A security improvement programme can deal with the most obvious weaknesses and still become outdated surprisingly quickly. People change roles, suppliers are replaced, businesses acquire new teams and new workflows appear. The controls and recovery assumptions that made sense at the end of a project may no longer reflect how the organisation operates.
Sustaining resilience doesn't require a constant succession of major programmes. It does need a modest, dependable way of noticing changes, examining the evidence and deciding what should happen next. The process must be light enough to survive ordinary business pressures, but strong enough to expose important assumptions that are no longer true.
Connect oversight to business change
Scheduled reviews are useful, though a calendar should not be the only reason to revisit an arrangement. A significant supplier change, a new essential application, an acquisition or an incident can justify an earlier discussion. Agree who updates the service dependencies and who determines whether the change needs further assessment.
Imagine an illustrative company launching a customer portal that gradually becomes its main ordering channel. A recovery target originally set for a supporting application might no longer be acceptable once that application has become central to sales. The technology may not have changed at all; its importance to the business has.
Service owners are essential to spotting these shifts. They can explain how commitments to customers, operating capacity or financial exposure have changed in ways that an IT asset inventory won't necessarily reveal.
The useful trigger is not change for its own sake, but a change that alters a dependency or consequence assumed at the last review. A supplier replacement may move administrative control, a new customer commitment may shorten the acceptable recovery period and an acquisition may add accounts outside an existing security process. I would make this a question in ordinary project and supplier approvals: which essential services are affected, which evidence needs updating and who decides whether further assessment is required? That is more dependable than hoping an annual review discovers the problem afterwards.
Keep the leadership review focused
I would make the regular discussion about decisions, rather than about filling a dashboard. What has materially changed? Which important facts are still unknown? Where is a decision overdue, and what have exercises or incidents shown? The supporting records can hold the detail so that the meeting stays manageable.
The Toolkit's Dashboard deliberately shows counts, not an overall security score. That distinction is important. One unresolved dependency affecting a critical business service may matter more than a dozen completed actions. The meeting needs to consider the significance of the exceptions, not just how many tasks have been closed.
There is no universal meeting schedule. As a practical starting point, I would suggest a monthly action review and a quarterly leadership discussion, with more frequent attention during significant change or remediation. The Cyber Governance Code of Practice also recommends exercising arrangements involving relevant stakeholders at least annually [S1].
It helps to separate decisions from routine delivery reporting. Actions progressing within an approved scope can remain with operational teams. Leadership can concentrate on accepted risks nearing expiry, material unknowns, changed dependencies and trade-offs requiring authority. A short meeting that resolves these matters is worth more than a detailed dashboard nobody challenges. The owner should explain the business consequence of each significant exception, not simply its status. One untested recovery method for the primary ordering channel can deserve more attention than several less material overdue actions.
Manage exceptions as decisions with an expiry
Not every weakness can be corrected immediately. An old application may need replacing, or a supplier may be unable to provide a requested service without a broader change. These constraints should be recorded clearly: the accountable business owner, the reason for deferral, any interim measures and the date for reconsideration.
An overdue task is not the same as an accepted risk. Acceptance needs an explicit decision by someone with the necessary authority. When the agreed date arrives, the business should consider the remaining exposure against its current circumstances rather than allowing the exception to carry on indefinitely.
Evidence has a shelf life as well. A successful restoration test shows what worked under a particular set of conditions, at a particular time. Changes to identities, infrastructure, software or suppliers may make that result less relevant. The evidence record should include its scope and when another check is needed, not just a permanent “complete” label.
Every accepted exposure needs a reason that can be challenged later. Perhaps replacement would interrupt a critical trading period, or a temporary control has been agreed while a supplier completes work. Record that context alongside the owner, interim measure and review date. When the date arrives, the business should ask whether the original assumptions still hold. Has the service become more important? Has the supplier's position changed? Has an exercise exposed a weakness? Renewal of an acceptance needs an explicit judgement; silence should not make the decision on leadership's behalf.
Bring AI into the same operating discipline
This is where the Practical Security and Resilience work connects naturally with the Practical AI series. A workflow that can examine customer data, send communications or update business records creates dependencies and permissions that need to be understood and controlled.
The first distinction is between an assistant that proposes text and a system authorised to do something with it. Actions with greater consequences may need tighter permission limits, meaningful approval, monitoring and a practical means of stopping or recovering the process. The service review should include data sources, integrations and service accounts, not just the AI model itself.
The NCSC's secure AI system development guidance considers security during design, development, deployment and ongoing operation [S12]. For a business buying an AI-enabled service, I would ask how the supplier and implementation team will maintain those controls when the underlying models, data or integrations change.
An AI pilot can become an operational dependency through apparently modest changes: a connector is introduced, permissions are broadened or a workflow begins writing to business records. The important question then moves beyond the quality of the generated text. Who is permitted to approve actions, how are errors detected and how can the business stop or reverse the work? These are familiar governance questions. The AI service belongs in the dependency map when essential operations rely on it, and its evidence should be revisited when its model, integrations or supporting data change.
Preserve learning and ownership
After a test or incident, the important question is whether the agreed improvements were made—and whether they resolved the problem. A recorded lesson without a decision or follow-up remains only an observation. Actions need business owners, a reason for being done and evidence of completion.
Over time, credible resilience comes from these fairly ordinary habits. Material changes prompt a review. Unknowns remain visible until they have been examined. Someone owns the difficult decisions. That is a more dependable basis for confidence than a dashboard showing everything as green.
I would favour a short, traceable chain: the event or change, the finding, the decision, the implemented action and evidence of its result. It helps retain learning when staff or suppliers change without creating a cumbersome process. A successful restoration test demonstrates a particular outcome under particular conditions; it is not a permanent assurance. Somebody should own the next review and recognise when changed dependencies have made the old evidence less relevant. Maintaining that link between action and consequence is what turns a one-off project into sustained management practice.
Sources and further reading
[S1] UK Government. Cyber Governance Code of Practice. Read original guidance
[S12] NCSC. Guidelines for secure AI system development. Read original guidance
The source collection records these references as checked on 28 October 2026. Source attribution does not imply endorsement.
Scenarios are illustrative, not client case studies. This article provides general business guidance, not a security audit, certification or substitute for sector-specific technical, legal or contractual advice.
