Why identity and access management must protect people through account recovery, role changes and dependency outages

🧭 Executive summary
A successful login does not settle who controls an account, whether its permissions remain justified or whether the legitimate person can recover access safely. Identity and access management (IAM) brings together the policies, processes and systems governing who may access what. It needs to work throughout the human-account lifecycle. My argument is that leaders should buy and govern that whole service, not regard a working sign-in screen as the finished job.
The evidence supports that broader view. The United Kingdom’s (UK) National Cyber Security Centre (NCSC) connects appropriate access, proportionate authentication and joiner, mover and leaver processes with making legitimate work easier in its identity and access management guidance. Authentication means checking control of a registered means of signing in, rather than proving that every later action is appropriate. Joiner, mover and leaver processes grant, adjust and remove access as someone’s relationship with the organisation changes.
The United States’ (US) National Institute of Standards and Technology (NIST) treats digital identity risk as more than a problem for the organisation running the system. Its fourth-revision Digital Identity Guidelines also ask designers to consider the people using the service, including their privacy, experience and routes to redress. Redress means a way to challenge and resolve a problem with the identity service. I regard these as complementary parts of a dependable service. Individual design choices can still involve trade-offs that need to be explained and tested.
The Cyber Security Breaches Survey 2025/2026 supplies UK business and charity context, not a measure of a particular account’s risk or proof that an IAM product works. The joint Scattered Spider advisory revised in July 2025 describes support requests used to change account control. Its reporting supports scrutiny of recovery decisions, not blame directed at individual help-desk staff or an assumed explanation for every retail incident.
The practical decisions are to name an accountable business owner, agree risk-based access-removal timescales and fund usable recovery and controlled emergency access, an approved exceptional route when ordinary access is unavailable. Ownership must cover enrolment (establishing an account and its means of signing in), authentication, permissions, account recovery (regaining account control) and removal. Use strong authentication and make elevated administration limited and temporary where appropriate. Check that removal reaches applications and existing sessions, meaning continuing signed-in interactions, rather than stopping at a disabled central account.
Design around the people who actually depend on the service: customers, colleagues, contractors, residents, beneficiaries and volunteers. They will not all have the same devices, connectivity, accessibility needs or support arrangements. Require suppliers to explain their responsibilities, limitations, dependencies and full-service costs. Ask what happens when the ordinary route fails, without assuming that a pricing model, outsourcing decision or commercial incentive caused an incident.
Start with the most consequential human-account journeys and their owners. Agree representative, authorised tests of recovery, role changes and dependency failures, meaning loss of a service the identity process needs. Examine usability, rejected impersonation, actual access removal and restoration separately. Record the conditions, exceptions and untested assumptions, then correct, retest and review after material changes. These are proposed acceptance practices to validate locally, not guarantees of continuity, security or universal recovery success. The customer assurance question is whether the supplier can show dependable outcomes in conditions representative of your environment.
Why a valid login does not prove appropriate access

IAM answers several related but different questions. Identity proofing concerns the confidence needed in a claimed identity. Authentication checks control of an associated authenticator, a means used to sign in, such as a password or passkey. The credential records the binding between the account and its authenticators. A passkey is a device-managed digital key used instead of a password. Authorisation decides what the account may do. NIST separates identity proofing, authentication and federation, the arrangements through which one service relies on identity information from another, in its Digital Identity Guidelines. Keeping these distinctions clear prevents a strong login method being mistaken for a complete protection programme.
Authentication provides risk-based assurance, a level of confidence assessed against risk, not certainty about every action that follows. NIST’s authentication guidance describes checking that a claimant controls authenticators associated with an account. That does not, by itself, establish that a permission remains justified after a transfer, that a recovery request is legitimate or that the organisation can safely operate without one of its suppliers. Those are additional questions, requiring their own evidence and decisions.
Consider an illustrative example, not a reported incident. A colleague moves from payroll into another department. Their passkey still works perfectly, but their old payroll permissions remain. The authentication technology has done its job; the business process has not. The appropriate response is not another reminder about passwords. It is an effective role-change process that removes obsolete access, verifies the result and assigns responsibility when a connected application does not update.
Customers face a related problem from the other direction. Someone who loses a telephone may be the legitimate account holder while lacking the normal means of proving it. Imagine a recovery screen that only says, “enter the code sent to your phone”. That is illustrative wording, not a quotation from a named service, and it does not resolve the lost-phone problem. Conversely, a process that accepts a convincing story without adequate checks may restore access to the wrong person. The design question is how to offer a safe alternative, not whether security or customer service should win.
I therefore use “operating system of trust” as a business metaphor, not a claim that identity replaces every other safeguard. It describes the connected decisions that let people act, challenge suspicious requests and withdraw authority when circumstances change. The boundaries matter: this article concerns human accounts and their supporting processes. Machine-account lifecycle and cross-environment detection deserve separate treatment. Devices matter here only insofar as their condition, loss or unavailability affects a person’s access.
This approach follows the NCSC’s emphasis on who needs access and under what conditions. It also changes the executive conversation. Instead of asking whether IAM software has been installed, ask whether the organisation can demonstrate the right outcome for the person, the service and the information involved, including when the ordinary route fails.
What recent research and developments actually establish
The Cyber Security Breaches Survey 2025/2026, published on 30 April 2026, found that 43 per cent of businesses and 28 per cent of charities reported identifying a breach or attack in the 12 months before their interviews. Phishing, deceptive messages intended to trick people into revealing information or taking unsafe actions, was reported by 38 per cent of businesses and 25 per cent of charities. Those percentages describe the respective business and charity populations represented by the survey, not just respondents who reported an incident. Fieldwork, the survey data collection, took place from August to December 2025, so the look-back periods vary with interview dates. The 2026 publication date is not the measured period, and these figures are not a live measure of November 2026 activity.
The technical report describes a random-probability telephone and online survey (using a sample selected through known selection probabilities) of 2,112 UK businesses and 1,085 UK registered charities, with results weighted, meaning adjusted to represent those populations. Education institutions were also surveyed, but their findings are published separately. Sole traders, public bodies other than education, and businesses with no information-technology capacity or online presence were excluded. The main report warns that unidentified or unreported events may be missing. These are self-reported, recognised experiences, not an independent forensic count of successful breaches. They establish neither an account-takeover probability for a particular customer, meaning the likelihood of someone else gaining control of that account, nor the causal effectiveness of an IAM product. I would use them to explain why the issue matters, not as a prevalence estimate for the whole public sector or a verdict on any supplier.

