segments. It can also
support containment by revoking specific sessions or workload
credentials rather than disabling entire business systems.
Threat intelligence can inform policies when it identifies malicious
infrastructure, compromised credentials, newly exploited vulnerabilities
or suspicious behaviours. However, external indicators should not be
treated as unquestionable facts. Intelligence feeds can be incomplete,
delayed or inaccurate. Organizations should evaluate the source,
confidence and relevance of indicators before using them to trigger
disruptive controls. The most effective monitoring strategy combines
threat intelligence with internal context and incident response
expertise.
7. Implementation Framework
A zero trust programme should begin with executive sponsorship and
clear accountability. Senior leadership should define the business
outcomes expected from the programme, such as reducing unauthorized
access to sensitive data, improving remote access security, limiting
lateral movement or strengthening cloud governance. Security teams
cannot implement zero trust effectively if application owners,
infrastructure teams, human resources, procurement and business units do
not participate. Governance should establish decision rights, funding,
risk acceptance procedures and a mechanism for resolving conflicts
between security and operational requirements.
A steering group can coordinate the programme across cybersecurity,
IT operations, enterprise architecture, data governance, privacy, legal
and business functions. It should approve priorities, review progress
and monitor risks. Technical work should be assigned to accountable
owners, not left as a general responsibility of the security department.
Application owners should identify the users and service identities that
require access, while data owners should define classification and
acceptable use. Human resources and procurement should support reliable
joiner, mover and leaver processes for employees and third parties.
7.2 Discover and classify
assets
The organization should create an inventory of users, devices,
applications, workloads, APIs, databases, cloud resources and external
connections. Discovery should combine automated tools with owner
validation because technical inventories may not identify business
purpose or data sensitivity. Unknown assets are difficult to protect.
Unmanaged servers, forgotten test environments and undocumented service
accounts can become entry points for attackers. The inventory should
record ownership, location, dependencies, criticality, exposure and
lifecycle status.
Asset classification should identify high-value systems and data
stores first. A risk-based programme does not need to apply the same
control intensity to every resource at the same time. Systems that hold
regulated data, support financial transactions or provide administrative
control over the environment should receive early attention. The
organization should also identify critical dependencies, including
identity services, DNS, certificate authorities, endpoint management and
logging platforms. These services may be targets because compromise of a
foundational control can undermine many other protections.
7.3 Strengthen identity
foundations
Identity governance should establish authoritative identity records,
consistent authentication, controlled provisioning and timely
deprovisioning. Multi-factor authentication should be deployed for
remote access, privileged accounts and sensitive applications, with
phishing-resistant methods prioritized where feasible. Shared user
accounts should be eliminated or tightly controlled. Privileged access
management should separate routine accounts from administrative
accounts, restrict privileged sessions and record high-risk
activities.
Non-human identities should be included in the inventory and
governance model. Every service account, API key, certificate and
automation credential should have an owner, purpose, scope and
expiration or review mechanism. Secrets should be stored in approved
systems rather than source code or configuration files. Credential
rotation should be automated where possible, and unused identities
should be disabled. These steps are essential because service identities
may have persistent access and can be overlooked during conventional
user access reviews.
7.4 Introduce device
compliance controls
The next stage is to improve visibility and control over endpoints.
Organizations should identify which devices are managed, which are
personally owned, and which are unknown. Endpoint management can enforce
encryption, patching, screen-lock settings, malware protection and
configuration baselines. Device health signals should be made available
to access policy systems. Where a device cannot meet required standards,
the organization may provide restricted access to selected applications
or require remediation before access is restored.
Device compliance policies should be realistic and support business
continuity. A policy that blocks all access whenever a minor update is
pending may cause users to seek insecure workarounds. Risk tiers can
distinguish between low-risk deviations and serious security failures.
For example, a device with an outdated non-critical application might be
allowed limited access, while a device with active malware or disabled
endpoint protection should be isolated from sensitive resources.
Exceptions should have an owner and expiry date.
7.5 Segment
networks and modernize remote access
Organizations should map communication flows between applications,
workloads and data stores before introducing segmentation. Abruptly
blocking traffic without understanding dependencies can disrupt business
services. Flow records, application documentation and interviews with
system owners can help identify legitimate communication paths.
