Regulation (EU) 2024/1689 · Artificial Intelligence Act ● Annex III: — days

NIS2, the Cyber Resilience Act and the AI Act: One Compliance Map

Three EU regimes now touch the same AI system: NIS2 secures the operator, the Cyber Resilience Act secures the product, the AI Act secures the decision. This map shows which rule fires when, and how one evidence set satisfies all three.

Renomira Team · 12 min read
Network switches densely cabled in a 19-inch rack with status lights
Photo: Dsimic / CC BY-SA 4.0 via Wikimedia Commons

Ask a mid-market CISO which EU regulation covers the AI-enabled access-control system they deployed last quarter and you will usually get two answers, both incomplete. The truth is that NIS2, the Cyber Resilience Act (CRA) and the AI Act can all apply to the same system at once — each regulating a different layer: NIS2 binds the operator of the system, the CRA binds the product that ships with embedded software, and the AI Act binds the decision-making behaviour of the model inside it.

The good news: once you see them as one stack, the evidence overlap is substantial. Most of the work you do for one regime counts for another — provided you set the programme up that way from the start.

What each regime actually regulates

Three different questions, three laws:

  • NIS2 asks: is this organisation an essential or important entity, and are its network and information systems secure? It binds operators — hospitals, energy and transport operators, digital providers, public administration, and their key suppliers — with risk-management measures, supply-chain security, and incident reporting duties enforceable against management.
  • The Cyber Resilience Act asks: does this product with digital elements ship securely, and are its vulnerabilities handled for its lifetime? It binds manufacturers placing products on the EU market — secure-by-design defaults, vulnerability handling, SBOMs, and reporting of actively exploited vulnerabilities.
  • The AI Act asks: what does this AI system do to people, and is that acceptable? It binds providers and deployers of AI systems — risk classification, prohibited practices, transparency, human oversight, logging and conformity.

A single AI product therefore has up to three legal personalities: the vendor who builds it (CRA manufacturer, AI Act provider), the customer who operates it (NIS2 entity if in scope, AI Act deployer), and the software supply chain both depend on.

The scope test in order

Work through these questions once per system and you will know which regime fires:

  1. Are we an NIS2 entity? Check your sector against the Annex I/II lists (essential vs important) and your size threshold — medium-sized enterprises (50+ staff, €10M turnover) are generally caught where the sector is. Sectoral national transposition decides the fine print; your national authority maintains the register.
  2. Is this a product with digital elements? If it has remote-data-processing capability — which most AI products do — the CRA applies to the product layer, regardless of whether you call it SaaS.
  3. Is this an AI system, and at what risk tier? Annex III classification decides whether you carry high-risk duties, or only transparency and prohibited-practice screening.
  4. Who is who? The same company can be a CRA manufacturer and an AI Act provider and an NIS2 operator across different products. Map roles per system, not per company.
Dimension NIS2 Cyber Resilience Act AI Act
Regulates The organisation The product The AI decision
Core trigger Sector + size (essential/important) Products with digital elements on EU market AI system placed on market / put into use
Signature duty Art.-level risk-management measures, 24h early-warning reporting Secure by default, vulnerability handling, SBOM Risk classification, Art. 50 transparency, Art. 26 deployer duties
First evidence artefact Asset + risk register Technical file + SBOM AI register + classification record
Primary owner in a mid-market firm CISO Product engineering DPO / compliance
Deadline phase-in In force since transposition (Oct 2024) Main obligations land around the end of 2027 Phased: prohibitions + literacy 2026, high-risk duties 2027–2029

(Timing for the CRA and the AI Act has been under active review — including the Commission’s Digital Omnibus discussion — so treat the phase-in column as planning guidance and confirm the current calendar before fixing contractual dates.)

One evidence set, three regimes

Here is the part that saves budget. The artefacts overlap far more than the legislation’s structure suggests:

  • Your asset inventory is your AI register. NIS2 demands an inventory of network and information systems; the AI Act demands a register of AI systems with role, purpose and risk tier. One table with the right columns serves both — add owner, hosting, data categories, and classification outcome.
  • Vulnerability handling is post-market monitoring. The CRA’s obligation to monitor, patch and report vulnerabilities maps almost one-to-one onto the AI Act’s provider duty to monitor a high-risk system after putting it into service, including substantial-modification triggers.
  • Incident clocks must be harmonised. NIS2 expects an early warning within 24 hours of becoming aware of a significant incident; the AI Act’s serious-incident clock for high-risk systems runs to notification within 15 days, with the exact sequencing defined by the reporting rules. One triage process feeding both clocks beats two on-call runbooks that disagree about what “aware” means.
  • Training and policy evidence doubles up. NIS2 management accountability, the AI Act’s Article 4 AI-literacy duty, and CRA secure-development training all resolve to the same records: who was trained, on what, when.

