The Reporting Duty Is a Detection Problem
The NIS2 reporting rules read like a documentation exercise with three notifications, a few defined fields each and a government portal. That reading misleads. The obligation asks you to notice a significant security incident and file an early warning within 24 hours of becoming aware of it. The deadline hangs on the word aware, because the 24-hour clock starts when you become aware, not when you finish investigating. If your logging and monitoring cannot surface a serious compromise for three days, you have not yet missed the reporting deadline, but you have absorbed three days of damage and will have to explain whether your detection meets the risk-management duties.
The question is therefore whether you can detect a significant incident in time, and less whether the form gets finished in time. Most Mittelstand firms that NIS2 pulled into scope cannot do that today. These are manufacturers, waste operators, food producers and digital providers who never thought of themselves as regulated infrastructure. They run firewalls, antivirus and backups, but they collect no logs centrally, have no alerting tuned to what "significant" looks like for them, and no one whose job is to look. The reporting duty makes that gap legally visible.
Are You In Scope? Nobody Sends a Letter
NIS2, formally Directive (EU) 2022/2555, widened the old NIS regime from a short list of named operators to whole sectors, sorted by size. Nobody adds you to a register and nobody notifies you that you are covered. The directive works on self-identification: you have to determine yourself whether you fall in scope, register with the competent authority and comply, whether or not anyone tells you to. "We were never contacted" is not a defence.
Essential vs important entities
Scope turns on your sector and your size. The directive splits covered organisations into essential and important entities, which differ mainly in supervision: essential entities face proactive oversight and important entities are supervised reactively, after something goes wrong, while the reporting duties are effectively the same. As a rough guide: large enterprises (around 250+ employees, or over €50m turnover and €43m balance sheet) in the high-criticality sectors are essential; medium enterprises (roughly 50+ employees, or over €10m in turnover and balance sheet) land as important, as do large enterprises in the "other critical" sectors. Some entity types are in scope regardless of size: DNS providers, TLD registries, qualified trust service providers, certain telecoms, and operators designated as critical.
Do not estimate this. The sector annexes and the size calculation have edge cases such as group structures, partial activities and headcount versus turnover tests, and these decide the classification. A mistake costs you in both directions, with unnecessary cost if you over-scope and unbudgeted liability if you under-scope. I write as an engineer, not as a lawyer. The classification is a legal determination, and a qualified adviser and the competent authority's own guidance help with it.
The self-identification duty
Because you assess scope yourself, the work starts with a defensible, written scoping decision and not yet with technology. It records which entity, which sectors and which size class are covered, and it contains the reasoning, dated and kept. If a supervisor ever asks why you did or did not register, you answer with that document. It costs little today; after an incident you have to reconstruct it at high cost.
The Article 23 Reporting Cascade
Article 23 of the directive lays out a staged cascade instead of a single report. Each stage has its own deadline and its own required content, and the later stages update the earlier ones as your understanding improves.
Each stage has to contain the following:
- Early warning — within 24 hours of awareness. It is a short first signal, not an analysis. It flags whether the incident is suspected to be caused by unlawful or malicious action, and whether it could have cross-border impact. That is roughly all it needs, which is why the bottleneck is noticing and not writing.
- Incident notification — within 72 hours. It updates the early warning and adds an initial assessment: severity and impact, and, where available, indicators of compromise. With it you commit to a first defensible picture of what happened.
- Final report — no later than one month after the incident notification. It describes the incident in detail with its severity and impact, and names the type of threat or root cause that likely triggered it, the mitigation measures applied and ongoing, and cross-border effects where relevant. If the incident is still ongoing at the one-month mark, you report progress then and file the final report within a month of handling it.
A "significant" incident triggers the cascade, and the directive defines it as one that has caused or is capable of causing severe operational disruption or financial loss to you, or considerable material or non-material damage to others. The competent authority may also request an intermediate status update at any point. You may know the 72-hour clock from the GDPR. NIS2 applies the same reflex to operational security, and a single event can trigger both duties. What the GDPR otherwise asks of a system is covered in AI and GDPR in Practice.
Germany: The Law Is Now in Force
For a long time Germany had the directive but no national law. The EU transposition deadline expired on 17 October 2024, and Germany missed it by more than a year. The Bundestag passed the NIS-2-Umsetzungs- und Cybersicherheitsstärkungsgesetz (NIS2UmsuCG) on 13 November 2025, and it entered into force on 6 December 2025. The duties are now live German law, implemented through the BSI-Gesetz.
The competent authority is the BSI (Bundesamt für Sicherheit in der Informationstechnik). Registration and reports go through a joint reporting office of the BSI and the Federal Office of Civil Protection and Disaster Assistance (BBK), and § 32 BSIG calls the stages there “frühe Erstmeldung”, “Meldung” and “Abschlussmeldung”. The transposition expands the population of covered organisations from a few thousand KRITIS operators to a figure the legislature put at roughly 29,000–30,000 entities. A mid-sized manufacturer or supplier who has never dealt with the BSI now has to register itself and be able to report. The statutory registration deadline of three months under § 33 BSIG expired on 6 March 2026; if you are in scope and not yet registered, do it without delay. Confirm the current mechanics against the BSI's own published guidance and not against a blog, including this one.
What "Able to Report in 24 Hours" Actually Requires
Technically one target remains: on any given day you can notice a significant incident and report it within 24 hours. That capability has four parts, and the most demanding one is not a document.
Detection is the long pole
Surfacing the incident takes most of the work and most of the money. For that you collect the logs of the systems that matter in one place and retain them long enough to use them. Most firms skip the next step and do not tune the alerting so that a significant event stands out from the daily noise. A log nobody queries only stores and detects nothing. This work is called detection engineering, and it overlaps heavily with securing the systems, which is why the same rules apply here as in securing software: know your assets, watch the boundaries, and assume you will have to reconstruct what happened.
Good logging design makes detection easier. If your systems, AI ones included, emit structured, queryable audit logs by default, you reconstruct what happened in an afternoon. I recommended the same for the EU AI Act Article 12 log schema: decide what a usable event record looks like before the incident, not during it. Writing detection rules is a discipline of its own. We built our validation layer for Microsoft Sentinel detections because turning "we want to catch X" into a validated detection query is slow, expert work, and because that work pays back every time an alert fires correctly.
The other three parts are lighter, but they do not cover the job on their own:
- An incident-response runbook. A short, rehearsed procedure settles who decides an event is a "significant incident," how you start and record the clock, who files, and how you track the 72-hour and one-month deadlines. Settle it on a calm afternoon and not at 2 a.m.
- A named responsible person. One owner answers for detection and reporting and has a deputy. Naming "everyone in IT" names no one, and the 24-hour clock does not pause for holidays or notice periods.
- Pre-drafted notification templates. Fill out the early-warning and 72-hour notification fields to a skeleton in advance, so that under pressure you complete a form and do not compose one. This part is paperwork, and you do it once, in advance.
I would not buy a large SIEM platform and a managed detection contract on day one just because a compliance deadline appeared. A mid-sized firm often spends heavily on tooling that way and still fails to detect the incidents that matter, because nobody tuned the alerts and nobody owns them. Start from the incident types that would be "significant" for your operation, instrument for those and build out from there. Size detection to fit instead of maximising it.