The joint Scattered Spider advisory, revised on 29 July 2025, offers a more specific account of the threat. Its authors include the NCSC, the Federal Bureau of Investigation (FBI) and the Cybersecurity and Infrastructure Security Agency (CISA). Its Overview introduces the relevant list of techniques as based on public reporting. It describes criminals approaching support teams as if they were legitimate account holders, seeking password resets and changes that moved sign-in control onto the criminals’ devices. That account includes the transfer of multifactor authentication (MFA), which combines different kinds of evidence, such as something a person knows and something they possess. The advisory draws on investigations as recent as June 2025. I use it here as evidence of reported methods, not proof that every publicly discussed retail breach followed the same sequence.
That distinction protects us from an unhelpful habit: taking a plausible attack story and repeating it until it sounds established. My interpretation is that help-desk recovery and changes to sign-in methods deserve specific controls because the advisory shows how those decisions can transfer account control. My recommendation is to test those operations directly, including requests that sound urgent or credible. It is not to blame the individual who answers the telephone, or to infer without evidence that outsourcing or cost reduction caused a particular compromise.

Authentication has also moved forward. The NCSC recommends passkeys over passwords where available, while retaining strong passwords and two-step verification (a second check when signing in) where passkeys are unavailable. Its original analysis, Passkeys are more secure than traditional ways to log in, explains its 2026 change of recommendation and considers credentials across creation, use, synchronisation (copying credentials across devices), revocation (invalidating a credential) and recovery. That is a useful reason to revisit older plans built entirely around passwords and repeated approval prompts. It is not a reason to stop examining recovery, device replacement or permissions.
The July 2025 final edition of NIST Special Publication (SP) 800-63B-4 supplies an updated technical reference for authentication and the management of sign-in methods. Its companion base guidelines emphasise a risk-based, customer-centred approach. These sources support modernisation with clear limits: standards guide the design, while representative testing, meaning tests under conditions reflecting the intended service environment, provides bounded evidence of what a particular implementation did under the recorded conditions. Neither an attractive roadmap nor a supplier’s declaration closes that gap.
Protect human accounts from joining to leaving

Access should begin with a justified relationship and cease when the relevant authority ends. That access decision is separate from decisions about retaining an account record or business information. Between those points, people change jobs, take temporary responsibilities, leave projects or stop acting for a supplier. The NCSC recommends explicit joiner, mover and leaver processes, including temporary accounts and third-party access. This is ordinary operational discipline with security consequences, not an administrative task to leave until someone has spare time.
Start by distinguishing populations. Employees may have an authoritative employment record, the agreed source of employment facts. Volunteers may have a coordinator and a placement end date rather than a payroll event. A contractor may be sponsored by a service owner. Customers may need to change a delegated representative or close an account. My recommendation is to document the relevant trigger and responsible owner for each journey, rather than pretending every relationship can be managed through the same human resources data feed, an automated supply of employment updates.
A role change needs an explicit decision about what is removed as well as what is added. The NCSC Cyber Assessment Framework’s identity and access principle calls for access rights to be reviewed and technically removed when no longer required, including after a role change or departure. The Information Commissioner’s Office (ICO) similarly addresses these matters in its access-control toolkit, covering movers, leavers, temporary permissions and records demonstrating removal. The ICO page is under review following legislative changes, so its status should be checked when setting formal obligations.
Agree what timely removal means in practice. For each risk class, define the authorised trigger, when timing starts, which applications are covered and what counts as completed removal. An urgent suspension and an ordinary planned departure may need different handling. Do not invent one universal number and call it best practice. Agree timescales around the potential harm, service obligations and available safeguards, then establish whether the supplier and internal teams can meet them.
The evidence should show the outcome, not just a request. In a proposed acceptance test, a check against the agreed conditions for accepting the service, an owner would record when removal was authorised, when the identity platform acted and when the relevant applications rejected further access. Existing sessions, temporary administrative permissions and separately managed accounts would be checked where present. These are recommended test criteria, not a claim that every product has the same architecture or access-removal behaviour.
Also plan continuity for the work left behind. Transfer responsibility for records and approvals through authorised processes, without keeping the former user’s access alive for convenience. Where an application cannot remove access within the agreed limit, record the gap, apply an approved alternative safeguard and assign a correction date. A reassuring dashboard is no substitute for knowing which permissions remain usable and who owns the decision to tolerate them.
Use strong authentication and make privileged access temporary

