Debunking Common Myths about Zero Trust
Use this page to challenge claims before they become policy, procurement commitments or comforting board assurances. The topics group related assumptions about purpose, migration, identity, resilience and evidence. They are not ranked by prevalence, and the statements are not attributed to particular suppliers or organisations. Read the myth that resembles the decision in front of you, then use the practical checks to discover whether the underlying assumption holds in your circumstances. The purpose is better judgement, not winning an argument by attaching a fashionable label to the opposing view.
In a workshop or proposal review, ask participants to describe the claim in plain language, identify the service and name the evidence that would support it. Separate what a product can do from what has been configured, tested and operated successfully. Record exceptions, dependencies and the consequences for people using the service. If a claim is only partly wrong, preserve the useful part and correct the overstatement. A VPN can still perform necessary work; an AI capability may offer value; a supplier example can teach useful engineering. None deserves either automatic endorsement or automatic dismissal.
Pair each challenge with an action: request a scoped demonstration, run an authorised test, name the accountable approver and responsible delivery team or revise the benefit claim. Use the FAQ section for fuller guidance on the corresponding delivery decision. Recommendations here are reasoned positions to test against local evidence, not substitutes for specialist assurance. Where the facts change, revise the conclusion. That is a more useful habit than defending yesterday’s certainty, and generally leaves fewer people having to explain an avoidable surprise later.
Purpose, scope and architectural choices
The reality is that Zero Trust removes automatic technical trust, not human dignity. The NCSC explicitly distinguishes Zero Trust from having no trust. A network connection or previous successful sign-in should not automatically justify every subsequent permission. That is different from assuming colleagues are dishonest. My view is that the distinction should shape both the policy and the way leaders explain it. An organisation cannot credibly claim to protect its people while communicating as though those people are the principal inconvenience.
Challenge the myth by asking what evidence an access decision actually needs. A service may require confirmation of identity, an appropriate permission and sufficient confidence in a device. None of those requirements establishes that the user is untrustworthy as a person. Explain the risk in terms of compromised accounts, mistakes and changing work, without turning ordinary staff into a threat category. Apply proportionate controls to senior leaders and technical administrators too. A policy that treats status as a permanent exception rather weakens the argument that it exists for everyone’s protection.
I would involve users in testing the design and make challenge and recovery routes clear. Ask whether the control permits legitimate work, whether refusals are understandable and whether support can resolve mistakes without unnecessary intrusion. Separate security monitoring from unrelated surveillance and involve privacy specialists where appropriate. These are practical recommendations, not a claim that a particular implementation is fair merely because it uses the Zero Trust label. Fair treatment needs to be visible in how decisions are made, explained, corrected and reviewed.
Examine access decisions and exceptions. Can the organisation explain their purpose, evidence and scope without personal suspicion? Do people know where to obtain help, and does feedback improve policy? Report technical uncertainty honestly. Seek justified access with respectful treatment. Distrusting everybody is not sophisticated security; it is poor leadership equipped with a policy engine and an unnecessarily difficult working relationship.
The reality is that products can implement controls, but cannot purchase accountability on your behalf. The NCSC rejects the single-product interpretation of Zero Trust and describes a strategic commitment involving multiple controls and designs. A platform can provide useful functions while leaving permissions, ownership, support or recovery unresolved. That does not make the product worthless. It means its contribution needs to be understood at the appropriate scale, without allowing a sales category to stand in for the architecture of the whole organisation.
Challenge a product-led proposal by starting with the service and risk. Ask what specific access decision the product changes, which identities and resources it covers and what remains outside scope. Require the supplier to describe required integrations, configuration and customer responsibilities. Compare those requirements with the capabilities and staffing available. I would also ask what existing controls must remain and why. If the answer is that the platform does everything, ask what “everything” excludes. The resulting pause can be more informative than the preceding collection of animated slides.
Turn claimed capabilities into acceptance tests. Demonstrate a permitted task, an inappropriate request, a missing signal, an access-withdrawal instruction and a dependency failure where relevant. Examine evidence and support arrangements, not just the visible success screen. Decide who will maintain policies, investigate alerts and resolve incorrectly blocked users after the implementation team leaves. Include ongoing cost and knowledge requirements. A feature that exists but cannot be operated reliably in the customer’s environment is not yet an effective control, however comprehensive its description in the proposal.
Assess products as components of an owned operating model. Record the risk addressed, evidence, limitations and remaining responsibilities, and review these as services change. Approvals should explain the expected contribution, not announce that Zero Trust has been acquired. Seek defensible access and resilience. A licence authorises software use; it does not settle accountability when that software’s decisions affect real people.
