Summary
- Article 23(3) defines a significant incident as one that has caused or is capable of causing harm — a possibility test with no number in it, applied by the entity that must report.
- The Commission's numeric version of that test, Regulation (EU) 2024/2690, binds exactly eleven provider categories; every other essential and important entity keeps the words, with no duty to record the determination or the date it became aware.
Read Article 23(3) of Directive (EU) 2022/2555 and the first thing you notice is what is not there: a number. An incident is significant, the directive says, if it "has caused or is capable of causing" severe operational disruption of the services or financial loss for the entity concerned; or if it "has affected or is capable of affecting" other natural or legal persons by causing considerable material or non-material damage (consolidated text). Both limbs are tests of possibility. Neither is measured.
Europe's most-copied incident-reporting chain hangs off that sentence. The early warning at 24 hours, the notification at 72, the intermediate report on request, the final report a month later — all of it begins when an entity concludes that the sentence applies to it. The question worth asking is who gets to apply it with numbers, who does not, and what the record shows afterwards.
One paragraph, two instructions. Article 23(11) is where the Commission was told what to do about the missing number, and it carries two different instructions in the same paragraph. The first subparagraph is permissive: the Commission "may adopt" implementing acts further specifying the type of information, the format and the procedure of a notification. The second is not. "By 17 October 2024, the Commission shall", with regard to DNS service providers, TLD name registries, cloud computing service providers, data centre service providers, content delivery network providers, managed service providers, managed security service providers, and providers of online marketplaces, of online search engines and of social networking services platforms, adopt implementing acts further specifying when an incident is significant. For other essential and important entities the verb slips back to "may adopt" (consolidated text, ELI presentation).
The mandatory act exists. Commission Implementing Regulation (EU) 2024/2690 was adopted on 17 October 2024 and published in the Official Journal on 18 October 2024, and its procedure record gives the legal basis as Article 23(11), second subparagraph, and Article 21(5), first subparagraph (in-force record, procedure page).
Eleven categories, and a closed list. Article 1 of that regulation, headed "Subject matter", is a single paragraph, and it performs the narrowing the whole piece turns on. It applies "with regard to" DNS service providers, TLD name registries, cloud computing service providers, data centre service providers, content delivery network providers, managed service providers, managed security service providers, providers of online market places, of online search engines and of social networking services platforms, and trust service providers — "the relevant entities" (Official Journal HTML, Official Journal PDF). The full official title reproduces the same enumeration, category for category (catalogue record).
Count them: eleven. A hospital is not one. Neither is a water utility, a postal operator, a public administration or an energy operator. They may be essential or important entities under NIS2 and carry every duty in Article 23 — the 24 hours, the 72 hours, the month — but the only Commission instrument that converts "significant" into figures does not reach them.
What the numbers look like. For the eleven, Article 3(1) supplies the test. Direct financial loss exceeding EUR 500 000 or 5 % of the relevant entity's total annual turnover in the preceding financial year, whichever is lower. Exfiltration of trade secrets as set out in Article 2(1) of Directive (EU) 2016/943. The death of a natural person. Considerable damage to a natural person's health. A successful, suspectedly malicious and unauthorised access capable of causing severe operational disruption. Recurring incidents under Article 4, or the service-specific criteria in Articles 5 to 14. Article 3(2) carves out scheduled interruptions and the planned consequences of scheduled maintenance (Official Journal PDF).
Set that beside Article 23(3) and the relationship is plain: the regulation narrows a possibility test. It does not replace a harm threshold, because the directive never wrote one.
Then the service-specific figures, where the drafting turns physical. For a data centre, Article 8 treats an incident as significant where a data centre service is completely unavailable; where availability is limited for more than one hour; where the integrity, confidentiality or authenticity of stored, transmitted or processed data is compromised as a result of suspectedly malicious action; or where physical access to a data centre is compromised.
For a content delivery network, Article 9 puts complete unavailability at more than 30 minutes, and limited availability at more than 5 % or more than 1 million Union users, whichever is smaller, for more than one hour.
Note the pairing inside Article 8. Complete unavailability carries no duration at all; limited availability carries an hour. The zero, the thirty minutes and the hour are each printed. That one limb sits at zero while its neighbour sits at sixty minutes is my comparison, not a contrast the Commission draws (Official Journal PDF).
Everyone else keeps the words. For an entity outside the eleven, Article 23(3) is the whole test, and it is a test of possibility applied by the party that must report. No figure, no table, no annex. There is also no published Commission gloss telling a hospital how the words land on its own services.
That is the finding, and it is structural rather than scandalous. Two delegations were printed in one paragraph. The one with a deadline produced an instrument binding eleven categories. The one about form and procedure, which would bind everyone, has produced nothing readable yet.
