Skip to main content
CYBERSECURITY

What is the Dutch Cybersecurity Act?

The Cyberbeveiligingswet is the Dutch implementation of the European NIS2 Directive. It imposes a duty of care, a reporting duty and responsibility for the supply chain on organisations in essential and important sectors. It has applied since 15 August 2026 and came with no transition period. That makes the party building your AI part of your own burden of proof.

In force since 15 August 2026 Supply chain and AI 8 min read

The Dutch Cybersecurity Act transposes Directive (EU) 2022/2555, known as NIS2, into Dutch law. Where its predecessor covered a small group of vital providers, this Act reaches a far broader set of organisations across eighteen sectors. The starting point is not a list of measures but a duty of care: you assess the risks to your own service provision, take appropriate measures, and must be able to demonstrate that you did so. Securing your supply chain is an explicit part of that duty, and that includes your ICT and AI vendors.

Last updated: 16 November 2026. The Act has been in force since 15 August 2026, together with its implementing decree.

15-08
in force since 2026, without delay
18
sectors covered by the Act
24h
for the first report after becoming aware
0
months of transition at introduction

Source: Cyberbeveiligingswet, Staatsblad 2026, 187, in force since 15 August 2026; sector scope from Annexes I and II of Directive (EU) 2022/2555; reporting deadlines from Article 23 of that Directive.

SCOPE

Who does the Act apply to?

The Act works with two categories, determined by sector and size. There is also a third group the Act does not name but does affect. Most AI builders sit in that third group.

Essential entity

The heaviest category: energy, drinking water, transport, banking, healthcare and digital infrastructure among others, from the larger company sizes upward. Supervision here is proactive, so the regulator can turn up without a specific trigger and ask for evidence.

Important entity

The same duty of care and the same reporting duty, but with reactive supervision: the regulator looks when there is a reason, such as an incident or a signal. This covers postal and courier services, waste management, food, manufacturing and digital providers.

Outside scope, inside the chain

Your vendors are often too small to fall under the Act themselves. That does not release you from managing their risk. The agency building your AI agent therefore usually sits outside the law, while accounting for it sits entirely with you.

Whether your organisation falls under the Act, and in which category, depends on sector, size and sometimes on a designation. That is a legal question, not the output of a formula.

SUPPLY CHAIN

Where the supply chain duty actually bites

The Act looks at you, not at your vendors. Your procurement and security file is therefore only as strong as the vendor you can least account for.

REGULATOR Your organisation essential or important entity duty of care, reporting, supply chain Cloud provider assessment on file Software vendor assessment on file Managed services assessment on file ? AI builder nothing on file The regulator addresses you, not your vendor. Your weakest file sets your position.
OBLIGATIONS

What the Act asks of you

Four blocks, of which the third and the fourth are the ones most often underestimated.

1

Duty of care: assess and document

You produce a risk assessment for your own service provision and take appropriate measures based on it. The Act prescribes no fixed list of controls, but it does name the subjects: risk management, incident handling, continuity and backups, access control, encryption, awareness and testing the effectiveness of what you put in place. Because you make the judgement yourself, the documentation is the evidence.

2

Reporting: 24 hours, 72 hours, one month

For a significant incident you send an early warning within 24 hours of becoming aware, a substantiated notification within 72 hours and a final report within a month. A notification to affected customers may be required as well. The clock starts when you become aware, not when it is clear what exactly happened.

3

Supply chain: your vendors are your risk

Supply chain security is an explicit part of the duty of care. You assess the quality of your vendors and record the arrangements, including the subcontractors they bring in. With AI solutions that is rarely a single party: there is a model provider, a hosting party and often a builder in between.

4

Management responsibility

The management body approves the measures, oversees their implementation and must follow training on the subject. That makes cybersecurity a board topic with its own agenda item rather than an IT topic. It is not the same as personal liability in the legal sense, and that difference is often overstated in the market.

Source: Directive (EU) 2022/2555, Articles 20 to 23, as implemented in the Cyberbeveiligingswet and its implementing decree.

AI IN THE CHAIN

Why AI projects are the weak point here

Not because AI is unsafe, but because AI projects grew up somewhere normal procurement control never passed by.

Started outside procurement

Many AI applications began as a pilot, on a department budget and without a vendor assessment. Then they simply stayed. In your records that vendor barely exists, while in practice they sit next to real data.

