Frequently Asked Questions about Zero Trust
Use these questions as a working guide to decisions, not a test of how many security expressions you can remember. The topics move from purpose and leadership through delivery, identities, people and resilience to evidence and future investment. Read the opening topics if you need a shared understanding; go directly to the relevant question if you are preparing a specific decision. The answers provide practical recommendations to adapt, not a universal implementation sequence or a guarantee of results.
For each question you use, identify the service concerned, invite the people affected and write down the decision needed. Record what you already know, what remains uncertain, who approves the decision, who carries out the action and when the evidence will be reviewed. Service owners, engineers, support staff, procurement and privacy colleagues will contribute different parts of the answer. Bring those perspectives together before choosing a control. Someone who explains an awkward dependency or a difficult user experience is supplying useful evidence, not obstructing progress.
Read the measurement and implementation-evidence questions together when evaluating benefits. Distinguish a feature being available, a control working in a test and an organisational outcome being observed. The myth-busting section complements this guide by challenging assumptions that can distort those decisions. Supporting guidance was consulted on 16 November 2026; older foundational or under-review material is identified where relevant. Check applicable requirements and product behaviour before implementation. The aim is an informed next step with accountable ownership, not a longer shopping list or a more decorative dashboard.
Zero Trust foundations and leadership
Zero Trust means that connection to a network does not, by itself, justify access to everything reachable from it. The NCSC architecture principles instead emphasise identities, device and service health, access policies and monitoring. In plain English, establish who or what is asking, what they need and whether the circumstances justify the request. This is a way of making access decisions, not a declaration that every colleague is secretly auditioning for a criminal career.
Start by choosing a service you understand. Ask its owner to identify the people, applications and suppliers that need access, the work they perform and the consequences of an incorrect decision. Separate permission to read ordinary information from permission to change sensitive records or administer the service. I would write those distinctions down before discussing products. Otherwise, the technology risks enforcing assumptions that nobody has properly examined, only rather more consistently than before.
Next, decide which evidence should influence access and how uncertainty should be handled. A known identity, a managed device and an approved task may justify one level of permission; missing device information may require a different route. These are illustrative policy choices, not universal rules. Test both legitimate work and requests that should be refused. Include a route for reviewing mistakes, because an unexplained refusal is not an acceptable support model for an essential service.
Finally, assign ownership for maintaining the policy as people, services and risks change. Keep records sufficient to explain important decisions without collecting unrelated personal information. Review whether controls actually limit unnecessary access while allowing the intended work. My practical test is whether a service owner can explain the permission, a technical team can enforce it and a user can obtain help when it goes wrong. If those conditions are absent, improve the design before describing the architecture as mature. Record the first review decision and make its accountable owner visible to everyone involved.
Zero Trust is the wider approach to architecture and access risk. Zero Trust Network Access, or ZTNA, is an application-access capability within that wider picture. As Section 3 explains, it can provide access to particular applications rather than broad network connectivity. The NCSC warns against treating Zero Trust as a single product. The distinction matters because an access service cannot, by itself, settle every question about permissions, data protection, endpoint security or recovery.
I would begin a ZTNA assessment with a precise use case. Identify the users, devices, application and current access path, then explain the exposure you want to reduce. An illustrative case might involve a supplier who needs one maintenance interface but currently receives much wider connectivity. Establish what should remain possible and what should become impossible. This gives the assessment a purpose beyond replacing a familiar acronym with a newer one and hoping the invoice represents progress.
Ask the provider to distinguish policy decisions from their enforcement. Which identity and device signals are actually available? What happens when a signal is missing, stale or contradictory? Can an existing session be restricted or revoked, and under which conditions? Do not infer these capabilities from a general statement about continuous verification. Test application compatibility, support arrangements and emergency access using authorised scenarios that represent the service, including legitimate activity that does not resemble a polished demonstration.
Keep the wider responsibilities visible throughout. Someone must still own accounts, review entitlements, investigate suspicious behaviour and restore the service after disruption. Record dependencies on identity systems and suppliers, together with the evidence available during an incident. Proceed when the proposed access model addresses the defined exposure without unacceptable operational consequences. ZTNA may be an excellent part of the answer. Calling it the entire answer merely conceals the work left for people who were probably busy enough already. Give each remaining responsibility an owner before approving the proposed access service.
I would use shared governance, name an accountable role with approval authority for each material decision, and assign delivery responsibilities to the teams doing the work. The NCSC describes adoption as a strategic commitment, involving changes to how systems are designed and operated. Leaders should agree priorities and acceptable trade-offs. Service owners should be accountable for service outcomes and access-risk decisions within their delegated authority. Technology teams should design and run the controls, while security and risk colleagues challenge assumptions and assess evidence.
