AI Act · Netherlands
EU AI Act enforcement in the Netherlands
The draft Implementation Act names eight market surveillance authorities — the most decentralised model in the Union. The AP takes recruitment and lending.
Trackers list the RDI as the Dutch AI Act authority. That is correct and incomplete in a way that matters: the RDI is the central contact point, not the body most deployers will deal with.
On 20 April 2026 the Dutch government published its draft Implementation Act — the Uitvoeringswet AI-verordening — for internet consultation. Rather than build a new regulator, it distributed the Act across eight market surveillance authorities already embedded in their sectors, on the reasoning that AI risk shows up inside a domain rather than in the abstract.
The eight, and what each holds
| Authority | Share of the Act |
|---|---|
| AP — data protection authority | Prohibited practices, transparency obligations, and the majority of high-risk systems, including recruitment, education, lending and government decision-making |
| RDI — digital infrastructure inspectorate | Critical-infrastructure AI. Central contact point for the Act |
| ILT — environment and transport inspectorate | Critical-infrastructure AI |
| IGJ — health and youth care inspectorate | Product-related high-risk AI |
| NVWA — food and consumer product safety | Product-related high-risk AI |
| NLA — Netherlands Labour Authority | Product-related high-risk AI |
| AFM and DNB | Financial-sector AI |
AI used in the administration of justice sits outside this list and is supervised by judicial bodies, mirroring the Act's own treatment of the judiciary.
The distribution that catches people out. The Netherlands Labour Authority is in the list, so organisations assume it covers AI in hiring. It does not — its share is product-related high-risk systems. Algorithmic candidate screening belongs to the AP, together with lending and education. If you are preparing for a labour inspectorate, you are preparing for the wrong conversation.
Two coordinators, not one
Eight supervisors create an obvious failure mode: the same system reviewed twice on inconsistent reasoning, or a gap that each body assumes another covers. The draft addresses it by making coordination a named function rather than an assumption. The AP and the RDI jointly coordinate, under cooperation protocols set out in law, covering information sharing and the alignment of supervisory activity.
So the Dutch model splits the two roles that most member states fuse. The RDI is the address — the single point of contact the Act requires. The AP is the substantive centre of gravity. A deployer writing to the contact point about a hiring system will reach the right building and the wrong desk.
The draft also provides for a mandatory AI regulatory sandbox, in which participants can test systems without immediate enforcement action. The rules themselves are not waived inside it — it buys supervised time, not an exemption.
A supervisor that started three years early
One Dutch feature has no equivalent elsewhere in the Union, and it changes what a first meeting looks like.
Since 2023 the AP has run the Department for the Coordination of Algorithmic Oversight (DCA) as a distinct unit, funded from €1 million in 2023 rising to €3.6 million by 2026. It publishes the AI & Algorithmic Risks Report Netherlands twice a year, setting out where it believes algorithmic risk is accumulating.
The practical consequence: the Dutch supervisor has been forming and publishing views on algorithmic risk for three years before the high-risk obligations bind. Its positions are readable in advance. That is an unusual advantage for a deployer willing to read them, and an unusual exposure for one whose system contradicts a position the regulator has already put in print.
The register nobody else has
At algoritmes.overheid.nl, Dutch government organisations publish descriptions of the impactful algorithms they operate — 1,536 entries when read on 23 September 2026, from municipalities, water boards, provincial bodies and national agencies.
It is not an AI Act instrument and entries are self-declared, so it is neither complete nor audited. It still does something no other member state offers: it makes a public body's algorithmic estate readable rather than inferable. If you sell into, partner with or process on behalf of a Dutch public body, what it has declared about its own systems is a matter of record before the first conversation.
Where the Netherlands stands
| Item | Position |
|---|---|
| Article 70 designation | Not complete — deadline was 2 August 2025 |
| Instrument | Draft Implementation Act (Uitvoeringswet AI-verordening), 20 April 2026, in internet consultation |
| Supervisory model | Eight market surveillance authorities; judiciary supervised separately |
| Central contact point | RDI |
| Coordination | AP and RDI jointly, under statutory cooperation protocols |
| Enforcement powers | Administrative fines; corrective orders; suspension or prohibition of use; withdrawal from the market; public warnings |
| High-risk obligations apply | 2 December 2027 (stand-alone) · 2 August 2028 (embedded) |
| Prohibitions in force since | 2 February 2025 |
What a Dutch deployer should do now
- Map each use case to one of the eight before you write to anybody. Hiring, lending, education and public-sector decisions go to the AP. Medical devices reach the IGJ, machinery the NLA, financial systems the AFM or DNB. The RDI is the contact point, not the default supervisor.
- Read the AP's published risk reports. Twice-yearly, three years deep. No other member state gives you this much of your future supervisor's thinking before the obligations arrive.
- Check the register if any counterparty is a Dutch public body. What they have declared is public, and a discrepancy between the declaration and the system you are integrating with is worth finding early.
- Start the six-month log retention now. Article 26 requires deployers to retain system logs for at least six months. Logs not kept cannot be reconstructed, so the system needs to be running correctly by roughly mid-2027 to have a compliant record on day one.
- Decide who owns the cross-authority answer. With eight supervisors and two coordinators, the risk is not a wrong answer but two inconsistent ones given by different parts of your organisation. Name one internal owner of the Dutch position.
Questions
Who enforces the EU AI Act in the Netherlands?
Eight market surveillance authorities under the draft Implementation Act published on 20 April 2026, with AI in the administration of justice supervised separately by judicial bodies. The RDI is the central contact point. The AP holds the largest substantive share: prohibited practices, transparency obligations and the majority of high-risk systems, including recruitment, education, lending and government decision-making.
Why does the Netherlands have two coordinating bodies?
Because the model separates the procedural role from the substantive one. The RDI is the central contact point, but the AP and the RDI jointly coordinate supervision through legally mandated cooperation protocols covering information sharing and the alignment of supervisory activity. Eight authorities cannot be coordinated by a letterbox, so the coordination function was given weight of its own.
We use AI to screen candidates in the Netherlands. Who supervises us?
The AP, the Dutch data protection authority. Recruitment sits inside the share of high-risk systems assigned to it, alongside education, lending and government decision-making. Note that the Netherlands Labour Authority covers product-related high-risk AI rather than hiring decisions, so the employment regulator is not the one most deployers expect.
What is the Dutch Algorithm Register?
A public register at algoritmes.overheid.nl where Dutch government organisations publish descriptions of the impactful algorithms they operate. It listed 1,536 entries when read on 23 September 2026. It is a transparency instrument rather than an AI Act requirement, and publication is self-declared, but it makes the Netherlands the only member state where a deployer's algorithmic estate in the public sector can be looked up rather than inferred.
Has the Netherlands completed its Article 70 designation?
No. The Implementation Act was published for internet consultation on 20 April 2026 and had not completed passage. The Article 70 deadline was 2 August 2025. The obligations are unaffected: the AI Act is a Regulation with direct effect, so high-risk duties arrive on 2 December 2027 for stand-alone systems regardless of where the Dutch bill has reached.
Related
Agent Trust Cloud
We publish this because the deployer side of the AI Act is widely misread, and an eight-authority split invites the specific mistake of preparing for the wrong regulator. Agent Trust Cloud is building a control plane for AI agents — identity, policy, evidence. The platform is in development and is not sold as a subscription today; what is available now is a fixed-scope assessment of the agents and machine identities you already have.