More links than it looks

Behind one AI feature there is often a model provider, a hosting party, a vector store and a builder. A vendor assessment naming only the builder does not cover the chain. The question per link is simple: where does the data sit and who can reach it.

Access that never closes

A pilot that quietly kept running usually still holds access keys, API tokens and an admin account. That is exactly the kind of open access the duty of care is about, and it is the easiest thing to clean up once it is mapped.

OUR POSITION

What we add to your file

We build AI solutions and supply the vendor file with them. You do not have to assess us afterwards, because the evidence is part of the delivery.

1

An accredited certificate you can verify yourself

W69 AI Consultancy is certified to ISO/IEC 27001:2022, certificate number IQ002141S0926WAC, issued on 14 September 2026 and valid until 13 September 2029. The scope is AI advisory and implementation consultancy. You can check the status independently in IAF CertSearch, no account needed. See the certification page.

2

A documented access arrangement

Who can reach which data, on what grounds, and how that access is closed once the engagement ends. Including the subcontractors we bring in and the region where data is held, so your own assessment does not stop at our front door.

3

An incident arrangement that fits your deadlines

You must be able to file a first report within 24 hours. That only works if your vendor informs you quickly enough, so our engagement terms state within what period you hear from us and what you receive when you do.

4

The AI Act in the same file

The same solution often touches two frameworks: the Cybersecurity Act and the AI Act. We deliver the documentation so it serves both, instead of researching the same thing twice. Read about the AI Act.

HONEST ABOUT THE LIMITS

What we do not claim

This subject attracts a lot of claims that do not survive a follow-up question. Four things you will not hear from us.

×

“Our software is NIS2 compliant”

It cannot be. The Act obliges organisations, not products. There is no mark that makes an AI solution NIS2 compliant. Anyone offering one is selling a claim the law does not recognise.

×

“ISO 27001 means you comply”

The Act designates no standard as proof of compliance. Our certificate makes your assessment of us demonstrable and considerably shorter. It does not replace that assessment, and it says nothing about your own duty of care.

×

“Your directors are personally liable”

The management body has its own duty of care and a training obligation, and supervision can reach board level. That is strict enough to take seriously. Selling it as personal liability goes further than the law does.

×

“We will tell you whether the Act applies to you”

That is a legal question, for your counsel or your regulator. We are also not certified to ISO 42001; that process is under way and it will appear on this site once it is done. What we do is make the technology and the documentation demonstrably sound.

FREQUENTLY ASKED QUESTIONS

Frequently asked questions about the Act

Yes. The Cyberbeveiligingswet entered into force on 15 August 2026, together with the implementing decree and the Critical Entities Resilience Act. There was no transition period, so the obligations applied from day one. The previous regime, the Wet beveiliging netwerk- en informatiesystemen, was repealed.

Usually not. Small agencies and independent builders tend to stay below the size thresholds. The duty of care for the supply chain still sits with you as the client, so you must be able to account for that vendor even though the law does not apply to them directly. That distinction is the one most often missed in AI projects.

No. The Act does not require a certificate from your vendors and does not designate any standard as proof of compliance. In practice procurement asks for one anyway, because an accredited certificate is the fastest way to substantiate a vendor assessment. A certificate does not replace your own assessment, it shortens it.

No. The Act places obligations on organisations, not on products. There is no mark that makes an AI solution or a piece of software NIS2 compliant. A vendor offering that is selling a claim the law does not recognise. What does exist is demonstrable control at the vendor, for example through ISO 27001, and a contract aligned with your reporting deadlines.

The NIS2 Directive uses three steps: an early warning within 24 hours of becoming aware, a substantiated incident notification within 72 hours and a final report within one month. You only meet those deadlines if your vendors are contractually required to inform you quickly enough, because the clock starts when you become aware and not when the cause is clear.

The management body must approve the security measures, oversee their implementation and follow training on the subject. Supervision can therefore reach board level. That is stricter than many organisations assume, and it is not the same as personal liability in the legal sense. What applies in your situation is a question for your legal counsel.

This page provides general information and is not legal advice. Last updated on 16 November 2026.

NEXT STEP

Start with the list, not with the policy document

In an hour we walk through the AI tools and builders sitting in your chain, and name the ones you cannot account for today. After that you know where the file is genuinely thin.

RELATED

Further reading

Home Services AI Scan Sectors WhatsApp