The clocks, and what anchors them. Article 23(4)(a) requires an early warning "without undue delay and in any event within 24 hours of becoming aware of the significant incident". Point (b) requires the incident notification "within 72 hours of becoming aware". Both hang on awareness — and awareness of the significant incident, meaning the significance determination has already happened inside the entity before either clock starts.
Article 23 then does something a reporting regime rarely does. It does not require the entity to record, document, retain or publish that determination, and it does not require the entity to record the date on which it became aware. The only assessment artefact the article names is the "initial assessment of the significant incident, including its severity and impact" that must travel inside the 72-hour notification (consolidated text, ELI presentation).
The two intervals that carry a number are therefore measured from an event the instrument never asks anyone to timestamp.
Two documents that do not start at the same place. Article 23(4)(c) requires an intermediate report on relevant status updates "upon the request of a CSIRT or, where applicable, the competent authority". It has no interval of its own; it begins when somebody asks. Article 23(4)(d) requires the final report "not later than one month after the submission of the incident notification under point (b)" — a month that runs from the entity's own filing, not from the incident and not from the authority's request.
Two of the four documents run on intervals the entity controls by acting or by being asked. Two run from a moment nothing documents.
The two 24-hour phrases. Article 23(5) is the authority's side of the early warning: a response "without undue delay and where possible within 24 hours of receiving the early warning", including initial feedback and, on request, guidance or operational advice on possible mitigation measures. Set the phrases beside each other — the entity's "in any event within 24 hours" and the authority's "where possible within 24 hours" — and the qualifier does the work. One is a ceiling. The other is a target with an exit.
Who tells the public. Not the entity, on the directive's own words. Article 23(7) provides that where public awareness is necessary to prevent a significant incident or to deal with an ongoing one, or where disclosure is otherwise in the public interest, the Member State's CSIRT or competent authority — and where appropriate those of other Member States concerned — may, after consulting the entity concerned, inform the public about the incident or require the entity to do so. A discretionary power, and it names the authority as its holder (ELI presentation).
What a stranger can see. Article 23(9) is the only overview the directive builds. Each single point of contact submits to ENISA, every three months, a summary report on notifications received, with anonymised and aggregated data on significant incidents, incidents, cyber threats and near misses; ENISA in turn reports findings to the Cooperation Group and the CSIRTs network every six months. Article 23(10) routes information to competent authorities under Directive (EU) 2022/2557 for entities identified as critical entities.
A quarterly anonymised aggregate is a count. It cannot tell you whether any particular entity read Article 23(3) correctly on any particular day, and it was never meant to. Anyone working through the European Commission directory entry will find the rules and the intervals, and will not find the file that would test a single case.
The other delegation, still open. Three dated facts, kept apart.
On 26 May 2026 the NIS Cooperation Group — Member States, the Commission and ENISA — agreed common templates for incident reporting, and the Commission said that "as a next step, the European Commission plans to adopt these templates through an implementing act, making them mandatory for all Member States" (Commission announcement). The announcement does not state that the act has been adopted, and it does not publish the templates.
On 20 January 2026 the Commission transmitted a proposal, COM(2026) 13, which would add to Article 23 a requirement that, when adopting an implementing act under the first subparagraph, it include ransomware information: whether the entity detected a ransomware attack, the attack vector, and whether mitigation measures have been implemented (COM(2026) 13).
And the negative, bounded in the same breath: across the nine official documents I read for this piece, none dated later than 21 November 2026 prints an adopted first-subparagraph act on the type of information, the format and the procedure.
A divergence neither page reconciles. Article 23(11), second subparagraph, enumerates nine provider types. Trust service providers are not among them. Article 1 of Regulation (EU) 2024/2690 and its full official title both include them. Both texts are printed, both are dated, and neither page I read reconciles the two (consolidated text, catalogue record).
What this adds. BTW has published on the ENISA registry's scope against its access clause under the NIS2 proposal, and nothing on Article 23. The four existing articles on this Commission object read Sweden's 5,000-connection threshold, the Intelsat/SES clearance and sustainable-finance FAQs. My own reporting-chain work reads DORA Article 19, the Cyber Resilience Act's Article 14, GDPR Articles 33 and 34, CIRCIA, the Australian payment-anchored clock and 23 NYCRR 500.17(b). Every one of those asked where the intervals sit. This one asks which entities the only numeric significance test actually binds, and what remains where it does not.
What would change the answer. Two things. An act under the "may adopt" limb of Article 23(11), extending the numeric criteria to other essential and important entities. Or a first-subparagraph act that arrives with templates attached, which would bind the shape of every notification without touching the trigger. Until one of them lands, the sentence that starts Europe's 24-hour clock is written by the entity, in a document nobody is required to keep.
Member Briefing
Read the full trend briefing
Members read the full briefing with the source notes behind every finding.
Strategic Circle
A daily briefing in your language, built from changes we have verified.
Join for USD 20/monthCancel anytime.
Members also get the full research library, a daily briefing on the networks they follow and the private forum.