The reality is that ZTNA addresses application access within a wider Zero Trust approach. Section 3 explains the distinction, while the NCSC principles cover identities, health, policies and monitoring across the architecture. An application-access service may be useful without resolving every permission, data or recovery issue. The mistake is not buying ZTNA. It is assuming that the purchase has completed responsibilities that were never within its scope, leaving the remaining work invisible until an incident makes it rather conspicuous.
Challenge the claim by asking the proposer to name the resources and identities covered. What request does the service assess? Which signals inform the decision, and where is it enforced? What is outside the protected path? I would require a simple scope statement and dependency map before approving the design. Avoid treating a platform diagram as evidence that all applications and access routes are integrated. Mark unsupported services and retained arrangements explicitly, with owners and a plan appropriate to the remaining exposure.
Turn the proposed benefit into an authorised test. Demonstrate that a legitimate user can perform the required task and cannot obtain unnecessary application access through the assessed route. Test signal loss, support and emergency arrangements where relevant. Ask what happens after an account is withdrawn, including any existing session. These checks evaluate a specific implementation; they should not be replaced by assumptions about what the ZTNA category usually promises. Record failed cases and limits alongside successful demonstrations so that the review reflects the actual service.
Examine the wider operating model: permission review, endpoint protection, investigation and recovery. Agree how responsibilities connect and expose unresolved gaps. Value ZTNA for the problem it demonstrably addresses while retaining broader governance. A component does not become less useful when its limits are acknowledged. It becomes easier to supervise without asking a product label to carry responsibilities it cannot fulfil. Make the remaining work visible in the same service review and funding decision.
The reality is that Zero Trust can address on-premises and cloud resources. NIST SP 1800-35 explicitly covers distributed resources across on-premises and multiple-cloud environments. The approach concerns justified access to resources, not a compulsory hosting destination. Moving an application may change available options, but it does not settle who should use it, what they should be allowed to do or how access is recovered. A change of landlord is not, by itself, an access-control strategy.
Challenge the myth by separating the hosting decision from the access requirement. Identify the service, its users, data, administration paths and operational constraints. Ask what assurance is needed and where policy can be enforced in the current environment. Then compare options, including any benefits or limitations associated with a proposed move. I would avoid assuming either that cloud services are inherently suitable or that on-premises systems are inherently unsuitable. Each claim needs a specific architecture and evidence rather than a broad preference about geography.
For a mixed estate, map identities, signal sources and enforcement coverage across environments. Test a representative permitted task and an inappropriate request on each relevant path. Check whether account changes and access withdrawal behave consistently enough for the intended service, while recording legitimate differences. Include the support and recovery dependencies created by external services. These are proposed assessment steps, not a claim that one central mechanism must control every resource. The goal is a coherent, explainable policy model, not uniformity pursued regardless of practical constraints.
Decide using service requirements, risk, cost and capability. Record unsupported components and agree appropriate interim controls. Verify reduced exposure without unacceptable disruption. A cloud move may be justified, but its reasons still need explaining. Retaining on-premises services should not excuse inherited trust either. Judge the access decision wherever the resource operates, keeping hosting choices accountable to actual organisational needs. Keep that reasoning available when hosting or supplier arrangements are next reviewed.
The reality is that network location alone should not establish access. The NCSC principles include not trusting any network, including the organisation’s own. That applies to office connections as well as remote ones, and to services and devices as well as people. Physical presence may provide context, but it should not become an unlimited permission. My advice is to examine the access decision wherever it occurs rather than assume the office boundary has already answered every relevant question on behalf of the service owner.
Challenge the claim by comparing what an identity can reach from different environments. Include ordinary users, administrators, suppliers and relevant automated processes. Ask which differences are justified by policy and which reflect inherited network arrangements. Identify the resource and action involved rather than treating all internal connectivity as one category. These are proposed assessment steps, not a claim that every office network is flat or every remote arrangement is stronger. A useful comparison needs the actual architecture and its controls, including retained protections that still serve a purpose.
I would test representative permitted and prohibited requests through approved scenarios. Check whether identity, health and resource requirements are enforced as intended in each environment. Include support and emergency access so that stronger checks do not create an unexamined route around the policy. Record coverage gaps and exceptions, and involve service owners when differing arrangements affect continuity. Avoid pursuing identical implementation simply for tidiness. The aim is a coherent basis for access decisions, with differences explained by service needs and evidence rather than familiarity with the building or connection type.
Prioritise meaningful exposure across office and remote paths, including administration and machine access where relevant. Explain changes and keep controls proportionate. Resource-centred access is not location blindness: location may be evidence, but not the entire argument. The kettle can remain familiar. Sensitive systems deserve an explanation that survives a compromised account or device appearing to come from the office.
