How to Navigate NIS2 Compliance for Your Business
· 9 min read · Updated
Lire en français
Start with the inventory, then keep walking past the server room
Achieving NIS2 compliance, which means meeting the obligations of the EU directive on the security of network and information systems, is essential for businesses operating within the EU, particularly for those managing critical infrastructure, and the first step is a thorough inventory of your IT and OT assets, to understand where you are exposed.
That first step is where most programmes quietly go wrong, because the inventory, even when it is done well, stops at the door of the workshop.
The assets that were never on anyone's list
Run the inventory properly and you will produce two lists.
The first is familiar, because it holds servers, applications, endpoints, identities and network equipment, and every item has an owner, a patch cycle and, usually, an access policy.
The second list is the one nobody has ever compiled. Machines, electrical cabinets, test benches, furnaces, presses, autoclaves, drives, in other words equipment where unauthorised use is not a confidentiality problem but a safety, counterfeiting or sabotage problem, and for most of these you cannot produce a list of who is entitled to operate them, and you cannot produce a log of who actually did.
This list is not missing through negligence, because it fell outside everyone's scope, too physical for IT and too digital for maintenance, and in a manufacturing, energy or water entity it is frequently longer than the governed-application list it sits beside.
Compiling it is, in our view, the most useful thing you can do in your first quarter of NIS2 compliance work, because everything downstream (risk analysis, access control policy, incident classification, supplier clauses) is scoped by it.
What Article 21(2) actually asks of you
NIS2 lists ten categories of risk-management measure, and three of them reach the shop floor directly, so they are worth reading with an OT asset in mind rather than a server.
- (c) business continuity and crisis management. An unauthorised start-up mid-cycle is a continuity event, not just a security one.
- (i) human resources security, access control policies and asset management. Note the pairing, because access control policy and asset management are named in the same breath, which is precisely the pairing that breaks when the asset is a press with a key switch.
- (j) multi-factor authentication or continuous authentication solutions. Written for logical access, but nothing in the text confines it there, and continuous authentication is exactly what an energised machine needs and almost never has.
The directive also frames the whole of Article 21(2) as an all-hazards approach, aimed at protecting network and information systems and the physical environment of those systems, and that phrase is doing more work than most NIS2 compliance programmes give it credit for.
The word "physical" is nearly absent from Article 21, which is why it gets missed
Read Article 21 looking for physical security and you will barely find it, and that absence has led a great many programmes to conclude, reasonably enough, that NIS2 is about networks and leaves doors and machines to somebody else.
The Commission does not read it that way, because its Implementing Regulation (EU) 2024/2690, adopted in October 2024, translates Article 21(2) into more than a hundred auditable technical requirements, and one of its thirteen annex sections is devoted to physical and environmental security, covering supporting utilities, perimeter and physical access controls, monitoring, and periodic review of those measures. The regulation attaches this to points (c), (e) and (i) of Article 21, even though physical security is not named there.
Be clear about scope, because this is where commentary gets sloppy, since that implementing regulation binds a specific subset of entities, namely DNS and TLD operators, cloud and data centre providers, CDNs, managed service and managed security providers, online marketplaces and search engines, social platforms and trust service providers, so if you run a factory it does not apply to you.
What it does give you is the Commission's own reading of what Article 21(2) means in practice, written down and directly applicable somewhere in the Union, and ENISA's technical implementation guidance, published in June 2025, mirrors its structure and includes the kind of evidence an auditor will expect to see. When your national supervisor or your customer's auditor asks what "appropriate measures" looks like for access control and asset management, this is, as far as we know, the most authoritative answer currently in writing, and it treats the power supply and the physical perimeter as in scope.
Planning as though physical access to machines is outside NIS2 is a bet against the direction of travel.
Incident response planning, tested against the shop floor
A robust incident response plan is a cornerstone of NIS2 compliance, because your plan should set out clear procedures for detecting, reporting and mitigating incidents, and the directive puts a clock on it.
The 24 / 72 / one-month cascade
Article 23 requires an early warning within 24 hours of becoming aware of a significant incident, a fuller notification within 72 hours including an initial assessment of severity and impact, and a final report within one month, and regular exercises are what make those windows survivable, because a plan that has never been rehearsed is a document, not a capability.
You cannot report what you cannot attribute
Here is where OT assets break the plan, and it is worth walking through before it happens rather than during.
At 22:40 on a Friday, a five-axis machine runs for forty minutes and produces eleven parts, and on Monday someone notices, so now start the clock. The chain that should be able to answer what follows is the one we describe in IAM at the end of the power cord.
Within 24 hours you owe an early warning, and within 72 hours you owe an initial assessment of severity and cross-border impact, but to produce either you need to answer three questions, who started it, under what entitlement, and was this unauthorised use or a scheduling error.