Translate that division into decisions people can recognise. Who approves access policy for a critical service? Who funds remediation of an identity-data problem? Who can accept an exception, and who reviews it later? Who decides whether a pilot is ready to expand? I would record the named accountable decision owner, responsible delivery teams, evidence required and escalation route for each significant decision. Avoid a committee structure that discusses everything but cannot authorise the work needed to resolve a known risk.
Include procurement, privacy, support and affected users early enough to influence the design. A contract can affect access to incident evidence; a recovery procedure can determine whether someone completes an essential task. These are not secondary concerns to be handed over once the architecture is finished. Ask service owners to describe the customer or public impact of both compromise and control failure. That gives technical decisions a practical context and helps prevent security becoming an internal exercise in protecting its own dashboard.
Review governance through actual decisions rather than attendance lists. Check whether exceptions have owners, disagreements are resolved and lessons change delivery plans. Give the board a concise account of exposure, progress, unresolved dependencies and decisions requiring sponsorship. Retain enough detail for technical assurance without expecting directors to administer the policy engine. My test is whether the right person can make a timely, informed decision when circumstances become awkward. Asking the security team to deliver everything alone is not accountability; it is optimistic delegation with an incident waiting to clarify the arrangement.
Potentially, but a smaller organisation should begin with its needs rather than an enterprise transformation label. The NCSC recommends considering specific risks, existing capabilities and targeted alternatives before committing to a wider Zero Trust programme. My advice is to prioritise controls the organisation can operate reliably. A modest, well-supported access arrangement is preferable to an ambitious collection of services that nobody has time to configure, review or recover when something goes wrong.
Choose the services whose loss or misuse would hurt customers, staff or the organisation most. Identify who administers them, which suppliers can connect and whether access is removed when work ends. Ask whether existing subscriptions already provide useful identity, authentication or logging capabilities. Do not assume an available feature is correctly configured, but do not buy a second product before understanding the first one either. Enterprise-sized invoices are not a recognised maturity measure, however confidently the sales presentation implies otherwise.
I would select a manageable improvement, assign an owner and establish a support route. That might mean stronger authentication for important accounts, removing unnecessary administrator permissions or limiting a supplier to the application they actually maintain. These are illustrative priorities, not a prescribed sequence. Test the change with ordinary users, including someone who loses an authenticator or works outside office hours. Include the cost of setup, training, reviews and recovery, not just the monthly licence charge.
If internal capacity is limited, ask a service provider to explain responsibilities and evidence access in plain language. Retain enough understanding to challenge the arrangement and make decisions about your own services. Review whether the improvement addresses the original risk without creating disproportionate disruption or dependency. Expand only when the operating effort is sustainable. Smaller organisations deserve proportionate protection, not a scaled-down imitation of every large-company committee. The useful outcome is more controlled access and greater confidence in continuity, with an honest account of what remains uncovered.
Planning, investment and existing technology
Start with a service that matters and an access risk you can explain without opening a product catalogue. The NCSC migration guidance recommends understanding users, devices, services and data, alongside methodical phased planning. Its adoption guidance also asks whether targeted controls could address the actual problem. My preference is a bounded pilot where ownership is clear, dependencies can be investigated and the organisation can learn without placing essential delivery under unnecessary strain.
Bring the service owner, technical team, security colleagues and representatives of affected users into the initial discussion. Agree what harm you are trying to prevent and which access path contributes to it. Map routine work, privileged activity, supplier connections and emergency requirements. An illustrative starting point is restricting contractor access to one application. Do not select a pilot solely because the technology is convenient or its users are unlikely to complain. That would produce a rather flattering but unhelpful experiment.
Before making the change, capture a baseline. Record existing permissions, tested access paths, support demand and any known exceptions. Define what success would look like, including legitimate work continuing and inappropriate access being refused. Agree who can stop the pilot, restore the previous arrangement or approve a temporary exception. Keep test scenarios safe and authorised. Where the service has safety implications, involve the relevant engineers and operators rather than treating their concerns as obstacles to the security programme.
At the review, examine evidence against the original objectives. Describe failures, missing observations and user difficulties alongside improvements. Decide whether to adapt, broaden or stop the approach, and record why. A useful pilot should leave behind an owned policy, a tested support route and a realistic account of operating effort, not merely a completed installation. I would extend the pattern only when those lessons are understood. Starting small is not a lack of ambition; it is a way of ensuring ambition remains connected to reality.
No. Replacement should follow a demonstrated gap or a justified change in operating model, not the arrival of a fashionable label. The NCSC describes Zero Trust as a combination of controls and designs, rather than a single purchase. Existing identity systems, firewalls, endpoint protections and monitoring may still perform useful functions. My starting question is what each control actually does for the services in scope, and whether that function remains necessary during and after the proposed change.
