Only if what the system does with personal data falls under one of the 17 categories, and the condition of that category is met. The list of the Dutch Data Protection Authority (Autoriteit Persoonsgegevens, AP) names no technology as such: it looks at the type of processing, and for most categories only when that processing is large scale or systematic.
Imagine you are the data protection officer (DPO) of a Dutch municipality. A project manager reports that the enforcement department wants to deploy an AI system, and asks: "Is this on the AP list? Then we know whether we need a DPIA." That is a good first question, but not the last one. If the processing is not on the list, a DPIA can still be required.
In short
Under Article 35(4) GDPR the AP adopted a list of 17 types of processing for which a data protection impact assessment (DPIA) is mandatory (Government Gazette 2019, 64418). Twelve categories apply to large scale processing and/or systematic monitoring, observing and influencing behaviour only to large scale and systematic processing, and for covert camera surveillance by employers the duty applies even in incidental cases. Health data apply only to large scale processing, profiling to a systematic and extensive evaluation of personal aspects, and two categories (blacklists and partnerships of public authorities) have no scale condition at all. If your processing is not on the list, you test it against Article 35(3) GDPR and then against the nine criteria of the European guidelines (WP248rev.01): with two or more criteria a DPIA is needed in most cases. If the system is an Annex III high-risk system under the AI Act, some deployers will also need a fundamental rights impact assessment (FRIA) from 2 December 2027, which may refer to the DPIA. If a system was placed on the market or put into service before that date, the duty applies only after a significant change in its design, and for systems intended to be used by public authorities by 2 August 2030 at the latest (Article 111(2)).
What the AP list is, and what it is not
The GDPR requires the controller to carry out a DPIA where processing is "likely to result in a high risk" to the rights and freedoms of natural persons, and to do so prior to the processing (Article 35(1) GDPR). Paragraph 3 names three cases where a DPIA is required in particular. Paragraph 4 has each supervisory authority establish its own list. The Dutch list is the decision of the AP of 19 November 2019. On its document page the AP says it plainly: you have to decide yourself whether you must carry out a DPIA, and the list can help you do so.
Three points are in the decision itself. First, the starting point is that for processing on the list you carry out a DPIA before you begin. Second, the list is not exhaustive: processing that is not on it can still, given its nature, scope, context and purposes, result in a high risk. Third, each category states which criteria from the European guidelines the AP took into account. Those guidelines are WP248rev.01 of the former Article 29 Working Party, which the European Data Protection Board (EDPB) endorsed on 25 May 2018 (Endorsement 1/2018).
The EDPB gave an opinion on the draft list in 2018 (Opinion 16/2018). It asked the AP, among other things, to add biometric data for unique identification, and to make location data subject to a DPIA only in combination with another criterion. According to the decision, that procedure (the consistency mechanism) led to the extra category of biometric data and to some textual changes. A summary of the decision is also on our page on the Dutch list of mandatory DPIAs.
The 17 categories and their conditions
The decision is published in Dutch; the descriptions below are our translation. Read the condition with each category. That is where "looks like it" becomes "is on it".
Large scale and/or systematic. These categories concern "large scale processing" and/or "systematic monitoring" (sometimes worded slightly differently, such as "large scale and/or systematic use"):
- 1. Covert investigation: investigation without informing the data subject in advance, for example by private investigators or for fraud prevention. For covert camera surveillance by an employer against theft or fraud by employees, the duty also applies in incidental cases, because of the unequal balance of power.
- 3. Fraud prevention, for example by social services or the fraud departments of insurers.
- 4. Credit scores: processing that leads to or uses assessments of creditworthiness.
- 5. Financial situation: financial data from which income, assets or spending patterns can be derived.
- 6. Genetic personal data.
- 9. Camera surveillance: monitoring of publicly accessible areas with cameras, webcams or drones.
- 10. Flexible camera surveillance: cameras on clothing or helmets, dashcams of emergency services.
- 11. Employee monitoring: monitoring of employees' activities, such as email, internet use and GPS in company vehicles.
- 12. Location data of or traceable to natural persons, for example by (scan) cars, navigation systems or phones.
- 13. Communication data, including metadata, except where necessary to protect the integrity and security of the network, the service or the end user's terminal equipment.
- 14. Internet of things: data generated by connected devices, such as smart meters and smart cities.
- 16. Observing and influencing behaviour: here both conditions apply together, large scale processing in which behaviour is systematically observed or influenced, including online behavioural advertising.
- 17. Biometric data for the purpose of identifying a natural person.
Large scale only. Category 7, health data, covers large scale processing, for example by healthcare institutions, occupational health services, reintegration companies, educational institutions, insurers and research institutes. Individual physicians and healthcare professionals are excluded on the basis of Recital 91 GDPR. That recital also names the lawyer, next to the physician and other healthcare professional.