Strong authentication remains essential. It should be matched to the consequences of getting the decision wrong, not to somebody’s job title or ability to tolerate inconvenience. The NCSC recommends multifactor authentication and additional safeguards for privileged accounts (accounts with authority to administer systems or other users), including separate accounts for administration and everyday work. A director’s email account, a finance approval and a system administrator’s console present different risks, but none deserves an automatic exemption because its user is busy or influential.
A passkey is an authenticator based on a digital key rather than a password that a person must create and remember. The NCSC explains that passkeys resist phishing and can be used through familiar device-unlocking methods. For leaders, the practical point is that a better control can also be easier to use. Ask how the proposed method works across the devices, shared working arrangements and accessibility needs that actually exist, rather than assuming a consumer demonstration proves enterprise suitability.
Do not turn that recommendation into an absolute promise. The Scattered Spider advisory demonstrates why changing the registered authentication method can itself become an attack route. My conclusion is that an organisation must examine the strength of enrolment, replacement and recovery alongside the normal login. Phishing-resistant authentication is designed to resist deceptive sign-in attempts. Buying it does not remove the need to govern who can change it or to decide whether an account still needs its permissions.
Privileged access means authority to perform consequential actions, such as changing security settings or administering other accounts. The NCSC framework highlights permissions capable of materially affecting essential functions. Itsprivileged access management guidance describes approved temporary access and limiting permissions to the task, with approval arrangements chosen according to risk. The reasoning is practical: a shorter period and narrower authority reduce the opportunity available through that permission; they do not make misuse impossible. I recommend defining the purpose, approver, scope, duration, logging (recording relevant actions) and expiry, then testing the result. Check that approvals remain usable during urgent recovery. This is a risk-based design recommendation, not proof of a universal product benefit or a feature every service provides.
Make support staff part of that design. A person who can reset a privileged account or change its authentication method may affect control of sensitive systems even without routinely administering them. Give them clear verification procedures, an escalation route and permission to pause a suspicious request. Avoid making them choose between following the procedure and meeting a target that rewards closing every call quickly. That is a proposed management safeguard, not an allegation about any named service provider.
Finally, rehearse exceptional administration. Cloudflare’s account of an unavailable emergency-disconnect control is a reason to check whether the fallback relies on the failed service, not evidence that every fallback does. Emergency access should be authorised, narrowly scoped and observable, with an explicit process for ending it. Test whether the staff who must recover a service can reach the necessary instructions and credentials when ordinary access is unavailable. Record the restrictions and remaining risks. The aim is not permanent emergency privileges hidden behind a reassuring label. It is a controlled route that supports recovery without quietly becoming the easiest route into the system.
Devices: the other half of every access request

A valid user on an unhealthy device is an incomplete trust decision. Laptops, telephones, servers, sensors and specialist equipment hold credentials, tokens and data. They also provide attackers with persistence. A device identity allows an organisation to recognise the endpoint, associate it with an owner and management state, and evaluate whether it meets security policy.
Device identity should be rooted in cryptographic material protected by hardware where feasible, enrolled through a controlled process and renewed or revoked through a managed lifecycle. Registration alone is not evidence of continuing health. Access decisions should consider encryption, secure boot, operating-system version, patch state, endpoint protection, configuration and signs of compromise. The National Cyber Security Centre’s Signals and policy engines guidance recommends combining independent signals about user identity, device identity, device health and behaviour. No single signal is perfect; assurance comes from their combination.
This does not mean that every minor deviation should block work. A risk-based policy might permit email from a personally owned mobile device while preventing downloads, require a managed endpoint for sensitive data, and reserve privileged administration for hardened workstations. The National Cyber Security Centre’s bring your own device policy guidance supports defining permitted tasks, services, data and technical controls according to risk. Where old operational technology cannot support modern controls, the answer may be network isolation, brokered access, enhanced monitoring and a funded replacement plan. Labelling the exception is not the same as managing it.
Bring your own device arrangements require candour. The National Cyber Security Centre’s Bring your own device guidance explains that these arrangements can improve flexibility but require organisations to balance usability, management and security risk. Organisations should define which services may be accessed, what security information is collected, what can be remotely removed and how personal data is separated. Employees should understand the bargain before enrolment. Contractors should receive a usable controlled route rather than being encouraged, implicitly, to improvise.
Machines, workloads and artificial intelligence agents

Non-human identities are becoming a larger and less visible part of the technology estate. Applications call application programming interfaces, workloads authenticate to cloud platforms, deployment pipelines sign software, devices send telemetry and artificial intelligence agents act across services. These identities may use certificates, keys, tokens or secrets, and they often persist for years because rotating them risks disruption.
The governance principle should be simple: every non-human identity needs an accountable owner, stated purpose, minimum privileges, approved credential method, expiry or review date and observable use. The National Cyber Security Centre’s cloud-security guidance on identity and authentication calls for robust authentication and lifecycle management for service identities as well as users. Embedded secrets and shared service accounts should be replaced with platform-managed identities and short-lived credentials where possible. Applications should authenticate as themselves, not as a generic administrator or a developer. Production and development identities should be separated. Human impersonation by automation should be exceptional, explicit and traceable.
Artificial intelligence agents heighten the issue because identity defines what an agent can see and do. An agent that reads mail, updates records and initiates transactions is not merely a software feature. It is an actor in the control environment. Leaders should require bounded permissions, human approval for material actions, clear provenance, tamper-evident logs and rapid revocation. The Microsoft Digital Defense Report 2026 describes identity as the primary control plane spanning human and non-human identities, including applications and agents. Vendor evidence is not neutral, but the underlying governance challenge is real and increasingly visible across platforms.
Design account recovery around the person who needs help