Migration, delivery and supplier dependency
The reality is that a VPN and a Zero Trust approach are not mutually exclusive. The NCSC addresses this misconception directly. The relevant question is what access a connection permits and how that access is governed, not whether a familiar technology has become unfashionable. Some access paths may warrant replacement; others may remain necessary during transition. I would assess the service and controls before deciding. Switching something off is straightforward. Demonstrating that the replacement preserves protection and continuity takes more useful work.
Start by identifying the functions the existing VPN arrangement provides and the applications that depend on it. Record authentication, access scope, monitoring and support responsibilities without assuming that every deployment behaves alike. Ask which functions the proposed replacement covers and which need another control. The NCSC implementation collection includes guidance on deciding when a VPN can be removed. Use that discussion to frame local assessment, not as permission to skip the awkward inventory because the intended destination appears clear.
I would test representative workflows and failure conditions before retirement. Include administrators, suppliers, users with different devices and any application with unusual dependencies. Verify both permitted activity and requests that should be denied. Check recovery, logging and emergency support, including outside normal hours where necessary. Document what remains on the old route and why. During coexistence, examine whether the retained path bypasses restrictions introduced elsewhere. Temporary arrangements deserve explicit ownership and review conditions, otherwise they have an impressive habit of becoming the permanent architecture.
Remove the route when required functions are covered and the responsible owner accepts the evidence. Update support procedures and retain an appropriate recovery plan. Report exceptions honestly. The correction is neither “keep every VPN” nor “remove them all”, but a defensible service decision. A disconnected user is not automatically protected, and an outage should not be mistaken for successful access-risk reduction. Record the retirement decision, its evidence and the owner responsible for remaining exceptions.
The reality is that understanding the existing estate is not an instruction to demolish it. The NCSC principlesexplicitly recognise legacy integration constraints, and its implementation guidance includes mixed-estate approaches. Some components may need replacement, but that conclusion should follow an assessment of the limitation and available controls. I would be wary of a programme that postpones every useful improvement until a complete rebuild. Perfect starting conditions are appealing, but they are not a dependable delivery strategy.
Identify the particular service and access problem. Establish which users and processes need access, what the system can support and what protections already exist. Separate an unsupported authentication method from a lack of monitoring or a risky administration path. Those limitations may require different interventions. Ask whether a controlled access route, isolation or another compensating measure can reduce exposure while the service continues. Record the remaining risk, because an overlay should not be allowed to turn a known limitation into a conveniently forgotten one.
I would prioritise a bounded improvement with a clear owner, acceptance tests and recovery arrangements. Check necessary workflows and prohibited access before enforcement changes. Where safety or essential operations are involved, involve the relevant specialists and use approved test methods. Keep interim exceptions visible and specify the conditions that should trigger reconsideration. Do not label every old component unacceptable simply because replacement is technically possible. Equally, do not treat operational familiarity as evidence that the current exposure is reasonable. Both positions avoid the actual assessment.
Agree the longer-term decision: retain with managed constraints, upgrade, replace or retire. Include funding, ownership and review as support changes. Seek phased improvement without pretending every limitation has disappeared. An ageing application may be inconvenient; an untested replacement can be inconvenient at greater expense. Progress means better-supported access, not merely a newer inventory and an impressive announcement. Revisit the decision when service requirements, supplier support or technical constraints materially change.
The reality is that delivery projects can finish while access governance continues. The NCSC explicitly says that organisations do not complete Zero Trust. Services, identities and requirements change, so policies need maintenance and review. That does not mean a programme should operate without scope, milestones or financial discipline. My preference is bounded delivery followed by explicit operational ownership. Continuous improvement must not become the phrase used whenever somebody asks whether a project has delivered what was agreed or who is paying for the next phase.
Challenge the myth by separating project acceptance from ongoing control responsibilities. Define the services and capabilities the phase will deliver, the evidence needed for acceptance and the limitations that remain. Name the accountable service owner, then identify who maintains policy, reviews permissions and handles incidents afterwards. Record unresolved exceptions with accountable owners rather than moving them into a generic future-work list. A handover is not complete merely because the implementation team has stopped attending meetings. The operating team needs knowledge, authority, documentation and realistic capacity to sustain the arrangement.
I would establish review triggers linked to meaningful change, alongside an appropriate routine schedule. A new supplier, changed application dependency or altered service purpose may justify reassessment. Test important access and recovery paths after relevant changes, and use support feedback to identify recurring friction. Track whether temporary permissions expire and whether monitoring still produces useful evidence. These are practical recommendations, not a requirement for endless redesign. Stable controls can remain appropriate, but their continued suitability should be considered rather than assumed simply because nobody has opened a change request.
Report what was accepted, what improved, what remains outside scope and who owns operations. Justify further investment separately. Seek bounded delivery followed by sustainable ownership. Leaders should expect accountable projects and maintained controls. The ribbon-cutting is optional; policy ownership is not. Closing a project must not leave carefully introduced protections without anyone responsible for keeping them useful.