A condition of its own. Category 15, profiling, requires a "systematic and extensive evaluation of personal aspects" based on automated processing, such as performance at work, pupils' performance, economic situation, health, preferences, reliability or behaviour. It has no scale condition. Note the difference with Article 35(3)(a) GDPR: that case also requires decisions based on the evaluation with legal effects or a similarly significant effect. Category 15 does not.
No scale condition. Two categories apply regardless of size:
- Blacklists (category 2): data on criminal convictions and offences, unlawful or disruptive behaviour or poor payment behaviour that is processed and shared with third parties, such as warning lists of insurers, retailers and telecom providers.
- Partnerships (category 8): municipalities or other public authorities that exchange special category or sensitive personal data with public or private parties, such as data on health, problem debt, unemployment or youth care, for example in neighbourhood teams and safety houses.
Why AI is never on the list "as AI"
The words artificial intelligence do not appear in the decision. It describes processing: what happens to which data about whom, at what scale and for what purpose. Two questions therefore decide the outcome. Which category fits what the system does? And is the condition of that category met?
"Large scale" and "systematic" are not defined in the GDPR. The guidelines give guidance. For large scale you look at the number of data subjects (as a number or as a proportion of the population), the volume and range of data, the duration and the geographical extent. Systematic means, according to the guidelines: occurring according to a system, pre-arranged, organised or methodical, part of a general plan or carried out as part of a strategy (WP248rev.01, pages 9 and 10). An AI system that runs automatically every day meets that quickly.
Example 1: the municipality with cameras, a scan car and a risk model
Take a fictional municipality with three applications.
Smart cameras in the town centre. The municipality installs cameras with image analysis on the market square to measure crowding and flag groups forming. That falls under category 9: monitoring of a publicly accessible area with cameras. The cameras run day and night according to a fixed plan, so systematically. The processing is on the list. If the monitoring is also large scale, Article 35(3)(c) GDPR applies as well: systematic monitoring of a publicly accessible area on a large scale.
A scan car with number plate recognition. For parking enforcement a scan car drives through neighbourhoods, reads number plates and records them with a place and time. That is location data traceable to persons, and the list names "(scan) cars" as an example in category 12. The car drives a fixed route every day: systematic. The guidelines also give a camera system that monitors driving behaviour on highways and automatically recognises number plates as an example where a DPIA is likely to be required. Where municipal staff wear cameras, such as smart glasses or body cams, category 10 (flexible camera surveillance) can apply, with large scale and/or systematic use; we wrote about that in smart glasses at the municipality.
A risk model for social assistance reviews. The model gives every social assistance file a score and selects files for review. Two categories fit. Category 3, fraud prevention, names "fraud prevention by social services" explicitly, and a model that scores every file is systematic and usually also large scale. Category 15, profiling, is also met: the model systematically evaluates people's reliability. If the municipality shares the results with other parties in a partnership, and the data is sensitive, such as problem debt, category 8 applies too, without a scale condition. You can read more about this example in AI fraud signals in social assistance.
The AI Act side. The cameras and the scan car are not automatically high-risk under the AI Act. Annex III point 1 covers remote biometric identification, biometric categorisation and emotion recognition; reading a number plate or counting crowds does not belong there, as long as the system does not recognise or categorise people. The social assistance model does, in our reading, fall under Annex III point 5(a): systems used by public authorities to evaluate eligibility for public assistance benefits, or to reduce, revoke or reclaim them. The municipality as deployer must then use the provider's information for its DPIA (Article 26(9)) and, as a body governed by public law, carry out a FRIA (Article 27(1)). Because the model profiles people, the exception in Article 6(3) is not available. Those duties apply from 2 December 2027. If the system was already put into service before that date, they apply only after a significant change in its design, and for systems intended to be used by public authorities at the latest from 2 August 2030 (Article 111(2)). The DPIA applies today.
Example 2: the bank and the insurer
A bank that assesses credit applications with a model. The model scores every application for consumer credit. That is category 4, credit scores, and a bank that assesses all its applicants this way processes on a large scale and systematically. Category 15 is met as well. If the bank rejects applications on the basis of the score, Article 35(3)(a) GDPR will as a rule also apply. Under the AI Act a system to evaluate the creditworthiness of natural persons is high-risk (Annex III point 5(b)). The bank as deployer must then also carry out a FRIA, even as a private party (Article 27(1)), from 2 December 2027.
An insurer that screens claims for fraud. The model gives every claim a fraud score. Category 3 names "fraud departments of insurers" explicitly. If, after established fraud, the insurer places people in a warning register shared with other insurers, category 2, blacklists, applies. Scale does not count there: even a small register shared with third parties is covered. Under the AI Act, fraud detection on claims is in our reading not high-risk as such. Point 5(c) covers risk assessment and pricing in life and health insurance, and for creditworthiness (point 5(b)) systems used to detect financial fraud are expressly excluded. The DPIA then remains the main test.