Design recovery alongside authentication, not after the preferred login method has been chosen. That is my reading of the NIST Digital Identity Guidelines, which consider how identity-service decisions affect the people using the service, their privacy and experience, other ways of interacting with the service and routes to redress. My recommendation is to give recovery its own owner, support arrangements and acceptance criteria, meaning the agreed conditions the service must meet. That is how customer focus becomes a service design decision rather than a reassuring slogan.
Begin with real journeys. In illustrative scenarios, a customer loses their only telephone, a volunteer changes their number, a colleague cannot use biometric recognition (recognition using characteristics such as fingerprints or facial patterns), or a person has no reliable internet connection. These are design scenarios, not frequency claims. Ask the people affected and the teams who support them which circumstances matter in your service. Avoid collecting a larger dossier of personal information simply because the digital route has become awkward.
A recovery route needs a credible way to establish that the request is legitimate without relying exclusively on the unavailable method. The joint help-desk attack evidence supplies the reason for testing impersonation resistance; NIST’s risk-based approach supplies the reason for proportionate assurance. The exact method needs technical and privacy review. A familiar voice, apparent seniority or convincing urgency should not, by itself, settle a sensitive recovery decision.
Usability should be demonstrated, not asserted. In a proposed test, a legitimate participant would attempt recovery using an agreed alternative, with only the information and assistance that would be available in practice. Record whether they succeed, what explanation they receive, how much help they need and whether they can challenge a mistaken refusal. Repeat with relevant accessibility needs and different levels of digital confidence. For a service intended to support unassisted recovery, success only with an engineer beside the participant would not meet that objective. An assisted route can be appropriate if its availability, safeguards and staffing match the agreed customer need.
Test the security side separately. An authorised assessor should attempt an agreed impersonation scenario under controlled conditions, without targeting unsuspecting people or real accounts. The evidence should show which checks prevented an inappropriate reset, how escalation worked and what was recorded. Passing the legitimate-user test does not establish resistance to impersonation, and rejecting an impostor does not prove the process is usable. Both outcomes matter, and neither should be hidden inside a single average score.
Recovery should restore the right level of access, not automatically re-create every historic entitlement. My recommended completion checks include reviewing relevant permissions, dealing with the lost or compromised authentication method and confirming that the account holder knows what changed. Where access cannot yet be restored safely, explain the next step and provide an appropriate service alternative. People should not have to invent workarounds, borrow somebody else’s account or abandon an essential service to satisfy our process.

A dependency is another service your process needs in order to work. For human-account protection, that might include an identity provider (a service that authenticates people and supplies identity information to other services), messaging service, employment feed, device check or support platform. My recommendation is to map these dependencies for normal login, recovery, role changes and emergency administration separately. A fallback route is not independent merely because it has a different name or appears on a different screen.
In its account of the 12 June 2025 outage, Cloudflare describes a failure affecting both entry and account maintenance. It reports that every identity-based login through its Access service, which controls application access, failed during the disruption. User-identity changes also stopped reaching that service through its provisioning route, which distributes account updates. The provider reports a duration of two hours and 28 minutes and traces the interruption to storage used by Workers KV, its key-value storage service, where an identifying key is used to retrieve a stored value. That storage depended in part on another cloud provider. Cloudflare describes the event as an availability failure, not an attack. This is a supplier’s primary account of its own service, with a commercial interest in the explanation, rather than an independent incident investigation.
The Access row in that incident account explains the security decision: without the necessary policy or identity information, the service refused entry rather than admitting the user. That behaviour is called failing closed. A separate fallback was affected too. Cloudflare reports that the emergency disconnect override for WARP, its connectivity software, could not be used because the same dependency was unavailable. These remain observations reported by the provider. They do not establish identical business consequences for every customer or a design shared by every identity product.
By its August 2025 follow-up, Cloudflare reported that it had put all Workers KV data onto its own infrastructure and was using that infrastructure to answer every request, while retaining third-party redundancy arrangements. My judgement is that fair scrutiny must acknowledge those reported changes rather than freeze the story at the incident. It must also distinguish a provider’s remediation account from an independent test of a customer deployment. Neither that account nor this article guarantees freedom from another outage. Ask what changed, what was tested and which assumptions remain.
For leaders, the lesson is not to demand that every system keep accepting access regardless of missing evidence. That could undermine the very protection you bought. Decide which activities must stop, which limited alternatives are acceptable and who may authorise exceptional access. A customer unable to transact, a frontline worker unable to reach a record and an administrator unable to repair a service may require different responses. Those differences need approval before an incident, not improvisation under pressure.
Restoration needs its own checks. In a proposed exercise, reconnecting a failed dependency would be followed by verification of queued role changes, pending access removals and temporary permissions. A green status page would not be treated as evidence that every outstanding change had completed. The end condition should include both resumed service and reconciled access. Test the people and communications around that work as well as the technology, with clear safety boundaries and a named authority to stop the exercise.