The reality is that dependency must be understood and managed, not renamed. The article notes that multiple suppliers can introduce integration and accountability challenges, and that an open-source implementation is not the same thing as a standard. Choice can be valuable, but a long supplier list does not establish practical independence. I would ask what the organisation can actually change, export, operate and recover. Those capabilities matter more than the reassurance offered by a sourcing label or an impressive number of contractual relationships.
Challenge the myth by mapping the service’s technical and operational dependencies. Identify identity, policy, enforcement, monitoring and support responsibilities, including where suppliers depend on one another. Name the accountable incident decision owner and the teams responsible for actions across those boundaries. For an open-source option, examine the skills, maintenance and support arrangements needed to operate it, rather than assuming access to code answers every practical question. For a proprietary option, examine documented interfaces and portability rather than assuming dependence is necessarily unmanageable. Compare specific arrangements and consequences, not competing articles of procurement faith.
I would test export and exit before relying on promises about them. Can policy and relevant evidence be obtained in a usable form? What must be rebuilt or translated to move elsewhere? What knowledge, time and assistance would be required? Include a realistic scenario involving an unavailable provider or an urgent need for incident records. Document costs and limitations, with responsibilities agreed in contracts and operating procedures. An export button may be useful, but it does not prove that another service can interpret the result or that your team can complete the transition.
Decide where integration justifies concentration and where alternatives matter. Retain internal understanding and review dependencies as services evolve. Seek workable exit and accountability, not automatic preference for a supplier count or licence model. Several contracts do not make a coherent service. Independence matters when choices can be exercised without discovering that essential knowledge, records or permissions were never retained.
Identity, authentication and least privilege
The reality is that authentication is one part of identity and access security. The NCSC principles also address identity knowledge, health signals, authorisation and monitoring. MFA can strengthen evidence about a sign-in, but does not by itself establish that every permission remains appropriate. As the article explains, roles change and projects end. Proving that someone is the right person is useful. It does not prove that they still need the access accumulated during every previous assignment.
Challenge the myth by reviewing a small, meaningful set of accounts. Identify the owner, current purpose, permissions and recovery arrangements. Include sensitive functions, administrators and supplier access, while distinguishing machine identities that need a different authentication approach. Check whether joiners receive justified access, movers lose permissions they no longer need and leavers are removed through an owned process. These are proposed governance checks, not an assumption about any organisation’s current state. Record unknowns and exceptions rather than treating MFA coverage as a substitute for understanding them.
I would test both permission and withdrawal. Can an authenticated user reach only the resources appropriate to the approved task? When access is withdrawn, are new sign-ins denied and existing sessions handled as required? Examine help-desk recovery and exceptional access too. A policy can appear strong while a routine reset process bypasses its intended protection. Use safe, authorised scenarios and focus findings on process and control improvement. The point is not to catch staff out, but to understand whether the organisation’s stated access rules survive ordinary operational demands.
Measure authentication, entitlement quality and withdrawal separately, alongside support and continuity. Assign owners to unresolved permissions. Treat MFA as part of an identity lifecycle, not its finishing line. Organisational memory is valuable; permanent accumulated privilege is less so. Explain what was authenticated, what was authorised and how authorisation ends, rather than stopping at the presence of a second factor. Review the whole lifecycle when service requirements or account recovery arrangements change.
The reality is that MFA methods resist different attacks. CISA’s foundational 2022 guidance, now archived, distinguishes phishing-resistant authentication from number matching, which it recommends as an interim mitigation for push-notification fatigue. The article makes the same distinction. A single MFA-enabled total can therefore conceal materially different protections. I would ask which methods are used for which services, rather than assume that two prompts, two screens or two pieces of information necessarily provide the same assurance.
Begin by mapping supported methods across important accounts and applications. Identify where stronger options are available, where compatibility limits the choice and where an exception needs review. Consider enrolment, recovery, accessibility and device availability alongside technical properties. Do not make a policy that works only for people with a particular personal phone or working pattern. Where an interim method is necessary, explain the reason and intended review. Prioritisation should direct improvement, not create permanent categories of people whose protection is assumed to be somebody else’s problem.
I would require suppliers to explain the attacks a method is intended to resist and the conditions supporting that claim. Test the complete operating process using approved scenarios, including lost authenticators and urgent account recovery. Check that support staff can verify a request without casually defeating the stronger method. Record implementation limitations separately from user guidance. Training is useful, but it should not be asked to transform a method into something its design does not provide. The distinction is technical and operational, not a judgement about users’ intelligence or effort.