Segmentation should begin with high-value environments and
administrative pathways, then expand as the organization gains
confidence in its dependency map.
Remote access should be based on application and resource
requirements rather than broad network membership. A user who needs one
internal application should not automatically receive access to an
entire subnet. Secure access gateways, application proxies and
identity-aware access systems can provide more targeted connectivity.
Virtual private networks may still be appropriate for particular use
cases, but access should be limited and monitored. NIST SP 800-46 Rev. 2
provides guidance for securing telework and remote access technologies
(Souppaya & Scarfone, 2016).
7.6 Integrate
applications and data controls
Applications should be prioritized according to business criticality,
data sensitivity and exposure. Legacy applications may not support
modern authentication protocols or granular authorization. Integration
options include identity-aware proxies, federation gateways, application
modernization, network isolation and compensating controls. The
organization should avoid assuming that a gateway can correct every
weakness inside an application. Application-level authorization, secure
session handling and protection against common software vulnerabilities
remain necessary.
Data controls should be aligned with classification. Sensitive
records may require encryption, restricted export, activity monitoring,
data loss prevention and stronger approval workflows. Data access
reviews should identify permissions that are no longer required. Where
feasible, access should be granted at the level of records or fields
rather than entire databases. Backups and recovery environments should
also be protected because they may contain copies of sensitive
information and can be targeted during ransomware incidents.
7.7 Automate, measure and
improve
Automation should be introduced after the organization understands
its policies and data sources. Useful early use cases include automatic
disabling of departed employees, expiry of temporary permissions,
detection of dormant service accounts, removal of non-compliant devices
from sensitive access and alerts for unusual privileged activity. More
advanced automation may use risk scores to require step-up
authentication or terminate suspicious sessions. Each automated action
should have an audit trail and a defined process for reversal when the
decision is incorrect.
Implementation should be measured using operational and security
indicators. Examples include the proportion of critical applications
using centralized identity, percentage of privileged accounts protected
by phishing-resistant authentication, percentage of managed endpoints
meeting baseline requirements, number of unknown assets, proportion of
service identities with assigned owners, age of unresolved access
exceptions and time required to revoke access after termination. Metrics
should measure risk reduction and operational effectiveness, not merely
the number of products installed.
8. Comparison
with Traditional Security Approaches
Table 1 compares common characteristics of perimeter-centred
security and zero trust architecture. The comparison is not intended to
suggest that traditional controls are obsolete. Firewalls, VPNs and
network monitoring remain important. The distinction concerns the
assumptions that govern access and the degree to which decisions are
tied to individual resources and current context.
Table 1. Traditional perimeter security and zero trust
architecture
Dimension | Traditional perimeter emphasis | Zero trust
emphasis
Trust assumption | Internal network is generally trusted after entry |
No implicit trust based solely on network location
Access unit | Network or broad application zone | Specific resource,
action and session
Identity | Often one factor among several perimeter controls | Central
input to each authorization decision
Device state | May be checked at connection time | Evaluated and
reassessed as context changes
Authorization | Often broad after network admission | Least privilege
and policy-based resource access
Segmentation | Network zones and perimeter boundaries | Fine-grained
segmentation and workload-level controls
Monitoring | Emphasis on boundary traffic and alerts | Correlated
identity, device, application and data telemetry
Response | Block at network boundary or isolate broad segments | Revoke
sessions, restrict identities and contain specific resources
The comparison reveals that zero trust extends rather than simply
replaces established security practices. A firewall can function as a
policy enforcement point, and a VPN may provide encrypted connectivity,
but neither should be treated as sufficient evidence that a user is
authorized to access every resource reachable through the connection.
Similarly, multi-factor authentication is a valuable control but does
not eliminate the need for authorization, device assessment and
monitoring after login.
