# NIS2 in healthcare: what IT teams must instrument

> Operational NIS2 evidence for a healthcare IT team: Articles 21 and 23 of Directive (EU) 2022/2555, the 24 h / 72 h / one-month clocks, national CSIRT versus GDPR Art. 33, and why Annex I is not automatic essential-entity status.

Author: Ala Ben Aicha

Canonical page: https://alabenaicha.me/insights/nis2-healthcare-it-teams

Updated: 2026-09-19

## Direct answer

NIS2 is not a single certificate. Article 23 requires 24 h, 72 h, then one month for essential or important entities. Annex I healthcare is not automatic essential-entity status. CSIRT and authority follow national transposition.

## A directive, not the EHDS / MDR / GDPR stack

The [EU healthcare compliance](https://alabenaicha.me/insights/eu-healthcare-compliance-landscape) page maps **four regimes**. **This page isolates NIS2** for the team that must instrument detection, escalation and evidence. Pin the text: [Directive (EU) 2022/2555](https://eur-lex.europa.eu/eli/dir/2022/2555/oj) (NIS 2). Transposition deadline: 17 October 2024; national measures applicable from 18 October 2024.

**Health** is listed in **Annex I**. That does not make every vendor, hoster or clinic an **essential entity**. Entity type, size (thresholds from the Annex to Recommendation 2003/361/EC, applied via Art. 3), specific inclusions and the **national transposition law** all count. A CISO does not “declare” status in a README: it is an organisational decision, with legal counsel.

The [French-language sibling](https://alabenaicha.me/insights/nis2-healthcare-instrumentation) covers ANSSI and CERT-FR. Do not copy those names onto every member state. The national CSIRT and competent authority are **transposition-dependent**. Start from the Commission’s [NIS2 overview](https://digital-strategy.ec.europa.eu/en/policies/nis2-directive) and the member-state page that names *your* single point of contact.

## Incident clocks: NIS2 Art. 23 is not GDPR Art. 33

[Article 23](https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32022L2555) ([ELI](https://eur-lex.europa.eu/eli/dir/2022/2555/oj)) requires notification of a **significant incident** to the CSIRT or the competent authority. The **letters** below are those of Art. 23(4); the 24 h / 72 h / one-month clocks remain the default deadlines of (a), (b) and (d):

| Clock                                       | Deadline (directive)                                                                                                  | Recipient (directive)       | Minimum content                                                                                                                                    | What this is not                             |
| ------------------------------------------- | --------------------------------------------------------------------------------------------------------------------- | --------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------- |
| Early warning                               | **24 h** after becoming aware — Art. 23(4)**(a)**                                                                     | CSIRT / competent authority | Significant incident; suspected unlawful or malicious act?; cross-border impact?                                                                   | An internal Jira ticket, a Slack post-mortem |
| Incident notification                       | **72 h** after becoming aware — Art. 23(4)**(b)**                                                                     | same                        | Update of (a) plus initial assessment (severity, impact, IoCs if available)                                                                        | The **GDPR** notification                    |
| Intermediate report                         | **Upon request** of the CSIRT / authority — Art. 23(4)**(c)**                                                         | same                        | Relevant updates on the situation                                                                                                                  | A final report “sent early”                  |
| Final report                                | **One month after (b)** — Art. 23(4)**(d)**                                                                           | same                        | Detailed description (severity, impact); type of threat or likely root cause; mitigation applied and ongoing; cross-border impact where applicable | An ISO 27001 certificate; Art. 23(4)(c)      |
| If the incident is **still ongoing** at (d) | Progress report **at that time**, then a final report **one month after the incident is handled** — Art. 23(4)**(e)** | same                        | (e) does not merge with (c): (c) is an authority request; (e) is the “still open at (d)” case                                                      | Silence until “we will see on Monday”        |
| **GDPR Art. 33**                            | **72 h** (**separate** track)                                                                                         | Data-protection authority   | Personal-data breach at risk                                                                                                                       | A NIS2 substitute                            |

(c) is not the one-month report. (d) is the final report, **one month after the notification (b)**, not one month after becoming aware. (e) is not “on request if it lasts”: if the incident is still open at the (d) deadline, you file a progress report *instead of* the final (d) at that deadline, then a final within one month of handling.

The two 72-hour clocks can coincide on the calendar and **diverge on recipient and content**. NIS2 goes to the CSIRT / designated authority; GDPR Art. 33 goes to the data-protection authority. One email “we had an incident” satisfies neither.

Trust-service providers: **derogation from point (b)** of Art. 23(4) — notification in **24 h** (not 72 h) for a significant incident that affects the trust service. Do not copy that derogation onto a hospital DPI.

## Art. 21: ten families of measures, one audit trail per family

Article 21(2) lists **proportionate** measures (not a miracle product):

| Letter | Art. 21(2) family                                         | Engineering evidence (example)                                |
| ------ | --------------------------------------------------------- | ------------------------------------------------------------- |
| (a)    | Policies on risk analysis and information-system security | Versioned risk register, named owner                          |
| (b)    | Incident handling                                         | 24 h / 72 h / 1-month runbook + dated exercise                |
| (c)    | Continuity, backup, recovery, crisis                      | Restore exercise with measured RTO/RPO — not a PDF policy     |
| (d)    | Supply-chain security                                     | Supplier inventory + clauses + failover test                  |
| (e)    | Acquisition / development / maintenance, vulnerabilities  | SDLC, CVE handling, documented patch window                   |
| (f)    | Assessing effectiveness                                   | **Scheduled** audit / penetration test, not a marketing badge |
| (g)    | Cyber hygiene and training                                | Training evidence for the management body (Art. 20)           |
| (h)    | Cryptography                                              | TLS / key / HSM inventory, not “we have HTTPS”                |
| (i)    | HR, access control, assets                                | Dated access review, minimal CMDB                             |
| (j)    | MFA, secured communications                               | MFA on EHR admin **and** on the incident-escalation path      |

Article 20 charges the **management body**. A CISO who “signs NIS2” in Confluence is not Art. 20.

## Integration exercise (no real data)

In a test environment, cut **one** downstream system (HL7 feed, FHIR API, or Mirth destination). Time: detection, “significant?” classification, 24 h early-warning draft to an **internal** mailbox (not to a CSIRT), queue growth, recovery, reconciliation. The gap this exercise reveals — missing owner, no clock, no IoC — is the backlog item. Do not notify a fictitious incident to an authority.

## What this is not

This is not an essential-entity qualification, not an ENISA or national-authority opinion, not a substitute for legal counsel, not an ISO audit, and not clinical advice. I do not notify on your behalf. ANSSI / CERT-FR belong on the [French page](https://alabenaicha.me/insights/nis2-healthcare-instrumentation), not on every member-state runbook. [GDPR health-data architecture](https://alabenaicha.me/insights/gdpr-compliant-healthcare-data-architecture) remains a **separate** workstream.

If you need to instrument logs, queues and failover around care interfaces, that is [digital health interoperability](https://alabenaicha.me/services/digital-health-interoperability). For a bounded spike, use [contact with project intent](https://alabenaicha.me/contact?intent=project).