Report coverage by method and service, together with enrolment, recovery and unresolved access needs. Do not apply a general vendor statistic to overall cyber risk. Seek stronger authentication operated appropriately within the access model. A tick in the MFA column can start a useful conversation. It should not conclude that conversation before anyone asks what the tick represents. Keep method limitations visible when discussing coverage and approving temporary exceptions.
The reality is that reassessment need not mean repeated interactive authentication. The NCSC principlesdescribe decisions using several signals, including identity and device or service health. As the article explains, a policy may maintain, restrict or revoke access as circumstances change. The available responses and their timeliness depend on the implementation. Constant visible interruption is not evidence that the underlying decision is particularly intelligent, and a quiet session is not proof that no assessment is taking place.
Challenge the claim by asking what each prompt establishes. Which action is being protected, what evidence changed and why is an interactive check appropriate? Distinguish an ordinary task from a sensitive administration action. Ask whether missing information should trigger a challenge, restricted access or another supported route. These are policy choices to examine, not instructions to choose the same response everywhere. I would expect the rationale to be understandable to service owners and support staff, not confined to whoever originally configured the platform.
Test representative workflows during a pilot. Record prompt frequency, legitimate task completion, failed access and recovery experience. Include users who change location, work shifts or use different approved devices. Investigate repeated challenges and incorrect restrictions alongside technical logs, with appropriate privacy controls. Do not interpret user frustration as proof that stronger protection is succeeding. Equally, do not remove every challenge merely to improve satisfaction. Assess whether the intervention contributes justified assurance for the action and whether its operational cost is proportionate to that purpose.
Adjust policies and support using the findings, then verify changed behaviour. Keep control tests alongside usability measures. Seek proportionate intervention with clear recovery. Endless prompts can teach people that their job is to get past prompts. A better system helps them understand when additional assurance matters while allowing ordinary work to continue without unnecessary ceremony. Agree who reviews recurring challenges and when unresolved usability problems must be escalated.
The reality is that a device posture result covers specified conditions, not every possible compromise. The article explains that assurance depends on the quality and freshness of the signals. The NCSC treats device and service health as evidence for access decisions, alongside other information. Passing a policy check can be useful without proving a device is uncompromised. I would resist replacing uncertainty with a reassuring label simply because the management console offers a convenient shade of green.
Ask what the check actually observes. Which devices are covered, which conditions are assessed and how recently was the evidence obtained? What happens if information is missing or a device has not reported? Separate a configured policy from demonstrated enforcement on the services in scope. Record unsupported operating systems, exceptions and other coverage gaps. These questions should be answered through the implementation’s documentation and authorised tests, not assumptions about what a product category normally includes. The result should describe the confidence available, not imply a certainty nobody has established.
I would test a representative set of conditions in an approved environment: a device meeting policy, one deliberately failing a supported condition and one with unavailable or stale information. Check permitted work and refusal behaviour, including the support route for a legitimate user. Do not introduce unsafe configurations into production merely to create a demonstration. Agree how a changed health signal affects an existing session and verify the behaviour the implementation actually supports. Record the time and limits of the observation so that results remain interpretable later.
Use posture within least privilege, monitoring and response. Review policies as devices and services change, and investigate inconsistent results. Treat posture as a bounded observation. A passing result may justify access under policy; it is not a lifetime certificate of good behaviour. Explain what the evidence covers, when it becomes stale and what happens when confidence is insufficient.
The reality is that least privilege should enable required work without attaching unrelated permissions.Section 2 describes access appropriate to the task, duration and changing role. My recommendation is to assess both unnecessary exposure and the ability to obtain justified permission. A control that repeatedly prevents legitimate delivery is not vindicated merely because it grants less access. Equally, convenience alone does not justify permanent broad permissions. The useful design question is how to support the work while keeping the scope explainable and reviewable.
Begin with real tasks and the resources needed to complete them. Involve users and service owners rather than defining roles solely from existing account templates. Separate ordinary work from occasional sensitive actions, then decide how additional access should be requested and approved. Record why the permission is necessary and when it should expire or be reviewed. I would investigate repeated exceptions as possible evidence that the standard role is wrong. Copying the broadest existing permission set may be quick, but it is not a particularly searching analysis of need.
Test a representative workflow before enforcing narrower permissions. Include unusual but legitimate work, urgent support and recovery from an incorrect refusal. Set response expectations for access requests according to service needs, with an escalation route when delays matter. Avoid creating an informal network of helpful administrators who routinely bypass the official process. That can conceal design problems and make accountability difficult. Use the feedback to improve roles and approval routes, while retaining controls over exceptional access and ensuring that temporary permission does not quietly become a permanent entitlement.