Create a simple control map for the pilot. Record the risk addressed, the owner, the systems covered, the evidence produced and the consequences if the control fails. Identify overlaps without assuming that apparently similar tools provide identical protection. Ask where there are genuine gaps, conflicting policies or unsupported dependencies. A catalogue of products is not enough. The point is to understand which decisions and protections the organisation can rely on, including any functions that are not obvious from the product name.
Where replacement looks sensible, define acceptance tests before removing the existing arrangement. Include legitimate access, prohibited access, monitoring, recovery and failure conditions. Compare support effort and user experience as well as technical capability. I would use a phased transition with clear authority to pause or roll back where practical. Record temporary duplication and its end conditions, because a migration can otherwise leave both old and new complexity in permanent residence. Neither nostalgia nor novelty should receive an automatic casting vote.
Retire a control only when its required function is demonstrably covered or a responsible decision establishes that it is no longer needed. Preserve necessary records and update operating procedures, contracts and support knowledge. Review whether removing the old path leaves an unintended bypass or an overlooked service without protection. The result should be a clearer, supportable architecture, not simply a smaller procurement list. Security equipment does not become ineffective when its branding becomes unfashionable, and replacement is not progress unless the evidence explains what became better.
Identify the exact constraint before deciding that an older system must be replaced or can be ignored. The NCSC architecture principles emphasise understanding the estate and legacy integration limitations. A system may lack modern authentication, useful monitoring or a supported way to restrict connections. Those are different problems and may require different responses. My starting point is to understand the service it supports, its dependencies and the consequences of interruption. Age is context, not a complete risk assessment.
Bring the service owner and people who actually operate the system into the discussion. Map necessary users, supplier activity, administration routes and data flows. Ask which controls already reduce exposure and where an access decision can realistically be enforced. The NCSC’s implementation collection includes guidance on mixed estates, recognising that not every component can implement every principle directly. I would assess a controlled access route, isolation or additional monitoring against the specific limitation, rather than assume an overlay makes the underlying weakness disappear.
Document compensating controls, residual exposure, accountable ownership and a review date. Test that the arrangement permits necessary work and does not create an alternative path around the intended restriction. Check recovery and emergency maintenance before enforcement changes. For systems linked to physical processes, involve engineers and safety specialists and use approved testing methods. Avoid active experiments simply because they would be routine elsewhere. The point of reducing risk is not to demonstrate enthusiasm by interrupting the service before an attacker has the opportunity.
Agree a realistic longer-term decision: upgrade, replacement, retirement or continued operation with an explicitly accepted constraint. Include funding, dependencies and evidence needed to revisit that decision. Review exceptions when suppliers change support or service requirements evolve, not only when a calendar reminder happens to appear. An exception should describe a managed position, not provide permanent shelter from scrutiny. Progress may be incremental, but it should remain visible. Calling a workaround a compensating control is reasonable; declaring the legacy problem solved requires evidence that the actual limitation has been addressed.
There is no responsible universal price or completion date for Zero Trust. Scope, legacy dependencies, existing controls and operating capability all affect the work. The NCSC warns that adoption can be costly, disruptive and resource-intensive when objectives are poorly understood. My preference is to estimate a defined phase with explicit assumptions, then revise the wider plan using evidence from delivery. A precise figure produced before anyone understands the services may look reassuring, but precision and accuracy are not interchangeable professional qualifications.
Define the outcome and scope before requesting prices. Identify the identities, applications, access paths and supporting services involved, together with what will remain outside the phase. Ask technical teams to identify integration work, data-quality problems, testing and potential service changes. Include discovery where uncertainty is significant rather than burying it inside an optimistic delivery allowance. Compare a targeted improvement with the broader architectural option. The sensible investment may be a smaller control change if it addresses the material exposure without creating disproportionate operational demands.
Build a whole-service estimate. Include licences, integration, configuration, training, support, monitoring, resilience exercises and ongoing reviews. Account for internal staff time, supplier dependencies, temporary parallel operation and the work needed to retire replaced arrangements. I would distinguish one-off expenditure, recurring cost and uncertain items, with an owner for validating each assumption. Ask what happens to cost if the number of users, data volume or required features changes. A favourable licence price can be entirely genuine while telling you remarkably little about the cost of operating the service.
Use decision points rather than treating an initial estimate as an unchangeable pledge. Agree acceptance criteria, budget boundaries and conditions for pausing or expanding. Compare actual effort and outcomes with assumptions, explaining differences openly. Do not claim financial return from unverified avoided breaches. Seek an affordable improvement and an honest forecast, not a confident number followed by a quietly lengthening list of exclusions.
Identity and third-party access
Choose multi-factor authentication by examining the attacks it resists, the services it supports and the experience of the people using it. MFA combines evidence from different types of factor rather than relying on a password alone. CISA’s foundational 2022 guidance, now archived, recommends phishing-resistant MFA and distinguishes it from number matching, an interim mitigation for push-notification fatigue. I would preserve that distinction while checking current provider documentation and the authentication options actually supported in the environment.