Table 2. Example access policy decisions
Scenario | Context and evidence | Illustrative policy outcome
Employee reads a public policy document | Verified identity; compliant
device; low-sensitivity resource | Permit standard access and log the
event
Finance officer approves a high-value payment | Strong authentication;
managed device; authorized role; transaction exceeds normal threshold |
Require second approval and enhanced audit
Contractor accesses a project workspace | Sponsor-approved identity;
time-limited contract; restricted device posture | Permit only the
assigned workspace until contract expiry
Service workload calls a customer database | Valid workload identity;
approved API; narrow data scope | Permit specified operations and deny
unrelated queries
Endpoint reports active malware | Device health signal indicates
compromise | Revoke sensitive sessions and isolate endpoint for
remediation
Administrator changes identity policy | Privileged identity; sensitive
configuration action | Require phishing-resistant authentication,
approval and detailed logging
These examples show why policy design should be connected to business
meaning. The objective is not to challenge users continuously without
reason. It is to apply stronger controls when the resource, action or
context justifies them. A well-designed system should make routine
legitimate access efficient while imposing additional checks on
sensitive or unusual activity.
9.
Organizational, Technical and Ethical Challenges
9.1 Legacy systems and technical debt
Many organizations operate applications that were designed before
modern identity federation, device posture assessment and fine-grained
authorization became common. Some legacy systems rely on shared
accounts, embedded credentials, outdated protocols or network location
as the main access control. Replacing these systems may be costly or
operationally risky. A pragmatic programme should classify legacy
systems, identify compensating controls and establish modernization
priorities. An application proxy or isolated network segment may reduce
exposure temporarily, but it should not become a permanent excuse to
ignore underlying weaknesses.
Technical debt also appears in undocumented integrations and custom
scripts. A system may depend on a service account that no current
employee understands, or a scheduled job may connect to a database using
credentials stored in a configuration file. Enforcing strict policies
without discovering these dependencies can cause outages. Discovery,
testing, staged enforcement and rollback planning are therefore
essential. Zero trust transformation should be treated as an enterprise
architecture programme, not as a quick configuration exercise.
9.2 Complexity and interoperability
A mature zero trust environment may involve identity providers,
endpoint management, cloud security tools, firewalls, proxies, API
gateways, data classification systems, logging platforms and security
orchestration. These systems must exchange accurate information and
apply compatible policy semantics. Poor integration can create
inconsistent decisions, blind spots and duplicated administration.
Vendor-specific implementations may also make it difficult to move
policies or identity data between platforms.
Open standards and documented interfaces can reduce some
interoperability problems, but standards do not guarantee consistent
implementation. Organizations should define an architecture that
specifies authoritative sources for identity and device information,
policy ownership, logging formats and integration requirements.
Procurement should evaluate how a product fits the existing
architecture, what data it collects, how it exports information and
whether it supports migration. Avoiding unnecessary duplication is
important because each additional control plane creates operational
overhead.
9.3 Cost, skills and organizational capacity
Zero trust can require investment in identity governance, endpoint
management, network redesign, application modernization, monitoring and
staff training. Licensing costs are only one part of the total cost.
Organizations must also budget for discovery, integration, testing,
policy engineering, support and ongoing maintenance. Smaller
institutions may have limited specialist staff and cannot implement
every capability at once. A maturity-based approach allows them to
prioritize controls with the greatest expected reduction in risk.
Skills are a significant constraint. Zero trust programmes require
collaboration among network engineers, identity specialists, cloud
architects, application developers, data governance professionals,
privacy officers and incident responders. Training should be
role-specific. Developers need to understand workload identity and
authorization, while help-desk staff need procedures for authentication
recovery and device remediation. Senior managers need to understand that
zero trust changes business processes as well as technical
infrastructure.
9.4 User experience and accessibility
Security controls that create excessive friction can encourage
insecure behaviour. Users may share credentials, disable controls or
move work to unapproved services if legitimate tasks become
unnecessarily difficult. Authentication should be proportionate to risk,
accessible to users with different needs and supported by reliable
recovery processes. Risk-based step-up authentication can reduce
repeated prompts for low-risk activity while strengthening controls for
sensitive actions.
Accessibility should be considered during the selection of
authenticators and device policies. Not every user can rely on the same
biometric or mobile device method. Alternative secure methods should be
available, and recovery procedures should not undermine the security of
the primary process. Usability testing with representative users can
identify obstacles before broad deployment. The goal is to make secure
behaviour the easiest practical path.
9.5 Privacy and employee monitoring
Zero trust systems may collect detailed information about identity,
device health, location, application usage and user behaviour. Such
information can improve security, but it can also create privacy risks
if collected excessively or used for purposes unrelated to security.