If the only record is the machine's hour counter (unattributed, unsigned, reset at the last service) you cannot answer, so you will file a notification that says something happened and you are investigating, and you will still be saying that a month later in the final report. That is not a reporting failure caused by bad process, it is a reporting failure caused by a missing log that was never designed to exist.
What a usable OT incident record looks like
The requirements are unlike your SIEM's, the tool that collects and analyses your IT logs, because volume is low, a few dozen events per machine per month, while unit value is very high, and the consumer is rarely a real-time detection engine, since it is a supervisor, an insurer, a prime contractor or a regulator, reading it much later.
That points to four properties:
- Local. The network was very likely down, or absent by design, at the moment that matters.
- Timestamped to a trusted clock. An OT clock that is twelve hours out or still on winter time makes every other property worthless.
- Chained and signed, so that a missing entry is visible rather than merely absent.
- Exportable without depending on any central infrastructure.
Attribution is the hard part, and it is an identity problem before it is a logging problem, because a log can only record the identity that the enforcement point actually established, which is why access control and evidence have to be designed together rather than in sequence. We have written separately about what a proof log proves and what it does not.
Supply chain: the pressure that arrives before the regulator does
Article 21(2)(d) makes supply chain security an explicit obligation, which means your customers' NIS2 compliance obligations become your questionnaire.
In practice this is the pressure that reaches a plant manager first, because automotive, aerospace, energy and defence primes are already sending cyber clauses down to tier-2 and tier-3 suppliers, and those clauses increasingly ask for things the shop floor cannot produce, such as a list of who may operate a given asset, evidence that qualifications were current, or a record of physical interventions by third-party technicians.
Answering "we don't track that" is a commercial problem long before it is a regulatory one.
Three places the standard playbook does not transfer
Briefly, because each deserves its own treatment.
Fail-closed is not automatic. In IT, denying on doubt (what is usually called failing closed) is the default, but cutting power to a ventilation system or a cooling unit is a safety decision before it is a security one, so the meaning of "deny" has to be set equipment by equipment, jointly with operations and HSE.
Connectivity is not an assumption. A decision point that calls a remote service is a decision point that abstains during an outage, precisely when risk peaks, so decisions have to be rendered locally and remain nominal, not degraded.
Most legitimate subjects are not yours. Temps, tier-2 subcontractors, OEM engineers flying in from another continent, and you have to verify an entitlement and tie it to a work order without ever having provisioned the person, often with no directory on the supplier's side.
Where transposition actually stands, and why it does not change your deadline
The directive required national law by 17 October 2024, and four member states met it. In November 2024 the Commission opened infringement proceedings against 23 of them, and issued reasoned opinions to 19 in May 2025. By mid-2026 the large majority had adopted transposing legislation, but a handful, including Spain and the Netherlands, still had not completed the process, and in July 2026 the Commission referred France, Ireland, Spain and the Netherlands to the Court of Justice, asking for financial penalties.
It is tempting to read that as breathing room, but it is not, for three reasons, because your customers' clauses do not wait for a parliament, because entities operating across several member states are already in scope somewhere, and because the obligations arriving in late-transposing states are, by design, the same ones. The organisations that used the delay to compile the OT inventory will be ready, while the ones that used it to wait will be doing the same work under a supervisor's timetable rather than their own. The rest of the 2027 calendar, including the Machinery Regulation and the Cyber Resilience Act, is covered in what to expect in 2027.
A practical sequence for NIS2 compliance on the shop floor
- Extend the inventory to OT assets with no access policy. Not every machine, but those where unauthorised use carries physical, safety or regulatory consequence.
- Define the meaning of "deny" for each of them, with operations and HSE in the room, because this is the hardest deliverable and it conditions everything else.
- Require an enforcement point that can decide offline. In our experience this single criterion eliminates unsuitable options faster than any feature comparison.
- Design the evidence trail with the access control, not after it, because you can only log the identity you actually established.
- Rehearse the 24-hour clock against an OT scenario, not an IT one, since that is where the plan will break, and better in an exercise.
And one thing not to do, do not connect equipment to the network for the sole purpose of governing it, because creating an attack surface in order to monitor the attack surface you just created is a poor trade. Physical access control does not require the machine to be smart, it requires the power outlet to be.
What this does not give you
Compliance work invites overstatement, so to be plain about the limits, controlling and logging access at the power supply answers the access control, asset management and evidence parts of Article 21(2), but it does not answer network segmentation, vulnerability handling, backup and recovery, cryptography policy or supplier due diligence. It is one column of the NIS2 compliance matrix, not the matrix.
It happens to be the column that is empty on almost every site, which is why it is worth starting there, and why it is worth being honest about where it stops.
Last reviewed: August 2026. NIS2 transposition and enforcement statuses change, so verify against your national competent authority before relying on any date here.