Begin with important accounts and services, but avoid turning prioritisation into a permanent exemption for everyone else. Identify administrators, sensitive business functions, contractors and recovery staff. Check compatibility, accessibility, device availability and working conditions before choosing methods. Ask whether people must use personal equipment and what alternative is available. A solution that assumes everybody has the same phone, dexterity, connectivity or confidence is not yet a complete design. Good assurance needs a workable route for legitimate users, not just an impressive preferred method.
Test the whole lifecycle: enrolment, ordinary use, replacement, loss, suspected compromise and departure. Pay particular attention to recovery and help-desk verification. I would require an approved process for restoring access that does not casually bypass the protection being introduced. Use authorised exercises to check how staff handle an unexpected request to reset an important account. Keep the exercise focused on the process rather than embarrassing individuals. Record what evidence is needed, who can approve exceptions and how those exceptions are reviewed.
Measure coverage by method and service, not merely by a single MFA-enabled total. Track failed enrolment, avoidable prompts, recovery delays and unresolved accessibility problems alongside control tests. Where an interim method is necessary, document the reason, additional safeguards and migration review. Do not apply a vendor’s headline reduction in account compromise to overall organisational cyber risk. The practical aim is stronger, supportable authentication within a wider access model. A second factor is useful; an easily manipulated recovery process can still leave a rather conspicuous gap in the argument.
Treat supplier access as a defined service arrangement, not a favour that gradually acquires permanent permissions. The NCSC principles recognise policy-controlled access for guests and partner organisations. I would require an identifiable account, a named internal sponsor, a clear purpose and permissions appropriate to the work. A long relationship may provide useful context, but it does not establish what access is necessary today. Neither does a familiar logo explain who should authorise a particularly sensitive action.
Begin by listing the applications, administration interfaces and data each supplier needs. Separate ordinary support from exceptional maintenance and emergency work. Agree how access is requested, approved, reviewed and withdrawn, including when named supplier personnel change. Set time limits or review conditions appropriate to the task. Avoid copying broad permissions from a previous contractor merely because that makes onboarding faster. Check that suppliers can perform their legitimate work through the proposed route, with support and escalation arrangements that reflect the service’s actual operating hours.
I would include evidence access and incident responsibilities in the contract and operating procedure. Who can confirm which account performed an action? Who can restrict an active session? How quickly will each party provide relevant records? Use an authorised exercise to test an urgent withdrawal request and a legitimate emergency-maintenance request. Observe both successful control and operational friction. If the supplier depends on subcontractors or its own identity provider, make those dependencies visible rather than assuming that the customer’s access policy automatically controls the entire chain.
Review permissions against current work rather than the existence of an unexpired contract alone. Sample expired or withdrawn access to verify that enforcement matches the record. Report missing sponsors, unmanaged exceptions and unsupported session withdrawal honestly. Give supplier personnel clear instructions and a fair route to resolve mistakes. Effective control should make responsibilities easier to understand, not turn contractors into reluctant navigators of an internal maze. The intended outcome is narrower, explainable access that supports delivery and can be changed promptly when the business need or risk changes.
Machine identities represent software services, workloads or devices rather than individual people. The NCSC explicitly includes services and devices in its identity principles. Applications and automated processes need permission to perform tasks, and those permissions deserve scrutiny. The implementation will differ from a person signing in, but the governance questions remain recognisable: what is the identity for, what can it do and who is responsible? Automation should not provide a quiet exemption from access ownership simply because nobody has a staff photograph.
I would start with important services and record the identities they use to communicate, administer resources or retrieve information. Identify an accountable owner, business purpose, authorised scope and dependencies. Include identities created by integrations and supplier-managed services where they affect the service in scope. Distinguish what is known from what discovery has not yet established. An apparently complete register assembled from one directory may omit other environments, so state the coverage rather than allowing a tidy spreadsheet to imply certainty it has not earned.
Review whether each identity has permissions appropriate to its function and whether its credentials can be managed through a supported lifecycle. Ask how access is changed when an application moves, an integration ends or a credential is suspected of compromise. Do not impose human-style authentication prompts on a process that cannot answer them. Instead, require the technical team to explain the supported identity and authentication mechanism, how it is protected and how failures are detected. Test changes in an appropriate environment before affecting production dependencies.
Build retirement and incident handling into operations. Verify that unused integrations lose access and that owners understand withdrawal consequences. Investigate unexplained permissions and report important unknowns separately. Combine ownership measures with tested restriction or withdrawal for selected identities. A service account cannot attend the review meeting. That is a reason to identify its human owner, not to conclude that the review can be cancelled.