Make suppliers accountable for customer outcomes, not just features

Buying a service does not remove the need to understand who does what. Ask the supplier to distinguish its responsibilities from yours and those of any subcontractor: product design, configuration, identity verification, support decisions, application integration, recovery and access removal. My recommendation is to turn that division into observable outcomes and escalation routes. The term shared responsibility is a starting point for a conversation, not a satisfactory answer to an unresolved customer problem.
CISA’s Secure by Design programme places responsibility for customer protection on the product provider’s leadership. It argues against expecting consumers and small organisations to carry a disproportionate share of the security burden. The pledge’s Overview describes products arriving with secure defaults, meaning protective settings enabled from the outset, and with multifactor authentication, logging and single sign-on (SSO) available without an extra charge. SSO lets people use one organisational sign-in across connected services. These are CISA’s policy expectations, not this article’s findings about any particular supplier. Participation is voluntary and the pledge is not legally binding; a signature does not establish that the intended outcomes have been achieved.
That principle should inform a practical commercial discussion. Ask which subscription includes the controls you need, what support hours apply, whether recovery assistance carries additional charges and how evidence can be obtained. Compare the cost of the usable, supportable service, not a low headline licence price against a rival’s complete package. These are procurement recommendations. They do not establish that a particular pricing model is unsafe or that every organisation needs the most expensive tier.
Look at incentives on both sides. Does the contract reward rapid call closure without checking whether access was restored safely? Does the buying organisation demand a short implementation timetable while deferring recovery testing? Does a service credit meet an actual continuity need, or merely provide a financial remedy after disruption? These are questions to investigate, not explanations established by the cited incidents. A causal claim would need case-specific evidence, such as a documented decision to omit a relevant engineering safeguard, evidence that the commercial incentive drove that decision, and a credible link to the failure after considering other explanations. No such chain is established here. Technical complexity, implementation error and operational choices remain possible explanations; none is assigned to a named supplier without evidence.
Consider a report of 21 May 2025 for the Press Association (PA), published by The Independent. It attributes the short phrase “human error” to Marks & Spencer’s then chief executive, Stuart Machin, in reporting his account of entry through a third party. The wording is quoted from that secondary report; the original interview recording or transcript was not examined for this article. The report also describes disruption to online shopping. Separately, the company’s own Cyber update reports the taking of some personal customer information, but says that account passwords and usable card or payment details were not included. Neither account is a complete independent root-cause finding, and neither establishes that profit pressure, underinvestment or a supplier incentive caused the incident. My conclusion is to examine the process behind the reported explanation, not to use it as the end of the investigation.
I would ask suppliers for disclosure that helps customers make decisions: relevant limitations, material dependencies, incident learning and evidence of corrected weaknesses. Customers should reciprocate with accurate requirements, timely approvals and participation in testing. Responsible scrutiny is not hostility to suppliers. It is a way to make promises specific enough to be useful, and to avoid transferring unpriced risk onto the people least equipped to recognise or manage it.
Demand representative testing and evidence you can examine

The customer assurance question: “What evidence can you provide from testing in conditions representative of our environment that identity and access management remains usable and secure during account recovery, role changes and dependency outages, with access removed within our agreed timescales?” A feature brochure without relevant test records cannot demonstrate those outcomes. A reference customer can provide useful context, but experience with different devices, integrations, users and support arrangements does not by itself validate your environment.
Agree representativeness before the demonstration. Identify the account populations, important applications, authentication methods, support hours, accessibility needs and dependencies that the tests cover. Record differences from production. Use a safe test environment and non-sensitive accounts where possible, with explicit authorisation, a stop condition and a restoration plan. Where simulation cannot reproduce a material feature, label that limitation rather than treating an untested assumption as a pass. These are proposed acceptance practices, not reported supplier results.
Test recovery. Use an agreed lost-authentication scenario and an independently designed impersonation scenario. Check whether a legitimate person can follow the approved route, whether an inappropriate request is refused or escalated and whether the resulting permissions are correct. Record the support needed and the security decisions made. Set acceptable outcomes before the test; do not let the supplier redefine success around whichever part of the demonstration happened to work.
Test role changes and removal. Authorise a transfer or departure affecting the applications in scope. Record the start event and verify when obsolete access ceases in each relevant service, including established sessions where applicable. Test temporary privileges and a supplier-supported account as well as the easier centrally managed account case. Report exceptions individually. A quick average can conceal the one high-impact permission that remained active beyond the agreed limit.
Test unavailable dependencies. Under agreed controls, simulate the loss of a dependency needed for authentication, recovery or permission updates. Observe the intended denial, approved alternative or restricted operation. Check that support and emergency instructions remain reachable. After restoration, verify outstanding changes and remove exceptional permissions. The rationale follows the specific failures described in Cloudflare’s June 2025 incident account; the test design must still reflect your own architecture rather than reproduce somebody else’s outage for its own sake.
Ask for an evidence pack that records the date, service version, configuration, test conditions, expected result, observed result, relevant timestamps, exceptions and action owner. Include privacy-safe logs or equivalent records, not just screenshots of success messages. Use separate judgements for usability, security and timeliness. A formal supplier statement can inform confidence, but distinguish it from a witnessed test or independent assessment. None proves that future failures are impossible.
For UK personal-data processing within its scope, Article 32 of the United Kingdom General Data Protection Regulation (UK GDPR) requires technical and organisational measures appropriate to the risk, taking account of the processing context, state of the art and implementation costs. The ICO’s data-security guidance also addresses timely restoration and testing effectiveness. That supports the reasoning for recovery and outage exercises; it does not prescribe these exact scenarios, guarantee continuity or set one universal access-removal deadline. My recommendation is to use test findings to drive an agreed correction, risk decision and retest.
The linked ICO security guidance still carries a review notice following the Data (Use and Access) Act 2025. However, the ICO’s overview of that Act states that its provisions affecting data-protection law and the Privacy and Electronic Communications Regulations are now in force. A guidance-review notice should therefore not be read as postponing those duties, nor should that statement be expanded to every provision of the wider Act. Check the applicable legal regime, current text and sector-specific obligations for your processing. UK and European Union (EU) requirements are not treated as interchangeable here.