Measure unjustified privilege and legitimate task completion together. Check permission removal after role changes and temporary work, sampling actual access. Seek appropriate access, not the largest number of refusals. If ordinary permissions require exceptional persistence, improve the operating model. Security should reduce exposure without reducing service capacity. A narrow doorway is useful only if the right people can get through it.
The reality is that software and device identities also need ownership and appropriate permissions. The NCSC identity principle covers users, services and devices. Machine identities can support important interactions without a person signing in at each step. The controls will differ, but purpose, scope and withdrawal still matter. My concern is not that automation is undesirable. It is that an identity nobody remembers creating can become an access dependency nobody feels able to change, which is an awkward basis for assurance.
Challenge the myth by selecting an important application or integration and tracing its identities. Who owns each one? What work does it perform, what resources can it reach and what would happen if access were withdrawn? Distinguish supported automation from an account used informally because it was convenient. State the discovery scope and unknowns rather than imply that one directory reveals every service identity. I would prioritise unexplained access to sensitive resources and unresolved ownership before attempting a perfectly comprehensive inventory that postpones every practical improvement.
Ask technical teams to explain the supported identity and credential lifecycle. How are changes approved, credentials replaced and obsolete integrations retired? Do not require a machine to answer a human authentication prompt, but do require an appropriate mechanism and an owned process. Test important changes in an approved environment, accounting for dependencies and recovery. Include the incident scenario in which access must be restricted quickly. Record what can be done safely, who can authorise it and where the implementation cannot support the desired response without additional work.
Review permissions against current purpose and verify retired integrations lose access. Investigate activity in context and keep exceptions visible. Apply appropriate controls, not identical human and machine workflows. A service account cannot attend the meeting; its human owner matters more, not less. Automation should reduce manual work without removing the ability to explain why a powerful permission still exists.
Detection, privacy and operational resilience
The reality is that restricting connections and detecting misuse answer different questions. Section 2 presents segmentation, monitoring and response as complementary controls, and the NCSC principles retain monitoring alongside policy-based access. A boundary may limit a path without explaining everything happening through an authorised connection. My advice is to examine both what can communicate and whether teams can recognise and act on inappropriate activity. Smaller compartments can be useful, but they do not remove the need to understand what is happening inside them.
Begin by mapping the critical service and its necessary dependencies. Identify the boundaries intended to constrain access, the policy behind them and the evidence available when a request succeeds or fails. Ask whether monitoring covers the relevant identities, resources and actions, rather than assuming that an enforcement log contains every fact an investigator needs. Record unknowns and service exceptions. Do not measure segmentation simply by counting zones or rules. A complicated boundary map can still leave the important access path poorly understood or unnecessarily broad.
I would use safe, authorised scenarios to test both restriction and response. Check that an inappropriate connection is refused, that legitimate work continues and that relevant activity produces usable evidence. Exercise investigation of suspicious behaviour within a permitted path where appropriate. Who receives the signal, how is context obtained and who can authorise containment? Include a false alarm so that recovery and proportionality are examined too. A prompt response is not automatically a good response if it unnecessarily interrupts an essential service or leaves operators unable to understand the decision.
Review evidence across teams and suppliers. Resolve gaps in context, responsibility and recovery, and retest after relevant changes. Boundaries, visibility and response belong in one protection model. Segmentation constrains paths, monitoring helps explain activity and response enables action. None performs the others’ work through optimism alone. Smaller rooms are useful, but someone must notice the smoke and know what happens next.
The reality is that Zero Trust aims to reduce risk, not provide immunity. The NCSC describes reducing unauthorised access, lateral movement and data compromise, rather than guaranteeing that incidents disappear. The article’s assume-breach principle explicitly accepts that prevention can fail. My objection to an absolute ransomware claim is practical: it can distract leaders from the containment, continuity and recovery work needed when controls do not behave as hoped. Confidence should come from tested capability, not from declaring failure theoretically impossible.
Challenge the claim by asking which attack scenario is covered and what happens when one control fails. Identify the service, identities and access paths involved. Examine whether an authorised account could still perform damaging actions, what permissions constrain it and which observations might reveal misuse. These are questions for an approved assessment, not instructions to introduce harmful activity into production. Ask suppliers to state the limits of their protection and dependencies. A credible answer should identify assumptions and residual exposure rather than treating qualification as an unfortunate weakness in the presentation.
I would exercise detection, containment and recovery with the service owner and operational teams. Define safe scenarios, success criteria and authority to stop the exercise. Measure whether inappropriate paths are blocked and whether responders can act without causing disproportionate disruption. Include essential work continuing or being restored through the agreed recovery arrangements. Review evidence access across suppliers and the ability to make decisions when an important dependency is unavailable. Do not treat the presence of a response feature as proof that people, procedures and integrations are ready to use it effectively.
