Suppose an aluminium foundry whose executive committee receives, every year, the same page on its cyber resilience best practices: daily backups in place, disaster recovery plan up to date, crisis unit designated. Nothing on this page says whether a full restoration has ever succeeded, how long it took, or whether anyone has ever tried to convene the unit on a Sunday.
A resilience practice is judged on this difference. We present none of them as the best: a practice is worth the test that has proven it in the organisation that applies it. Four of them lend themselves to an assessment in calm times: backups genuinely restored, plans tested, dependencies known, a crisis decision prepared. For each, observable signs separate proven resilience from declared resilience, and each gap calls for a specific remedy.
A public framework for placing cyber resilience best practices
The online glossary of NIST, the US National Institute of Standards and Technology, defines cyber resiliency. In our paraphrase, it is the ability to anticipate, withstand, recover from and adapt to adverse conditions, stresses, attacks or compromises on systems that use or are enabled by cyber resources. The glossary gives as its source, among others, publication SP 800-160, volume 2, revision 1.
The same institute publishes a cybersecurity framework, the Cybersecurity Framework (CSF), whose version 2.0 dates from 26 February 2024. It arranges its expected outcomes under six functions: Govern, Identify, Protect, Detect, Respond and Recover. Its summary specifies that it does not prescribe how these outcomes should be achieved and refers, for that, to online resources.
Several of these outcomes, quoted here, speak of tests. The framework expects backups to be “created, protected, maintained, and tested” (PR.DS-11). It also expects their integrity to be verified before any restoration (RC.RP-03) and improvements to be drawn from tests and exercises, including those carried out with the suppliers and third parties concerned (ID.IM-02).
The assessment we propose sticks to what can be observed without waiting for an incident.
| Practice | Sign of declared resilience | Sign of proven resilience |
|---|---|---|
| Backups | A report states that backups ran | A full restoration has succeeded in an empty environment, and its duration is known |
| Business continuity and disaster recovery plans | A validated document, dated from when it was written | A dated test report, with the gaps identified and their follow-up |
| Dependencies | A list of applications | For each essential activity, the chain of components and suppliers, and the restart order |
| Crisis decision | A crisis unit organisation chart | Named holders and deputies, reached during a test without going through the company’s email |
Backups restored, plans exercised
An indicator that measures the writing of backups, never their reading back, signals declared resilience. The gap calls for a restore test: rebuilding in an empty environment the application the business has ranked highest, directory and administration tools included, then timing it. The duration recorded and the age of the restored data are then compared with the two objectives defined in our entry RTO RPO.
A second check requires no test: can a production administration account delete the backups? If so, an attacker who takes control of it can too. The remedy is a copy that these accounts cannot reach; our entry on immutable backup describes the mechanisms.
The same reasoning applies to plans. A business continuity or disaster recovery plan that no one has exercised describes an intention. A few checks on documents are enough to see this: does the document carry the date of a test, do the on-call numbers still answer, does the diagram take account of the latest migration? The remedy involves tests of increasing scope: a tabletop review, the actual failover of one application, then a broader exercise. None is closed until its gaps have an owner and a deadline.
For financial entities, the DORA regulation makes this an obligation: its Article 11, paragraph 6, requires ICT (information and communication technology) business continuity plans and response and recovery plans to be tested at least once a year. Entities that its Article 16 subjects to a simplified framework must, for their part, test their plans “regularly”.
Dependencies known before the incident
A recovery can stumble over a component that no one had linked to the activity: the directory on which all authentication depends, a name service, a licence checked with a vendor, a supplier that alone holds an access. When the inventory stops at applications, resilience is merely declared. It becomes verifiable when, for each essential activity, the chain is written down to the suppliers and the restart order follows from it.
The CSF 2.0 formulates this outcome for suppliers: they are known and prioritised by criticality (GV.SC-04), and the relevant suppliers are included in incident planning, response and recovery activities (GV.SC-08).
Closing this gap requires two pieces of work. The first is documentary: tracing the chain with the business and operations. The second is a matter of observation, because a routine incident can reveal a dependency that no diagram showed. Operations that trace user journeys and draw lessons from each incident, as described on our SRE and observability page, keep this map up to date. They also train teams in recovery, on a small scale, well before a disaster.
A crisis decision prepared in calm conditions
The last assessment concerns decisions. Isolating the network, suspending a workshop, triggering the recovery plan: for each, do a name and a written criterion exist?
A crisis unit described by functions, with no names or deputies, reachable only through the company’s email, falls under declared resilience. The opposite signs are checked one by one: named holders and deputies, written triggering thresholds, a means of contact that works with email down, a call test carried out outside working hours.
What is then missing is a mandate from executive management, followed by an exercise. The course of a crisis, its two units and its notifications are the subject of our article on cyber crisis management. Only the part that can be checked in calm times counts here.
These four assessments are carried out without an incident and without any new tool. Building the plans, the tests and the control of third parties they then call for are described on our cyber resilience page. A useful first test concerns the application the business has ranked highest: restore it in an empty environment, then present to the executive committee the duration observed alongside the duration promised.