Conclusion: Leadership decisions and practical actions

Senior leaders do not need to select every credential or inspect every access log. They do need to decide who owns the outcome, which harms matter and what evidence is sufficient. The NCSC’s identity and access principle provides a foundation for this recommendation: understand and control access, review it and remove it when no longer required. Its essential-function context is not a universal legal mandate for every organisation. Turn that into a manageable programme with clear decisions, rather than another list of technologies awaiting purchase.
First, appoint an accountable business owner. Give that role authority to coordinate service owners, human resources or volunteer management, customer operations, technology, security, procurement and suppliers. Keep specialist responsibilities where they belong. The owner should resolve gaps between teams and report unresolved risks, not personally approve every account. Include representatives of the people who use and support the service so the operating model reflects actual needs.
Second, agree the service outcomes. Define access-removal timescales by risk, recovery arrangements, support availability and tolerable interruption for important journeys. Identify which exceptional access is allowed, who approves it and how it expires. Distinguish customer-account needs from workforce and privileged-account needs. Do not promise instant restoration or universal access continuity if neither the architecture nor the evidence supports it.
Third, commission a baseline and representative tests. A baseline is a recorded starting position for the access journeys being examined. As a proposed first-month action, identify the most consequential human-account journeys, their owners and their dependencies. Check existing evidence before buying additional tools. Select recovery, role-change and outage scenarios with the supplier, and record what has not been tested. The first month is an illustrative planning window, not a legal deadline or a forecast that every organisation can complete the work within it.
Fourth, correct and retest the highest-risk gaps. Over a proposed following two months, prioritise recovery weaknesses, overdue permissions, uncontrolled privileged access and dependency arrangements that cannot support essential work. Use phishing-resistant authentication where appropriate and feasible, following NCSC guidance, but fund the supporting process as well. Sequence work around risk and change capacity, not a claim that the same timetable suits every council, company or charity.
Finally, make assurance continuous and intelligible. At an agreed review interval, examine access-removal exceptions, recovery outcomes, support demand, failed tests and outstanding corrections. Repeat relevant tests after material changes to the service, its dependencies or support arrangements. Explain what the evidence covers and what remains uncertain. A smaller set of honest measures is more useful than broad coverage figures that conceal weak recovery routes or untested applications.
My view is that trust should be earned through service behaviour, not demanded from the person who has lost access. A customer should not need specialist knowledge to recover safely. A service-desk colleague should not have to improvise identity checks. A departing worker’s permissions should not remain active because ownership fell between teams. We cannot promise risk elimination. We can insist on clear responsibility, proportionate safeguards and evidence that the process works when circumstances change. That is what makes identity a dependable operating system of trust.
References and further reading (much more here)
Research cut-off: 11 November 2026.
References contains cited publications, source pages and definition sources.
Further Reading contains additional background and identity navigation.
- Cybersecurity and Infrastructure Security Agency. Secure by design.
- Cybersecurity and Infrastructure Security Agency. Secure by design pledge.
- Data (Use and Access) Act 2025, c. 18 (UK).
- Department for Science, Innovation and Technology, & Home Office. (2026a, April 30). Cyber security breaches survey 2025/2026.
- Department for Science, Innovation and Technology, & Home Office. (2026b, April 30). Cyber security breaches survey 2025/2026: Technical report.
- Federal Bureau of Investigation, Cybersecurity and Infrastructure Security Agency, Royal Canadian Mounted Police, Australian Signals Directorate’s Australian Cyber Security Centre, Australian Federal Police, Canadian Centre for Cyber Security, & National Cyber Security Centre. (2025, July 29). Scattered Spider (Joint Cybersecurity Advisory AA23-320A). Cybersecurity and Infrastructure Security Agency.
- Hartman, J., & Desai, C. J. (2025, June 12). Cloudflare service outage June 12, 2025. Cloudflare.
- Information Commissioner’s Office. Access control.
- Information Commissioner’s Office. Data (Use and Access) Act 2025.
- Information Commissioner’s Office. A guide to data security.
- Marks & Spencer. Cyber update.
- National Cyber Security Centre. 10 steps to cyber security: Identity and access management.
- National Cyber Security Centre. Passkeys: What you need to know.
- National Cyber Security Centre. Principle B2 identity and access control.
- National Cyber Security Centre. Use privileged access management.
- National Cyber Security Centre. (2018, February 5). Phishing attacks: Defending your organisation.
- National Cyber Security Centre. (2026, April 23). Passkeys are more secure than traditional ways to log in.
- The Privacy and Electronic Communications (EC Directive) Regulations 2003, SI 2003/2426 (UK, as amended).
- Robinson, A., & Trautmann, T. (2025, August 8). Redesigning Workers KV for increased availability and faster performance. Cloudflare.
- Saker-Clark, H. (2025, May 21). Marks & Spencer blames ‘human error’ as cyber attack set to cost £300m. The Independent.
- Temoshok, D., Fenton, J. L., Choong, Y.-Y., Lefkovitz, N., Regenscheid, A., Galluzzo, R., & Richer, J. P. (2025). Digital identity guidelines: Authentication and authenticator management (NIST Special Publication 800-63B-4, final, July 2025). National Institute of Standards and Technology. Alternative access to the same final publication: online authentication text.