People, privacy and usable access
No. Policy can reassess available evidence without asking someone to complete a fresh interactive challenge for every ordinary action. The NCSC principles describe access decisions informed by multiple signals, including identity and device or service health. As the article explains, reassessment may maintain, restrict or revoke access. Which events trigger those responses, and how promptly they occur, depend on the implementation. Continuous verification is therefore a design objective to examine, not a promise of constant visible activity.
I would begin by identifying the actions that genuinely warrant stronger assurance. Reading routine information and changing a sensitive configuration need not be treated identically. Ask what signal justifies a challenge, what an additional prompt establishes and whether a less disruptive response would meet the need. Write the rationale in terms that a service owner and user can understand. Repeated prompts should not become a ritual that nobody can explain beyond saying that security requested them. Ritual has its place; access architecture benefits from a clearer purpose.
During the pilot, observe actual tasks rather than testing sign-in alone. Record prompt frequency, failed attempts, task completion and recovery experience across representative working patterns. Investigate repeated challenges and incorrect restrictions with the people affected. Check whether signal freshness, application integration or competing policies explain the problem. These are proposed investigation steps, not an assumption about a particular product. Include users with access needs and staff who move between locations or devices as part of legitimate work, because an unusual pattern is not automatically an inappropriate one.
Agree how support teams investigate refusals without granting unnecessary access. Review security tests alongside usability, retaining justified checks and adjusting avoidable friction. Explain material changes before wider deployment. The outcome should be proportionate intervention with workable recovery. If colleagues spend more time satisfying the access system than serving customers, investigate the design rather than congratulate the organisation on its exceptionally visible security.
Begin with the security question, not the maximum amount of information a monitoring tool can collect. The ICO’s monitoring-workers guidance warns that excessive monitoring can intrude into private life and undermine wellbeing. It is under review following legislative changes, so confirm the applicable requirements before implementation. My position is straightforward: protecting an organisation should include respecting its people. A technical ability to observe something does not, on its own, establish that observing it is necessary or appropriate.
For each proposed data source, record its purpose, the decisions it supports, who can access it and how long the information is needed. Ask whether a less intrusive source could answer the same question. Involve privacy specialists and, where appropriate, staff representatives before procurement or deployment fixes the design. Consider remote working and personal devices explicitly. Avoid assuming that a general security notice gives colleagues a meaningful understanding of detailed monitoring. Explain what happens in language that does not require legal or technical translation.
I would separate security investigation from unrelated management uses and require review before changing the purpose. Limit access to sensitive records and make the route for raising concerns visible. Where a data protection impact assessment is required, complete it before the processing begins, with specialist advice on scope and safeguards. Do not treat a completed assessment as permission to ignore later changes. Review supplier access, retention and incident arrangements too, particularly where monitoring information is shared or processed outside the organisation’s immediate control.
Check whether collected information actually contributes to useful decisions and remove unnecessary collection through a controlled review. Ask affected people about misunderstandings, disproportionate intrusion and practical difficulties. Investigate poor-quality signals before allowing them to drive consequential restrictions. Keep the evidence needed for legitimate security work, but resist the idea that every available observation must become a permanent record. Security logs should not quietly develop a second career as indiscriminate surveillance. A defensible programme can explain both why it collects information and why it deliberately chooses not to collect more.
Treat accessibility, ordinary usability and recovery as design requirements from the beginning. The article argues that user experience is part of control effectiveness, not a cosmetic addition after deployment. My recommendation is to define the work that must remain possible, then test whether people with different needs can complete it through appropriately protected routes. The objective is not to abandon assurance whenever someone encounters difficulty. It is to avoid designing a single idealised user and quietly excluding everybody who fails to resemble them.
Choose pilot participants for the range of services, tasks and working conditions they represent. Include different access needs, employment arrangements, shifts, devices and levels of connectivity. Ask people to complete realistic tasks rather than merely demonstrate that they can sign in once. Examine instructions, enrolment, routine authentication and the experience of a policy challenge. Provide a safe way to report difficulties without implying that the user is the problem. Someone who explains an awkward workflow is offering design evidence, not volunteering for remedial training.
I would test loss and recovery just as carefully as routine access. Establish alternatives for unavailable devices or authenticators, including support outside ordinary office hours where the service requires it. Decide what identity evidence support staff need, how urgent cases are escalated and who can approve temporary access. Keep exceptions limited and reviewable rather than allowing accessibility needs to become a permanent informal bypass. Work with affected users and relevant specialists to choose alternatives that meet the actual requirement rather than assuming one substitute works for everyone.
Measure task completion, incorrect refusals, restoration time and support demand, looking beyond overall averages. Pair those findings with access-control tests so convenience and protection remain visible together. Check that users understand the changes and know where to obtain help. Resolve recurring barriers before wider rollout. Requiring exceptional patience to obtain ordinary access is not a security principle; it is a service defect.