Organizations should define what telemetry is necessary, who can access
it, how long it is retained and under what circumstances it may be used.
Transparency and proportionality are important, particularly where
monitoring could affect employee rights or expectations.
Behavioural analytics should not be treated as infallible evidence of
misconduct. Unusual activity may result from legitimate work patterns,
travel, disability-related technology use or changes in job
responsibilities. Automated decisions should be explainable enough for
security teams to investigate and correct errors. Privacy review, access
controls on monitoring data and documented retention schedules help
prevent security telemetry from becoming an uncontrolled employee
surveillance system.
9.6 Availability and dependence on central services
Centralized identity and policy infrastructure can become a critical
dependency. If an identity provider, policy engine or certificate
service becomes unavailable, legitimate access may be interrupted.
Attackers may also target these services because they influence access
across many applications. Resilience planning should include redundancy,
protected administrative access, secure backup configurations, tested
recovery procedures and carefully controlled emergency access. High
availability must be balanced with the need to avoid fail-open behaviour
that grants broad access when policy systems fail.
Organizations should define failure modes explicitly. For some
low-risk resources, a limited cached authorization decision may be
acceptable for a short period. For highly sensitive systems, inability
to verify current policy may justify denying access. These decisions
should be based on business impact and documented risk acceptance.
Testing should include identity provider outages, network partitions,
expired certificates and loss of telemetry sources.
10. Zero
Trust in Cloud, Hybrid and Emerging Environments
Cloud computing makes resource-centred security particularly relevant
because applications and data may be distributed across multiple
providers and regions. Cloud platforms provide identity and access
management services, workload identities, security groups, logging and
encryption capabilities. These capabilities can support zero trust, but
cloud accounts and subscriptions must be governed consistently.
Excessive permissions, exposed storage, insecure API keys and poorly
configured trust relationships can undermine the architecture.
Hybrid environments create additional challenges because on-premises
systems and cloud services may use different identity stores, network
designs and logging systems. Organizations should establish common
identity lifecycle processes and consistent authorization principles
while recognizing that enforcement mechanisms may differ. A single
policy language may not be practical across every platform, but the
business meaning of the policy should remain consistent. For example, a
contractor should not receive broader access merely because a particular
application is hosted in a different environment.
Internet of Things and operational technology environments introduce
devices with limited processing power, long service lives and
specialized communication protocols. Some devices cannot support modern
endpoint agents or frequent authentication exchanges. A zero trust
approach in these environments may rely on device identity, network
segmentation, protocol allowlists, gateways and strict control of
administrative access. Risk assessments must account for safety and
availability, since aggressive security enforcement could disrupt
physical processes.
Artificial intelligence and automated decision systems also increase
the importance of workload identity and data governance. Machine
learning pipelines may access sensitive datasets, model repositories and
inference services. Each component should have a defined identity and
limited permissions. Training data, model artefacts and inference
endpoints should be protected against unauthorized modification and
extraction. Automated agents should not receive broad access simply
because they operate without direct human interaction. Their permissions
should be scoped to approved tasks, and high-impact actions should
require suitable controls and auditability.
Application programming interfaces are another important area. APIs
expose business functions and data to internal applications, partners
and external developers. API keys should not be treated as permanent
proof of trust. Strong designs use authenticated identities, narrowly
scoped tokens, request validation, rate limits, authorization at the
operation and data level, and monitoring for unusual usage. Tokens
should have appropriate lifetimes and be protected from leakage through
logs, browser storage or source repositories.
11. Evaluation, Metrics and
Maturity
A zero trust programme should be evaluated by the degree to which it
reduces risk and improves the organization's ability to control access.
Counting deployed products is insufficient. A company may own several
identity and security platforms while still having unmanaged service
accounts, broad administrative permissions and unknown assets.
Evaluation should therefore combine capability measures, operational
outcomes and evidence of policy effectiveness.
Useful identity measures include the proportion of users covered by
strong authentication, the proportion of privileged accounts managed
through privileged access controls, the percentage of service identities
with named owners and the time required to remove access after
termination. Device measures may include the percentage of endpoints
enrolled in management, compliance rates, patch latency and the number
of unknown devices attempting to access sensitive resources. Application
measures may include the percentage of critical applications using
centralized identity, the coverage of application-level authorization
and the number of high-risk legacy systems operating under compensating
controls.