The failure mode to avoid: three spreadsheets maintained by three teams. Assign one owner per system who assembles a single evidence bundle, with a compliance matrix (regulation → requirement → artefact → owner → review date) as the index.

Where the regimes genuinely diverge

Some duties cannot be pooled, and pretending otherwise creates gaps:

  • Management liability under NIS2 is personal — authorities can hold management bodies responsible and impose temporary bans on directors. Nothing equivalent exists in the AI Act or CRA.
  • The AI Act’s prohibited-practices screen (Art. 5) and fundamental-rights duties have no analogue in cyber legislation — an impeccably secured, perfectly patched system can still be an unlawful one if it manipulates or scores people.
  • Transparency to people — Article 50 disclosure that someone is talking to an AI — is unique to the AI Act. Security documentation never satisfies it.
  • Notified-body involvement differs: the CRA routes high-assurance products through conformity assessment bodies, the AI Act through its own Article 43 route, and NIS2 uses national supervision with no product certification at all.

Sequencing the work across 2026–2027

A pragmatic order of operations for a mid-market team:

  • Now: build the combined asset + AI register; run the NIS2 scope test; run the AI Act classification screen (Art. 5 first, Annex III second); open one incident-reporting runbook with all three clocks on it.
  • Next: gap-assess product security against CRA requirements if you ship software — secure-by-default, vulnerability disclosure process, SBOM capability — because the end-2027 phase-in arrives faster than procurement cycles.
  • Then: high-risk AI Act readiness — documentation, conformity route, registration — for systems going live before the 2027 duties apply.
  • Continuously: feed one evidence bundle to all three audits; re-run the scope test whenever a new sector, product or acquisition changes the answer.

How to check your own position in three minutes

The free exposure scan asks twelve questions about your AI usage and governance and returns an exposure score, your ranked exposures with the exact article that fired, and your first three moves — mapped to the dates that apply. Confirm your email and the full report unlocks.

Frequently asked questions

Does NIS2 apply to us if we only use AI, not sell it?

Very possibly yes. NIS2 turns on your sector and size, not on whether AI is your product — a logistics company or hospital using AI systems is assessed as an operator of network and information systems. The AI layer then adds its own deployer duties on top; the two analyses run in parallel and should share one asset inventory.

Is the Cyber Resilience Act the same as the AI Act?

No. The CRA secures products with digital elements — vulnerability handling, secure-by-default configuration, SBOM, update mechanisms — for everything from smart cameras to software platforms. The AI Act regulates what AI systems are allowed to do and how they treat people. An AI product can be fully CRA-compliant and still breach Article 5 of the AI Act, or the reverse.

Which regime has the shortest incident deadline?

NIS2’s early-warning clock — measured in hours from awareness of a significant incident — is the tightest in practice. The AI Act’s serious-incident reporting for high-risk systems runs on a multi-day horizon (up to 15 days depending on severity). Build triage to the shortest clock and let the longer deadlines fall out of the same process.

Who in the company owns the combined programme?

Split by artefact, not by regulation: the CISO owns the incident runbook and risk measures, product engineering owns the SBOM and vulnerability process, and the DPO or compliance lead owns the AI register, classification and transparency duties. What you need above them is one person who owns the compliance matrix that ties the three together.

Do SMEs get any relief from this overlap?

Partial relief, not exemption. The AI Act contains targeted simplifications for smaller companies — registration concessions and regulatory-smartbox preferences among them — and NIS2’s size threshold excludes micro entities, but the CRA and the AI Act’s core duties follow the product and the system, not your headcount. Smaller mid-market firms win on scope, not on exemptions.

How do we avoid maintaining three compliance programmes?

You cannot merge the laws, but you can merge their evidence: one register, one incident process, one training record, one policy set, indexed by a compliance matrix that shows which requirement each artefact satisfies. Re-audits then become a matrix walkthrough instead of three separate projects — which is also the cheapest way to answer the customer questionnaires that cite all three regimes in one form.

Next step

Put this guide to work on your own estate.

Run the free exposure scan to see which of these articles actually apply to your company — 12 questions, four minutes, no card. Confirm your email and the full report unlocks.

  • Free forever plan
  • No credit card
  • EU-hosted · GDPR-native
Run a free scan Sign up free

Build a live AI register in minutes — upgrade only when you need the audit pack.

ai act nis2 overlap obligationsnis2 and ai act compliance togethercyber resilience act vs ai act differenceoverlapping eu tech regulations aiai act and nis2 mapping checklisteu cyber rules ai compliance alignmentcra ai act product security obligationswhich eu rules apply to our ai stack

Keep reading