- Temoshok, D., Proud-Madruga, D., Choong, Y.-Y., Galluzzo, R., Gupta, S., LaSalle, C., Lefkovitz, N., & Regenscheid, A. (2025). Digital identity guidelines (NIST Special Publication 800-63-4, final, July 2025). National Institute of Standards and Technology.
- United Kingdom General Data Protection Regulation, Regulation (EU) 2016/679, art. 32 (UK, as amended).
Additional background and identity navigation are listed here, including official home pages, the speaker biography, advisory-author institution pages and a passkey topic index. Organisation-name links provide identity navigation; linked publication titles identify the specific page or work. These pages do not substantiate incident causes or technical outcomes.
Publications used for definitions in the article or glossary remain in References because they are cited sources, not merely optional background. Article-specific operational definitions remain labelled in the glossary; no missing definition destination is invented. The FBI link remains its verified official organisation profile rather than an unresolved home-page destination. For terminology background, the NIST Digital Identity Guidelines and NCSC identity and access guidance underpin the glossary; their complete entries remain in References because the article cites them.
- Australian Federal Police. Homepage [Official home page].
- Australian Signals Directorate. Australian Cyber Security Centre [Official centre home page].
- Australian Signals Directorate. Reveal their secrets. Protect our own. [Official directorate home page].
- Canadian Centre for Cyber Security. Canadian Centre for Cyber Security [Official home page].
- Cloudflare. Cloudflare [Home page].
- Cybersecurity and Infrastructure Security Agency. Home page.
- Department for Digital, Culture, Media and Sport. Department for Digital, Culture, Media and Sport [Official organisation home page; publisher-identity background]. GOV.UK.
- Department for Science, Innovation and Technology. Department for Science, Innovation and Technology[Organisation page]. GOV.UK.
- European Union. Your gateway to the European Union [Official home page; navigation only].
- Federal Bureau of Investigation. About [Official organisation profile].
- Home Office. Home Office [Organisation home page]. GOV.UK.
- Independent. The Independent [Home page].
- Information Commissioner’s Office. Information Commissioner’s Office [Home page].
- Marks & Spencer. Homepage [Corporate home page].
- Marks & Spencer. Our leadership [Stuart Machin profile; biography navigation only].
- National Cyber Security Centre. The National Cyber Security Centre [Home page].
- National Cyber Security Centre. Passkeys [Topic index; background navigation only].
- National Institute of Standards and Technology. National Institute of Standards and Technology [Home page].
- Press Association. Press Association [Home page].
- Royal Canadian Mounted Police. Royal Canadian Mounted Police [Official home page].
| Term | Definition | Source (link) | Categories |
| Acceptance criteria | The agreed conditions the service must meet. This is the operational meaning used in the article. | Definition-source unresolved. NIST risk-based design guidance supplies context, not a verified lexical definition. | Security assurance; Supplier assurance |
| Acceptance test | A check against the agreed conditions for accepting the service. | Definition-source unresolved. NIST guidance supplies context for the article’s proposed test, not this exact protocol. | Security assurance; Supplier assurance |
| Account provisioning | Distribution of account updates to connected services. In the case study, the provider reports that user identities were not updated. | Cloudflare incident account: Access impact | Account lifecycle; Architecture |
| Account recovery | Regaining control of an account when the ordinary means of signing in is unavailable. | NIST SP 800-63B-4 | Account recovery; Authentication |
| Account takeover | Someone other than the legitimate account holder gaining control of the account. | Joint Scattered Spider advisory (observed account-control context) | Identity and access management; Threats |
| Authenticator | A means used to sign in, such as a password, passkey or security key. | NIST SP 800-63B-4 | Authentication |
| Authentication | Checking control of a registered means of signing in; it does not establish that every later action is appropriate. | NIST SP 800-63B-4: Introduction | Authentication; Identity and access management |
| Authorisation | Deciding what an account may do. | NCSC identity and access management guidance | Identity and access management; Access control |
| Baseline | A recorded starting position for the access journeys being examined. This is the article’s operational meaning. | Definition-source unresolved. NCSC identity guidance provides lifecycle context, not a verified definition of this review baseline. | Security assurance; Governance |
| Biometric recognition | Recognition using characteristics such as fingerprints or facial patterns. | NIST SP 800-63-4 | Authentication; Human-centred security |
| Cloudflare Access | The provider’s service for controlling access to applications, discussed in the June 2025 incident. | Cloudflare incident account: Access impact | Product-specific; Access control |
| Cloudflare WARP | The provider’s connectivity software discussed in the incident. WARP is retained as a product name, not expanded as an acronym. | Cloudflare incident account: WARP impact | Product-specific; Service resilience |
| Credential | The record binding an account to its authenticators. It is distinguished here from the authenticator used to sign in. | NIST SP 800-63-4 | Authentication; Account lifecycle |
| Credential revocation | Invalidating a credential. | NIST SP 800-63B-4 | Authentication; Account lifecycle |
| Credential synchronisation | Copying credentials across devices, as discussed in the article’s passkey lifecycle explanation. | NCSC passkey analysis | Authentication; Account lifecycle |
| Dependency | Another service that the identity process needs in order to work. A dependency failure is loss of that required service. | Cloudflare incident account | Architecture; Service resilience |
| Emergency access | An approved exceptional access route used when ordinary access is unavailable. This is the article’s operational meaning. | Definition-source unresolved. NCSC privileged access guidance supports approval and scope controls, not a universal emergency-access definition. | Privileged access; Service resilience |
