State of compliance
The state of AI Act compliance in the Netherlands
Each month we measure the full Dutch algorithm register against five forms of evidence the AI Act expects of high-risk systems. The result is an index from 0 to 100 per round, using the same definitions every round, so the movement over time can be read directly.
Measured on: 2026-08-20 · Source: Algoritmeregister van de Nederlandse overheid
- index across all entries
- 71.9
- index for high-risk systems
- 67.3
- entries in the register
- 1536
- self-declared high-risk systems
- 41
First index round, no comparison with a previous measurement yet.
The finding of this round
Self-declared high-risk systems score 67.3, below the 71.9 of the register as a whole. In the category where the bar is highest, the register therefore shows less evidence. For 24 of the 41 high-risk entries no impact assessment is stated, and 26 of the 41 were not updated in the past twelve months.
The five indicators
Each indicator counts how many entries show the evidence concerned. The index is the unweighted average of these five shares, because any weighting is a judgement that makes the measurement harder to follow.
| Indicator | What we count | All entries | High-risk |
|---|---|---|---|
| Impact assessment stated | Does the entry state a completed impact assessment, such as a DPIA or IAMA? | 42.6%655 / 1536 | 41.5%17 / 41 |
| Legal basis stated | Is a legal basis for using the system stated? | 77.5%1190 / 1536 | 73.2%30 / 41 |
| Proportionality substantiated | Does the entry explain why the use is proportionate to its purpose? | 96.2%1478 / 1536 | 87.8%36 / 41 |
| Human oversight described | Does the entry describe how people intervene in the outcome? | 98.4%1511 / 1536 | 97.6%40 / 41 |
| Kept current | Was the entry updated in the twelve months before the measurement? | 45%691 / 1536 | 36.6%15 / 41 |
Organisations with the most entries
Willingness to publish is a signal in itself. Publishing a lot shows a willingness to be transparent; it says nothing about the quality of the underlying documentation.
| Organisation | Entries |
|---|---|
| Gemeente Amsterdam | 71 |
| Belastingdienst | 69 |
| Douane | 48 |
| Gemeente Utrecht | 45 |
| Gemeente Den Haag | 44 |
| Gemeente Rotterdam | 29 |
| Rijksdienst voor Identiteitsgegevens | 22 |
| Stichting Bureau Informatiediensten Nederland | 21 |
| Politie | 20 |
| Autoriteit Consument en Markt | 18 |
Earlier observations
2026-06-12
1462 entries, of which 41 high-risk; for 33 of those no impact assessment was stated.
Manual full download of the register. The August measurement reproduces these figures on every comparable field, so both rounds can be compared directly.
Method and limitations
The measurement retrieves the full register through the public interface of algoritmes.overheid.nl and counts per entry whether the fields concerned are filled in substantively. Fields holding only a dash, "to follow" or "not applicable" count as not filled, otherwise you measure form rather than substance. Current means: updated in the twelve months before the measurement date.
An important limitation: we measure the register, not the organisation. An empty field means the register shows no evidence. The impact assessment may well exist and simply not be stated. The register is also not a replacement for the EU database under article 71; it is a national, policy-based publication duty whose content runs ahead of what organisations will need to have demonstrably in order.
This index is also available in machine-readable form through the open API.