Network and workload measures can include the percentage of critical
workloads with documented communication paths, the number of unnecessary
network connections removed, the coverage of workload identity and the
frequency of segmentation rule reviews. Data measures may include the
percentage of sensitive repositories classified, the proportion of
sensitive data stores with access reviews, and the number of excessive
permissions identified and corrected. Monitoring measures may include
log source coverage, alert investigation time, mean time to revoke
compromised credentials and the proportion of automated actions with
documented review procedures.
Metrics should be interpreted carefully. A high rate of blocked
access attempts may indicate effective enforcement, but it may also
indicate poorly designed policies that prevent legitimate work. A
reduction in alerts may reflect improved security or reduced visibility.
Quantitative indicators should be complemented by interviews, access
reviews, incident analysis and controlled testing. Organizations should
periodically test whether a compromised account can reach unrelated
resources, whether privileged activity is attributable to an individual,
and whether a device reported as compromised is actually prevented from
accessing sensitive systems.
Maturity models can help organizations sequence improvements. CISA's
Zero Trust Maturity Model organizes capabilities across major pillars
and describes progression from traditional to more advanced states
(CISA, 2021). Organizations should use maturity models as planning aids
rather than as compliance scorecards. A smaller institution may achieve
meaningful risk reduction through strong identity governance, protected
endpoints and segmentation of critical systems even if it has not
automated every policy decision. The appropriate target state depends on
the threat environment, business model, regulatory obligations and
available resources.
12. Proposed
Integrated Implementation Model
The framework proposed in this article combines governance, asset
visibility, identity, device posture, resource protection, monitoring
and continuous improvement. It is designed as a sequence of
interdependent workstreams rather than a rigid product deployment order.
The organization begins by defining business objectives and identifying
critical resources. It then establishes reliable identity and asset
information, introduces stronger authentication and endpoint controls,
and progressively applies resource-level authorization and segmentation.
Monitoring and incident response are integrated throughout the
process.
Figure 1. Conceptual zero trust access flow
User or workload request
|
v
Identity and authentication validation
|
v
Device posture and workload integrity assessment
|
v
Context collection: resource, action, risk, time and environment
|
v
Policy engine evaluates explicit authorization rules
|
+------ Deny / challenge / request approval
|
v
Policy administrator configures the approved access path
|
v
Policy enforcement point mediates resource access
|
v
Continuous telemetry, session reassessment and audit logging
|
v
Revoke, restrict or renew access as conditions change
The flow begins with a request from a human or non-human identity.
The identity service validates the identity and authentication strength.
Device or workload systems provide evidence about the endpoint or
execution environment. Contextual information describes the requested
resource, operation and current risk. The policy engine evaluates the
evidence against policy. If access is approved, the policy administrator
establishes the required path and the enforcement point mediates the
session. Monitoring continues during access, allowing the system to
respond when risk changes.
Figure 2. Enterprise implementation workstreams
Governance and risk ownership
|
Asset discovery and data classification
|
Identity lifecycle and privileged access
|
Device compliance and endpoint protection
|
Application, API and workload authorization
|
Network segmentation and secure connectivity
|
Data protection, monitoring and incident response
|
Measurement, assurance and continuous improvement
The workstreams are mutually reinforcing. Asset discovery identifies
what must be protected. Classification determines the strength of
controls. Identity governance establishes who or what may request
access. Device and workload controls provide evidence about the
requesting environment. Application and network controls enforce
decisions, while data protection reduces the impact of unauthorized
access. Monitoring and measurement identify gaps and guide further
investment.
A practical implementation can be divided into four phases. Phase One
establishes governance, asset visibility, identity hygiene and a
baseline risk assessment. Phase Two strengthens authentication,
privileged access, endpoint management and logging for critical
resources. Phase Three introduces granular application authorization,
workload identity, segmentation and data-centric controls. Phase Four
expands automation, continuous assessment, advanced analytics and
regular assurance testing. Each phase should include pilot environments,
user feedback, technical validation and rollback procedures.