| Emergency disconnect override | An emergency way to disconnect the provider’s connectivity software; the article describes the WARP control reported unavailable during the incident. | Cloudflare incident account: WARP impact | Product-specific; Service resilience |
| Enrolment | Establishing an account and its means of signing in. | NIST SP 800-63-4 | Account lifecycle; Identity and access management |
| Evidence pack | A record of test date, service version, configuration, conditions, expected and observed results, timestamps, exceptions, ownership and supporting records. | Definition-source unresolved for this exact article-specific record. ICO testing-effectiveness guidance provides the rationale. | Security assurance; Supplier assurance |
| Failing closed | Denying access when the service cannot obtain the identity or policy information needed to make the access decision. | Cloudflare incident account: Access impact | Access control; Service resilience |
| Federation | Arrangements through which one service relies on identity information from another service. | NIST SP 800-63-4 | Architecture; Identity and access management |
| Fieldwork | The survey data-collection activity, distinguished from the later publication date. | Cyber Security Breaches Survey technical report | Research methods |
| Human resources data feed | An automated supply of employment updates used as an input to the access lifecycle. | Definition-source unresolved. NCSC lifecycle guidance supplies context; the feed wording is the article’s operational explanation. | Account lifecycle; Governance |
| Identity and access management (IAM) | The policies, processes and systems governing who may access what. | NCSC identity and access management guidance | Identity and access management; Governance |
| Identity proofing | Establishing confidence in a claimed identity. | NIST SP 800-63-4 | Identity and access management; Account lifecycle |
| Identity provider | A service that authenticates people and supplies identity information to other services. | NIST SP 800-63-4 | Architecture; Authentication |
| Joiner, mover and leaver processes | Processes that grant, adjust and remove access as a person’s relationship with an organisation changes. | NCSC account-lifecycle guidance | Account lifecycle; Governance |
| Key-value storage | Storage that retrieves a stored value using an identifying key. | Cloudflare Workers KV incident account | Architecture; Product-specific |
| Logging | Recording relevant actions so that access decisions and activity can be examined. | NCSC audit-record guidance | Security assurance; Access control |
| Multifactor authentication (MFA) | Authentication combining different kinds of evidence, such as something a person knows and something they possess. | NIST SP 800-63B-4 | Authentication |
| Passkey | A device-managed digital-key authenticator used instead of a password. | NCSC passkey explainer | Authentication |
| Phishing | Deceptive messages intended to trick people into revealing sensitive information or taking unsafe actions. | NCSC organisational phishing guidance | Threats; Human-centred security |
| Phishing-resistant authentication | Authentication designed to resist deceptive sign-in attempts; it is not presented as protection against every account threat. | NCSC passkey explainer | Authentication; Threats |
| Privileged access | Authority to perform consequential actions, such as changing security settings or administering accounts. | NCSC B2 identity and access principle | Privileged access; Access control |
| Privileged access management | Controls for elevated administration that can use approved temporary access and permissions limited to the task, with risk-sensitive approval arrangements. | NCSC privileged access management guidance | Privileged access; Governance |
| Privileged account | An account with authority to administer systems or other users. | NCSC privileged-account guidance | Privileged access; Account lifecycle |
| Random-probability sampling | Sampling through known selection probabilities, rather than treating a respondent group as automatically representative. | Cyber Security Breaches Survey technical report | Research methods |
| Redress | A route to challenge and resolve a problem with an identity service. | NIST SP 800-63-4: Introduction | Human-centred security; Account recovery |
| Representative testing | Testing under conditions reflecting the intended service environment, with differences and limitations recorded. | Definition-source unresolved for a single formal definition. NIST risk-based design guidance provides context for the article’s operational meaning. | Security assurance; Supplier assurance |
| Risk-based assurance | A level of confidence assessed against risk, rather than certainty about every action. | NIST SP 800-63B-4: Introduction | Authentication; Security assurance |
| Secure defaults | Protective settings enabled from the outset, rather than left for the customer to add. | CISA Secure by Design Pledge: Overview | Governance; Supplier assurance |
| Session | A continuing signed-in interaction, relevant when checking whether access has actually stopped. | NIST SP 800-63B-4 | Authentication; Access control |
| Single sign-on (SSO) | Using one organisational sign-in across connected services. | NCSC single-sign-on guidance | Authentication; Architecture |
| Survey weighting | Adjustment of survey results to represent the relevant populations; it does not turn self-report into an independent forensic count. | Cyber Security Breaches Survey technical report | Research methods |
| Two-step verification | A second check when signing in, retained as a fallback where passkeys are not available in the cited NCSC advice. | NCSC passkey explainer | Authentication |
| United Kingdom General Data Protection Regulation (UK GDPR) | The UK data-protection regulation referred to in the article for risk-appropriate security of in-scope personal-data processing under Article 32. | UK GDPR Article 32 | Data protection; Law and regulation |
| Workers KV | Cloudflare’s key-value storage service discussed as a dependency in its June 2025 incident account. | Cloudflare incident account | Product-specific; Architecture; Service resilience |

Leave a Reply