Operational resilience and managed services
Yes, but the approach must fit the operational environment rather than copy controls from office IT. CISA’s April 2026 guidance on Zero Trust for operational technology addresses legacy gaps, availability constraints and safety alongside identity, visibility and layered security. Operational technology, or OT, controls or monitors physical processes. That makes the consequences of a mistaken restriction especially important. My starting point is collaboration with engineers and operators, not a security deployment schedule that asks them to accommodate the architecture afterwards.
Identify the process, its operating limits and the access paths that matter. Understand how engineers, suppliers and automated components interact with it, including emergency maintenance. Use discovery and testing approaches approved for that environment. Ask what can safely be observed, changed or interrupted, and who has authority to make those decisions. Record dependencies between IT and OT without assuming that a familiar account or shared service is appropriate on both sides. The useful question is what supports safe, reliable operation under credible compromise and failure conditions.
I would begin with a bounded improvement, such as better governance of an understood remote-maintenance route, rather than indiscriminately introducing agents or automatic isolation. That is an illustrative starting option, not a universal prescription. Define permitted activity, monitoring, escalation and recovery with the process owner. Check how the arrangement behaves when identity services, connectivity or monitoring signals are unavailable. Test in an approved setting and sequence changes around operational requirements. A control that protects a dashboard while leaving operators unable to manage the process has missed the point rather comprehensively.
Assess security and operational evidence together, including access control, maintenance visibility, recovery and adverse effects on continuity or safety. Retain owned exceptions where technical limits remain. Use procurement to improve supported identity, logging and access capabilities as equipment changes. Adaptation is not inferior security. In OT, understanding physical consequences is part of technical competence, and protection must support the process rather than compete with it.
Ask how the service will operate when circumstances are inconvenient, not just how it appears during a demonstration. The article highlights accountability, evidence access, incident responsibilities, resilience and exit as central sourcing questions. My recommendation is to begin with the business service and the work you expect the provider to perform. Define which decisions remain yours, which activities are delegated and what information is needed to supervise them. A managed service should supply capacity or expertise, not make responsibility disappear behind a support portal.
Require a clear operating model with named internal decision owners and explicit provider delivery responsibilities. Who approves policy, enrols identities, investigates alerts, authorises restrictions and restores incorrectly denied access? What coverage exists outside normal office hours? How are urgent disagreements escalated? Ask which other services and subcontractors the arrangement depends on and what happens if those dependencies fail. Include privacy and procurement colleagues where data handling or contract terms need specialist review. Do not assume that the provider operating the platform also owns every task required to make the wider service effective. Write down the boundaries before a real incident discovers them.
I would test scenarios before commitment or wider rollout. Use an authorised exercise involving urgent access withdrawal, an unavailable identity service and an essential user incorrectly blocked. Ask the provider to produce the records needed to explain its decisions. Examine actual response and recovery steps rather than accepting a general statement that the platform is resilient. Agree how evidence will be obtained during an incident, including when several suppliers are involved. Record limitations, service tolerances and any functions that depend on an additional licence or separate customer action.
Examine exit as an operational task. What policy and activity information can be exported, and with what assistance? What knowledge must your team retain? Review outcomes, user experience and exceptions alongside availability. Keep an informed internal owner. Outsourcing can be sensible; outsourcing the ability to understand your own access risk is a less attractive bargain, whatever discount happens to accompany it.
The answer should be designed and rehearsed before the dependency fails. As the article stresses, resilience and emergency access belong in the assessment of the access model, not in an appendix discovered during an outage. I would identify which essential tasks must continue, which actions should stop and who can authorise an alternative route. Avoid a blanket assumption that every failure should either admit everyone or deny everything. Different services have different consequences, and a tidy universal rule can be operationally untidy rather quickly.
Map the dependencies involved in ordinary access: identity, policy, enforcement, connectivity, administration and support. Ask which components are needed to restore the others, including whether administrators could become locked out of the recovery process itself. Define service-specific tolerances and credible fallback arrangements with service owners. Include the people serving customers or managing essential operations, because they can explain the consequences of delay. Where physical safety is involved, use the relevant engineering and safety decision process rather than treating continuity as a technical preference to be balanced informally.
I would test a controlled failure scenario in an approved environment or exercise. Observe whether essential work remains possible, whether inappropriate access is prevented and whether responders can obtain sufficient evidence. Test communications as well as technical recovery. Emergency permissions should have named approval, appropriate scope, a way to record their use and a clear return to ordinary policy. Consider how those records will be captured if normal logging is affected. An emergency route that cannot be explained afterwards deserves scrutiny before it becomes the organisation’s favourite shortcut.