The model should be adapted to organizational size and complexity. A
university may prioritize student information systems, research
repositories, identity federation and access by external collaborators.
A financial institution may focus on privileged access, payment systems,
customer data and third-party connections. A small business may begin
with centralized identity, multi-factor authentication, managed
endpoints, cloud application controls and restricted administrator
accounts. The principles remain consistent, but the order and depth of
implementation should reflect risk and capacity.
13. Discussion
The analysis indicates that zero trust is most effective when it is
treated as an enterprise security architecture supported by governance
and operational discipline. The concept is sometimes reduced to a
product category or a marketing label, but this interpretation obscures
the work required to make access decisions accurate. A proxy or
identity-aware gateway may be one enforcement component, but it cannot
compensate for inaccurate identity records, unknown assets, excessive
permissions or weak application authorization. The architecture must
connect policy to reliable evidence and enforce decisions at meaningful
resource boundaries.
The literature also suggests that zero trust is not a rejection of
all network-based security. Network controls remain useful for reducing
exposure, limiting communication and protecting infrastructure. The
change is that network membership no longer serves as a sufficient basis
for trust. This distinction matters because organizations may be tempted
to replace one perimeter product with another and describe the result as
zero trust. A genuine transformation changes how identities are managed,
how applications authorize actions, how devices are assessed and how
activity is monitored.
A second implication concerns the relationship between security and
usability. Strong controls are more likely to succeed when they are
integrated into normal workflows. If authentication is unreliable,
device requirements are poorly explained or access policies are
inconsistent, users may develop workarounds. Risk-based policies can
help by reserving additional verification for sensitive actions and
unusual circumstances. The organization should also provide accessible
authentication options and effective support for recovery.
Third, zero trust depends on information quality. A policy engine
cannot make sound decisions if it receives incorrect role assignments,
stale device status or incomplete asset classifications. The
architecture therefore requires data governance and clear ownership.
Security teams should work with human resources, application owners,
data stewards and procurement to maintain accurate records. This may be
less visible than deploying a new security tool, but it is fundamental
to the reliability of the system.
Fourth, implementation should be incremental and evidence-driven.
Large-scale enforcement without discovery can disrupt essential
services. Organizations should begin with critical resources, test
policies in observation or report-only modes where available, identify
dependencies and then enforce controls in stages. Pilot projects should
be selected for their business value and technical representativeness.
Lessons from each pilot should inform policy templates and integration
standards before expansion.
Finally, the architecture should be evaluated in terms of resilience
and containment as well as prevention. A zero trust environment should
make it harder for an attacker to move laterally, use compromised
credentials across unrelated services or maintain access after a risk
signal is detected. Incident exercises can test these claims.
Organizations should measure how quickly they can identify the affected
identity, revoke sessions, isolate a device, restrict workload
communication and restore trusted operations.
14. Conclusion
Zero trust security architecture responds to the limitations of
traditional perimeter-based security in environments where users,
devices, applications and data are distributed across organizational and
cloud boundaries. Its central principles are explicit verification,
least privilege, resource-centred authorization, continuous monitoring
and the assumption that compromise may occur. These principles are
implemented through coordinated identity, device, application, network,
workload and data controls.
The article has shown that zero trust is not a single technology and
should not be equated with the deployment of a particular gateway,
firewall or authentication product. It is an architectural approach that
connects policy decisions to reliable evidence and enforces those
decisions throughout the information system. Its success depends on
asset discovery, identity governance, secure application design, device
management, segmentation, telemetry, incident response and
organizational accountability.
The proposed implementation model recommends a phased, risk-based
transformation. Organizations should begin with governance and
visibility, strengthen identity and endpoint foundations, apply granular
controls to critical resources, and expand automation and continuous
assurance as capabilities mature. They should measure outcomes such as
reduced excessive access, faster revocation, improved visibility and
smaller incident blast radius. They should also address privacy,
usability, availability and interoperability as core design
requirements.
Future work may examine empirical measures of zero trust
effectiveness across different organizational sectors, the economics of
migration from legacy architectures, privacy-preserving behavioural
analytics, and the application of workload identity to autonomous
software agents and operational technology. Further research should
distinguish between maturity indicators and demonstrated security
outcomes. Ultimately, zero trust is valuable not because it promises
that trust can be eliminated, but because it makes trust explicit,
limited, evidence-based and continuously reviewable.
