The Netherlands publishes a royal decree that will put certain AI developers under the Vifo investment-screening notification duty from 1 January 2027

What the decree is, and when it starts

The decree amends the existing Besluit toepassingsbereik sensitieve technologie, which the explanatory note calls the Besluit sensitief. That decree rests on the Wet vifo and widens or narrows what counts as sensitive technology beyond the EU export-control lists the Act uses as its base.

Article I changes two annexes. In part A it adds rows to the table in annex 2, which the note describes as listing technologies designated as sensitive technology under article 3 of the Besluit sensitief. In part B it inserts rows in the table in annex 3 after reference codes 3E201, 5A004 and 5D002 and appends rows AST 5 to AST 10. The note says the annex 3 table marks the sensitive technologies that are also very sensitive. Article II says the decree takes effect on 1 January 2027. The decree rests on articles 4 and 8 of the Wet vifo.

A note on method. The HTML page rendered the annex tables as empty cells, so our first reading took the AI scope from the explanatory note. We then opened the XML rendering of the same Staatsblad item, which carries the annex 2 row text for AST 5 to AST 10, and the AI scope below is read from it. The annex 3 rows in the XML are bare codes (AST 5 to AST 10), with no text.

What row AST 6 says

The operative row is headed "Artificiële Intelligentie (AI) -technologieën als bedoeld in bijlage 1, derde lid, van Verordening (EU) 2026/1386" and continues that they are covered "voor gebruik als in dat lid nader gedefinieerd". The row therefore defines AI scope by cross-reference to the third paragraph of annex 1 of Regulation (EU) 2026/1386 on the screening of foreign investments in the Union. We did not open that regulation.

The note describes what that cross-reference picks up. It says the decree follows, one to one, the definition of AI technologies in the revised EU foreign-investment screening regulation, covering a technology or know-how connected to an automated system that operates with varying autonomy and may show adaptiveness after deployment, and that infers from input how to generate outputs such as predictions, content, recommendations or decisions. On the note, it is limited to general-purpose AI models as defined in Article 3, point 63 of the AI Act, or systems based on them, suited to developing space or defence applications, and to general-purpose AI models with systemic risk within the meaning of Article 51 of the AI Act, or systems based on them. That is the note's description of the EU text, not the row's own wording.

The note says that, if information useful for interpreting the definition becomes available during preparations to implement the revised EU regulation, the Bureau Toetsing Investeringen will actively publish it where necessary. It does not say when.

Identification, impersonation and tracking systems

The row's second paragraph adds AI systems outside the first. It reads "Voor zover vallend buiten voornoemde Artificiële Intelligentie (AI) -technologieën, omvat AST6 tevens AI-systemen", again as meant in annex 1, third paragraph, of the EU regulation, in two groups.

The first group is systems designed for or used for identification of persons, groups and objects or impersonation of persons, based on analysis of image, speech or biometric information, for three purposes: identifying or recognising persons and groups; evaluating speech communication or identifying persons remotely and biometrically; or impersonation, which the row glosses as deepfakes, meaning audio or image material that imitates persons or shows a simulated setting. The second group is systems designed for or used for identifying persons, groups or objects from geographic data and information from networked sensors, such as GPS, WiFi, cameras and 4G networks, to analyse movement, location, traffic or event data for tracking persons, groups or objects, or to assess persons on essential security aspects, "zoals het screenen van personen op veiligheidsrisico's".

The note gives its reasons: a national risk analysis, concern about interference with fundamental rights, scope for misuse, and effects on the safety and stability of society. It also reports that an anonymous consultation respondent thought identification of objects from imagery might be too broad, citing spotting production faults on a conveyor belt, and says the definition was then narrowed to identification of persons and groups for which object identification from imagery may be used. We have not reconciled that account with the final wording, in which both groups still name objects.

The notification trigger and what it costs

On the note's account, a planned acquisition in a target that integrates or develops AI systems as described, as part of its products and services, would have to be notified under the Wet vifo if it gives significant influence or increases it, or gives control over the target. The Dutch is plain: 'gemeld dienen te worden onder de Wet vifo'. The notification goes to the Minister of Economic Affairs and Climate.

The note adds that for very sensitive technology the duty arises earlier than for ordinary sensitive technology, where the duty first bites on an intention to acquire control. For very sensitive technology, significant influence is enough.

The decree itself gives no percentages. In the Wet vifo as consolidated on wetten.overheid.nl, article 4(1) lists thresholds of one tenth, one fifth and one quarter of the votes, and allows an order to designate categories of targets to which only some of those apply. We read that through a search extraction of the consolidated page, not the raw statute, and we did not check which categories the Besluit sensitief designates. Counsel should confirm.

The note estimates 125 extra cases a year, 25 of them complex, from the decree as a whole, which covers more technologies than AI. That figure is an administrative estimate in the note, not a count of AI deals.

The EU regulation in the background

The note ties the decree to Regulation (EU) 2026/1386 of 17 June 2026 on screening of foreign investments in the Union, and says that regulation applies from 17 January 2028 after an 18-month implementation period. The Dutch government says it is running a separate track to align national law with it, and that the Besluit sensitief will be involved in that process where needed.

The practical point from the note is sequencing. AI enters the Dutch regime a year before the EU regulation applies, and the note says the EU text does not prevent Member States from bringing in technologies beyond its minimum scope.

What we did not verify

What we opened: the full Staatsblad 2026, 356 page on officielebekendmakingen.nl, including the preamble, Articles I and II, and the explanatory note, and its XML rendering (stb-2026-356.xml), which shows the annex 2 row text for AST 5 to AST 10 and the annex 3 rows as bare codes. For the Wet vifo, we read a search extraction of article 4 from the consolidated text dated 1 September 2025 on wetten.overheid.nl.

What we did not open: Regulation (EU) 2026/1386 and its annex 1, on which AST 6 depends; the existing Besluit sensitief; the raw text of articles 4 and 8 of the Wet vifo; and the Council of State advice. We did not read the text of the annex 2 rows for AST 5 and AST 7 to AST 10 beyond confirming that they are there.

What we refuse to claim: the exact percentage that triggers notification for an AI target, the penalty for failing to notify, the length of any standstill, or whether a given company is inside the scope. The note's account of the general-purpose-model limits is the note's description of the EU text, not a reading of that text.

Related briefings

• Netherlands: decentralised AI Act implementation
• EU AI Office: enforcement against general-purpose model providers
• EU implementing regulation 2026/1755 on GPAI model access and evaluation
• Europe AI regulation hub
• AI Regulation News hub