Report demonstrated results and untested areas. Assign owners to gaps and avoid converting blocked paths into a universal ransomware-risk reduction. Seek layered resilience with acknowledged limits. Making compromise harder and constraining consequences is worthwhile without embellishment. A guarantee that nothing will go wrong is not stronger assurance; it is a reason to examine the assumptions more carefully.
The reality is that useful monitoring needs a defined purpose, relevant evidence and appropriate limits. The ICO warns that excessive worker monitoring can undermine privacy and wellbeing. Its guidance is under review, so check current requirements with appropriate specialists before implementation. I would not accept “we might need it” as sufficient justification for collecting every available observation. A large collection of information is not automatically a better security capability, particularly if nobody can explain the decision it is supposed to support.
Challenge the myth by asking what question each source answers. Identify the risk, the intended use, the people affected, access restrictions and the reason for retaining the information. Consider whether a narrower source could meet the need. Include privacy specialists and, where appropriate, staff representatives before design choices become difficult to change. Explain monitoring in terms people can understand. A broad notice that security monitoring occurs is not necessarily a meaningful explanation of an intrusive practice. The organisation should be able to describe what it does without hoping nobody asks for clarification.
I would test whether collected signals actually help authorised teams recognise, investigate or respond to relevant events. Check data quality, context and missing coverage. Do not interpret an unusual pattern as proof of misconduct, and avoid turning security information into unrelated assessments of people. Establish review before changing purposes or sharing records more widely. Where a data protection impact assessment is required, complete it before the activity begins. Specialist advice is important because the appropriate safeguards and legal basis depend on the circumstances, not on the enthusiasm of the monitoring vendor.
Review unnecessary collection and retention through a controlled process. Assess user concerns alongside technical usefulness. Seek proportionate visibility with accountable use, not indiscriminate surveillance. Protect colleagues and customers too. A defensible programme explains what it observes, who may act and where it draws a line. Collecting everything merely creates an expanding responsibility; reasoned restraint can be competent security.
The reality is that controls must fit the physical process and its operating constraints. CISA’s April 2026 OT guidance addresses legacy limitations, availability and safety alongside identity, visibility and layered security. An office-IT control may offer a useful principle without being suitable for direct deployment in operational technology. I would require engineering and operational involvement before changing access or introducing interruption. A design that ignores the process it protects is not strengthened by being implemented consistently or described in particularly authoritative language.
Challenge the claim by asking what the equipment does and what happens if access is delayed, restricted or lost. Map engineering, supplier and automated interactions, including emergency maintenance. Identify the operating limits and approved methods for discovery and testing. Do not assume that an agent, scan or automatic isolation action appropriate elsewhere is acceptable here. Record the dependency on external services and the authority required for changes. These are questions for the people responsible for the process, not details to be inferred from a network diagram alone.
I would begin with a bounded improvement and a jointly agreed test plan. An understood maintenance route may provide a useful starting point, but the choice depends on local risk and feasibility. Define permitted work, monitoring, escalation and recovery together. Check how the arrangement behaves when identity, connectivity or signals are unavailable. Use an approved setting and change window rather than experimenting with a live critical process for convenience. Include the possibility of an incorrect refusal, because safe recovery from control error is part of the design, not an embarrassing exception to it.
Assess security alongside continuity and safety. Retain owned constraints and use procurement to improve supported capabilities. Adapt implementation without abandoning justified access. Engineers and operators understand consequences; they are not obstacles to purity. An industrial process can be unimpressed by a presentation. Competent protection means better control without unacceptable risks to the physical work people depend on it to perform.
Assurance, automation and evidence
The reality is that compliance depends on applicable obligations and evidence, not an architectural label. The article distinguishes the scope and effect of different legal and industry requirements. NIST’s implementation guide includes mappings to security standards and guidelines, useful for structuring an assessment but not determining an organisation’s legal position. I would treat any claim of automatic compliance as an invitation to ask which requirement, which service and which evidence. Without those details, the claim is too broad to provide dependable assurance.
Begin with the service and information in scope. Ask the relevant legal, privacy and assurance specialists to identify obligations where interpretation is needed. Map requirements to controls, owners and records, distinguishing what the access architecture supports from responsibilities elsewhere. Examine exclusions and organisational boundaries before reusing a supplier’s mapping. A product may offer a useful reporting function without covering the entire process required by an obligation. These are proposed assessment steps, not legal advice or a conclusion about any particular regime’s application to the reader.
I would seek evidence of operation as well as design. An access-review procedure describes intended behaviour; completed reviews and verified removal of obsolete permissions provide different support. Test whether records are complete enough for the assurance question and whether exceptions receive appropriate approval and follow-up. Confirm how evidence can be retrieved during an incident or assessment, including dependencies on suppliers. Avoid treating a polished report as proof that the underlying data is accurate. Formatting can make a weak assurance argument easier to read without making it any more convincing.