Review exercises against service requirements and resolve circular dependencies, unclear authority and unsupported applications. Retest after significant changes. Review emergency access and remove temporary permissions through the approved process. Assess continuity and control together: recovery is incomplete if it leaves an uncontrolled route open. Seek a recoverable service with accountable decisions, not a permanently unlocked side door labelled exceptional.
Measurement, assurance and future decisions
Measure whether the intended exposure changed, not how many products were deployed. I would combine access-risk measures with tested response, continuity and user experience. The following are proposed measures, not results already achieved or externally prescribed targets. Select a small set that fits the pilot, and define the service, identities and access paths covered. A green dashboard can be a useful summary, but it should be the conclusion of an evidence trail rather than a particularly cheerful substitute for one.
Establish a baseline before the change. Record persistent privileged permissions, supplier entitlements and inappropriate access paths found through authorised tests. Repeat the same tests afterwards and explain differences in scope. For access withdrawal, measure from an authorised instruction to verified denial, checking new sign-ins and existing sessions separately. Include unsuccessful withdrawals and unsupported applications. A lower count of administrator accounts is not enough if the remaining permissions still expose the service. Examine what access permits, who owns it and whether the entitlement remains justified.
For response, use safe, approved exercises to measure detection, investigation and verified containment as separate stages. Pair those results with legitimate task completion, incorrectly denied access, support demand and recovery experience. Test essential work during relevant dependency failures too. Define service-specific acceptance criteria with owners rather than inventing a universal percentage improvement. Review the slower and failed cases, not only averages. Record the baseline period, evidence sources, exclusions and any concurrent changes that could explain the result. These details determine what a comparison can reasonably support.
Present observed change, operational consequences and uncertainty together. A blocked test path does not establish a corresponding reduction in breach probability. Fewer recorded incidents may reflect changed detection or reporting. Invite independent challenge where practical, and use findings to adjust policy and expansion decisions. The objective is defensible improvement, with enough humility to notice when an attractive assumption has outlived the facts.
No. A Zero Trust architecture can support particular controls, but it is not a universal compliance finding. As the article explains, legal and regulatory obligations differ in scope and effect. NIST’s June 2025 implementation guide includes mappings to security standards and guidelines, which can help organise analysis. Those mappings do not determine the obligations of a particular organisation or prove that its implementation meets them. A useful reference is not the same thing as an assessor’s conclusion about your service.
Begin with the service, the information it handles and the jurisdictions or contractual requirements relevant to it. Ask legal, privacy and assurance specialists to identify the applicable obligations where interpretation matters. Then map each requirement to a control, accountable owner and evidence source. Distinguish obligations that an access control supports from those needing other processes or safeguards. Avoid copying a supplier’s framework mapping into the assurance pack without checking scope. Otherwise, a document about what a product could support can quietly become a claim about what the organisation has actually done.
I would test whether evidence demonstrates operation, not just configuration. An access-review policy shows an intention; completed reviews and verified removal of unnecessary access provide a different kind of support. Record coverage, review periods, exceptions and missing observations. Check whether records are sufficiently complete, protected and retrievable for the intended assurance purpose, with retention decisions informed by applicable requirements. These are proposed assurance steps, not legal advice about a particular regime. Where a control does not meet the requirement, document the gap and route it to someone empowered to act.
Review mappings when obligations, services, suppliers or controls change. Keep effectiveness, compliance and resilience distinct. A scoped assessment does not remove every risk, and an architectural improvement does not automatically satisfy every obligation. Ask what assessors examined and excluded. Seek traceable evidence and ownership. If the case depends mainly on a supplier saying “compliant”, the next task is to ask a better question.
A useful engineering reference is NIST SP 1800-35, published in June 2025. NIST describes 19 example implementations developed with 24 collaborators, including integration detail and lessons learned. These examples provide practical models for examining how components can work together. They are not 19 independent field trials establishing a universal reduction in breaches. That distinction is not criticism of the guide. It is a reminder to ask evidence to support the kind of conclusion it is actually equipped to support.
I would use an example to identify design choices, dependencies and questions for a local implementation. Compare the documented environment with your services, identities, devices and operating constraints. Examine the scenarios demonstrated and the limits of the testing. Ask whether required capabilities depend on integrations, configuration or assumptions absent from your environment. Document those differences before adopting the design. A successful demonstration is valuable evidence about the conditions demonstrated; it is not a transferable warranty covering every application, customer need or difficult supplier relationship you may encounter.
To establish local security outcomes, define a baseline and repeatable tests. Examples of proposed measures include fewer unintended access paths, less unnecessary standing privilege and faster verified access withdrawal. Pair those with legitimate task completion, support experience and continuity tests. Record the population covered, failed cases and missing observations. Compare like with like and explain concurrent changes that could affect the result. Where feasible, invite independent challenge. A claim that a tested access path is now blocked can be justified without pretending that breach probability fell by an identical proportion.