References
Bonneau, J., Herley, C., van Oorschot, P. C., & Stajano, F.
(2012). The quest to replace passwords: A framework for comparative
evaluation of web authentication schemes. 2012 IEEE Symposium on
Security and Privacy, 553–567. https://doi.org/10.1109/SP.2012.44
Cichonski, P., Millar, T., Grance, T., & Scarfone, K. (2012).
Computer security incident handling guide (NIST Special Publication
800-61 Rev. 2). National Institute of Standards and Technology.
https://doi.org/10.6028/NIST.SP.800-61r2
Cybersecurity and Infrastructure Security Agency. (2021). Zero trust
maturity model. U.S. Department of Homeland Security.
https://www.cisa.gov/zero-trust-maturity-model
Ferraiolo, D. F., & Kuhn, D. R. (1992). Role-based access
controls. In Proceedings of the 15th National Computer Security
Conference (pp. 554–563).
Gilman, E., & Barth, D. (2017). Zero trust networks: Building
secure systems in untrusted networks. O'Reilly Media.
Grassi, P. A., Garcia, M. E., & Fenton, J. L. (2017). Digital
identity guidelines (NIST Special Publication 800-63-3). National
Institute of Standards and Technology.
https://doi.org/10.6028/NIST.SP.800-63-3
Hu, V. C., Ferraiolo, D., Kuhn, R., Schnitzer, A., Sandlin, K.,
Miller, R., & Scarfone, K. (2014). Guide to attribute based access
control (ABAC) definition and considerations (NIST Special Publication
800-162). National Institute of Standards and Technology.
https://doi.org/10.6028/NIST.SP.800-162
Joint Task Force. (2020). Security and privacy controls for
information systems and organizations (NIST Special Publication 800-53
Rev. 5). National Institute of Standards and Technology.
https://doi.org/10.6028/NIST.SP.800-53r5
Joint Task Force Transformation Initiative. (2012). Guide for
conducting risk assessments (NIST Special Publication 800-30 Rev. 1).
National Institute of Standards and Technology.
https://doi.org/10.6028/NIST.SP.800-30r1
Kent, K., & Souppaya, M. (2006). Guide to computer security log
management (NIST Special Publication 800-92). National Institute of
Standards and Technology. https://doi.org/10.6028/NIST.SP.800-92
Osborn, B., McWilliams, J., Beyer, B., & Salton, D. (2016).
BeyondCorp: Design to deployment at Google. ;login:, 41(3), 28–34.
Rose, S., Borchert, O., Mitchell, S., & Connelly, S. (2020). Zero
trust architecture (NIST Special Publication 800-207). National
Institute of Standards and Technology.
https://doi.org/10.6028/NIST.SP.800-207
Saltzer, J. H., & Schroeder, M. D. (1975). The protection of
information in computer systems. Proceedings of the IEEE, 63(9),
1278–1308. https://doi.org/10.1109/PROC.1975.9939
Sandhu, R. S., Coyne, E. J., Feinstein, H. L., & Youman, C. E.
(1996). Role-based access control models. Computer, 29(2), 38–47.
https://doi.org/10.1109/2.485845
Shostack, A. (2014). Threat modeling: Designing for security.
Wiley.
Souppaya, M., & Scarfone, K. (2016). Guide to enterprise
telework, remote access, and bring your own device (BYOD) security (NIST
Special Publication 800-46 Rev. 2). National Institute of Standards and
Technology. https://doi.org/10.6028/NIST.SP.800-46r2
Stallings, W. (2018). Effective cybersecurity: A guide to using best
practices and standards. Addison-Wesley Professional.
Verizon. (2022). 2022 data breach investigations report. Verizon.
Ward, R., & Beyer, B. (2014). BeyondCorp: A new approach to
enterprise security. ;login:, 39(6), 6–11.
Whitman, M. E., & Mattord, H. J. (2021). Principles of
information security (7th ed.). Cengage Learning.
Zissis, D., & Lekkas, D. (2012). Addressing cloud computing
security issues. Future Generation Computer Systems, 28(3), 583–592.
https://doi.org/10.1016/j.future.2010.12.006