Review assessments when obligations or services change. Record residual risk and ask assessors what they tested and excluded. Seek a traceable chain from requirement to control to evidence. A badge should summarise a bounded conclusion, not silence questions. “Our provider says so” is no substitute for understanding the obligation and demonstrating that the organisation meets it.
The reality is that automation does not remove responsibility for the policy or its consequences. Section 6 calls for data quality, transparency, oversight and testing when considering AI-assisted capabilities. My recommendation is to define what a tool may decide, why that authority is appropriate and how mistakes are corrected. A system can act automatically while a named role within the deploying organisation remains accountable for approving its authority and overseeing outcomes. “The AI decided” may describe the mechanism, but it does not explain who approved the arrangement or why the affected service should accept the result.
Challenge the myth by distinguishing advice from action. Does the capability highlight information for an analyst, recommend a response or restrict access without further approval? Identify the inputs, operating limits and consequences of error. Ask what explanation is available to investigators and how uncertainty is represented. Do not assume that a human nominally reviewing an output provides meaningful oversight. I would require sufficient information, time and authority for that person to challenge the recommendation, together with an escalation route when the system behaves outside the intended use.
Test a bounded use case with approved scenarios and appropriate privacy safeguards. Compare results with the existing process and examine incorrect alerts, missed events and legitimate unusual activity. Check how affected users recover access and how teams suspend or reverse an inappropriate automated action. Include service owners where continuity matters and relevant specialists where safety is involved. These are proposed governance checks, not a verdict on every AI product. The necessary controls depend on the authority delegated and the consequences, not simply on whether the supplier has used the term AI.
Set evidence requirements before expanding autonomy and reassess after significant changes. Separate benefits from expectations and name an accountable decision owner and a responsible operating team empowered to act. Seek accountable automation serving a defined need. AI should not become the destination for decisions nobody owns. Sophisticated technology and responsibility are compatible; good governance makes that relationship clearer rather than providing an elaborate way to avoid it.
The reality is that implementation evidence and outcome evidence answer different questions. NIST’s June 2025 guide documents example architectures and integration lessons. That supports learning about designs and demonstrated behaviour, not a universal claim about financial return or breach probability. A successful example can justify investigation without establishing the same result everywhere. My objection is not to using other people’s work. It is to asking their evidence to carry conclusions it was never designed to support, then presenting the additional optimism as assurance.
Challenge a benefit claim by identifying what was observed. Was it successful integration, a blocked test path, changed incident experience or a modelled financial estimate? Ask about scope, baseline, conditions, exclusions and independent verification where available. Distinguish reported results from predictions and supplier interpretation. The practical evidence question in the FAQ section asks what can be learned from examples; this myth concerns the extra leap from those examples to a promised local benefit. Both require proportionate conclusions, but the latter also needs attention to transferability and attribution.
I would establish local measures before implementation. Repeat authorised access tests, review unjustified permissions and verify withdrawal or containment behaviour against a baseline. Include usability and continuity guardrails, failed cases and missing observations. Explain concurrent changes that might affect results. Fewer recorded incidents alone does not establish why the number changed. A financial case should identify assumptions and uncertainty rather than converting a general percentage into avoided losses. Use the implementation examples to improve design and testing, then assess the actual service rather than borrowing somebody else’s confidence.
Report observations, interpretation and forecasts separately. State what the evidence supports and what would strengthen it. Revise expansion decisions when results disappoint. The correction is disciplined judgement, not paralysis until perfect evidence appears. A blocked path can be valuable without proving a matching reduction in breaches. Evidence-led conviction accommodates uncertainty; defending a favourite assumption indefinitely is an expensive alternative to learning.
Turn myth-busting into better decisions
Take one live proposal or operating assumption and identify the claim most likely to affect the decision. Ask what it means, what it covers and what evidence would support it. If the claim is too broad to test, narrow it. Keep the useful part of the idea while removing unsupported certainty. The objective is a sound decision, not a victory over the person who first proposed it.
Choose a practical follow-up: request implementation detail, verify an access path, exercise recovery, clarify an owner or revise an overstated benefit. Use safe, authorised tests and involve service owners and affected people. Record results, limitations, next actions and review points. Where evidence does not support the claim, change the proposal rather than improve the wording until the gap becomes less noticeable.
Bring the findings back to the decision-maker and agree whether to proceed, adapt or stop. Carry unresolved risks into an owned action plan rather than leaving them in meeting notes. Use the FAQ guidance to turn the corrected understanding into delivery steps. Revisit the conclusion when the evidence changes. Reasoned conviction should help an organisation act with clarity, while leaving enough room to learn when reality declines to follow the presentation.