Financial return and reduced losses require further evidence and explicit assumptions. Do not turn implementation counts or quiet incident periods into confident avoided-loss figures. Separate observations, estimates and expectations. Use uncertainty to guide further testing. Borrow the engineering lessons, verify local behaviour and keep claims proportionate. A working installation is an achievement, but assurance must explain precisely what the evidence establishes and excludes.
Treat AI as a capability to evaluate, not an authority to obey. The article identifies possible uses in analysis while emphasising data quality, transparency, oversight and local testing. My recommendation is to start with a defined decision or investigative task, then ask whether the proposed capability improves it. Do not begin with a requirement to use AI and work backwards until an appropriate problem volunteers. Adding fashionable terminology to an access decision does not make the evidence beneath it more reliable.
Ask the supplier or internal team to explain the inputs, outputs and intended operating limits. What observation does the system make, what action can follow and who remains accountable? Distinguish a recommendation to an analyst from an automatic restriction affecting an essential service. Require an explanation of how incorrect alerts, missed events and uncertain inputs are handled. These are evaluation questions, not a claim that all AI systems share the same design. If the relevant behaviour cannot be described sufficiently for a safe deployment decision, narrow the proposed use.
I would begin with a bounded assessment using approved data and scenarios. Compare the proposed approach with the existing process, including work that legitimate users perform under unusual conditions. Examine whether analysts can understand and challenge an output, whether errors can be corrected and whether affected users can recover access. Set limits on automatic action according to the consequences of error. Involve privacy specialists where personal information is processed and service owners where continuity may be affected. Keep security testing separate from unsupported judgements about individual employees.
Define evidence requirements before expanding authority, and reassess after changes to data or use. Retain a practical way to suspend the capability where appropriate. Separate demonstrated benefits from expectations. Oversight requires information and authority to intervene, not merely a name on an approval record. The organisation must own the consequences. “The AI decided” describes delegation, not an acceptable end to accountability.
Post-quantum planning concerns cryptographic dependencies, not a replacement for identity governance or sensible access policy. NIST’s post-quantum cryptography guidance recommends identifying vulnerable algorithms and planning migration to its quantum-resistant standards. For a Zero Trust programme, the practical connection is the cryptography supporting identities, protected communications and other trust mechanisms. My advice is to make those dependencies visible in architecture and procurement decisions, rather than treating quantum resistance as an unrelated future initiative with no owner or service context.
Begin with important services and data whose sensitivity or lifetime makes migration planning significant. Ask technical teams and suppliers where cryptography is used, what can be inventoried and which dependencies remain unknown. Identify accountable owners and links to certificates, software, hardware and external services as relevant. Do not assume that a general statement about encryption answers the question about algorithms or changeability. An inventory should support decisions, not merely accumulate technical names that look impressive while leaving the organisation uncertain about what it can actually change.
I would ask suppliers for specific, reviewable migration information: supported standards, affected versions, dependencies, interoperability testing and expected customer actions. Treat a roadmap as a plan to examine, not evidence that migration has already occurred. Use an approved test environment to investigate compatibility, performance and recovery before changing important services. NIST’s migration project separates cryptographic discovery from interoperability work, a useful reminder that finding dependencies and proving a replacement works are different tasks. Connect both to ordinary change management and service continuity requirements.
Prioritise using current authoritative guidance and applicable obligations, not an invented quantum-computer arrival date. Review ownership, discovery coverage, tested options and supplier dependencies while addressing present access risks. Post-quantum capability cannot compensate for unjustified permissions. Seek an adaptable migration path for real services, not a speculative countdown or permission to postpone less glamorous identity work until the future becomes convenient. Keep migration decisions connected to service ownership and documented change approvals throughout.
Put the FAQ guidance into action
Choose one service and the questions most relevant to its next access decision. Bring the service owner, delivery team and affected users together. Agree the exposure to address, the required work to preserve and the accountable decision owner and the team responsible for the next action. Record unknowns openly. If you cannot explain the problem without naming a product, spend more time defining the problem.
Establish a baseline, plan a bounded change and test both protection and usability through safe, authorised scenarios. Include support, recovery, supplier responsibilities and continuity before widening deployment. Set acceptance criteria appropriate to the service and record exceptions with authorised accountable approvers, action owners and review conditions. Do not wait for a perfect estate, but do not disguise known gaps as completed work.
At the review, compare evidence with the original objective and decide whether to adapt, broaden or stop. Separate observed results from expectations, and assign follow-up actions with agreed review points. Use the myth-busting section to challenge any assumptions still carrying more weight than the facts. The immediate deliverable should be a better-informed decision and an owned action, not merely a meeting that agreed Zero Trust was important.