Example 3: the employer and the school
An employer that measures customer service performance with AI. The system analyses every call and gives staff a quality score. That is category 11, employee monitoring: systematic monitoring of employees' activities. Even at a small employer this can be systematic. In its opinion on the Dutch list, the EDPB pointed to the two criteria that meet here: vulnerable data subjects (employees) and systematic monitoring. The score itself is profiling (category 15). Under the AI Act a system to monitor and evaluate performance and behaviour at work is high-risk (Annex III point 4(b)), from 2 December 2027. A private employer that does not provide a public service does not need a FRIA for this; a public sector employer, as a body governed by public law, does (Article 27(1)). If the system was already put into service before 2 December 2027, those duties apply only after a significant change in its design, and for systems intended to be used by public authorities by 2 August 2030 at the latest (Article 111(2)). If the system also infers emotions from employees' voices, that has been prohibited since 2 February 2025, except for medical or safety reasons (Article 5(1)(f)).
A school that proctors online exams. A proctoring tool watches through the webcam and flags suspicious behaviour. If the tool recognises students by their face to establish who is taking the exam, it processes biometric data to identify someone (category 17), with large scale and/or systematic use. Note: processing biometric data to uniquely identify someone is prohibited in principle (Article 9(1) GDPR) and is allowed only where an exception applies, such as Article 29 of the Dutch GDPR Implementation Act (UAVG) for authentication or security. A DPIA does not make a prohibited processing lawful. If the tool infers emotions, the prohibition of Article 5(1)(f) AI Act also applies in education. Observing behaviour falls under category 16, but there large scale and systematic both apply; at a small institution, large scale may be missing. You then move on to the steps below. Under the AI Act, monitoring and detecting prohibited behaviour of students during tests is high-risk (Annex III point 3(d)). Whether the school must also carry out a FRIA depends on whether it is a body governed by public law or a private entity providing public services (Article 27(1)).
Not on the list? You are not done yet
Processing that is not on the list does not mean you need no DPIA. The decision says so itself, and the EDPB wrote in its opinion that every list further specifies Article 35(1) GDPR, which prevails in any case, so that no list can be exhaustive. You then move on to Article 35(3) GDPR and the nine criteria of the guidelines:
- evaluation or scoring, including profiling and predicting;
- automated decision making with legal or similar significant effect;
- systematic monitoring;
- sensitive data or data of a highly personal nature;
- data processed on a large scale;
- matching or combining datasets;
- data concerning vulnerable data subjects, such as children, employees and patients;
- innovative use or applying new technological or organisational solutions;
- processing that prevents data subjects from exercising a right or using a service or a contract.
According to the guidelines, in most cases a controller can consider that processing meeting two criteria requires a DPIA, and in some cases one criterion is enough. For many AI systems two quickly come into view, for example a score about people (criterion 1) with a new technology (criterion 8), though you must substantiate this per application. How to set up such a DPIA is in our step by step guide to the DPIA for AI. How the DPIA and the FRIA relate is in DPIA and FRIA side by side.
DPIA and FRIA: one file, two questions
The AI Act does not replace the DPIA. The deployer of a high-risk system uses the information the provider must supply (Article 13) to carry out its DPIA (Article 26(9)). If it must also carry out a FRIA, it may refer in the FRIA to the parts of the DPIA that already meet an obligation, or include those parts (Article 27(4)). The FRIA applies only to Annex III high-risk systems (Article 6(2)), and then only to bodies governed by public law, private entities providing public services, and deployers of systems for creditworthiness and for life and health insurance (Annex III point 5(b) and (c)). Systems for critical infrastructure (Annex III point 2) are excluded. The high-risk duties for Annex III and the FRIA apply from 2 December 2027, following the amendment by Regulation (EU) 2026/1744. For a system already placed on the market or put into service before that date, they apply only after a significant change, and for systems intended to be used by public authorities at the latest from 2 August 2030 (Article 111(2)).
The test in three steps
- Is the processing on the AP list, and is the condition met? Find the category that fits what the system does with which data, and read the condition: large scale and/or systematic, large scale only, a systematic and extensive evaluation, or no condition. If the answer is yes, you carry out a DPIA before you start.
- If not: does the processing fall under Article 35(3) GDPR? Those are profiling with decisions that have legal or similarly significant effects, large scale processing of special category or criminal data, and systematic monitoring of a publicly accessible area on a large scale.
- If not: does the processing meet two or more criteria of WP248rev.01? Then a DPIA is needed in most cases. If it meets one or more and you still do not carry out a DPIA, record why, including the DPO's view; where in doubt, the guidelines recommend carrying out the DPIA anyway. Where a DPO has been designated, you must seek the DPO's advice when carrying out a DPIA (Article 35(2) GDPR).
Want to run this for your own application? With the case check you describe the application in a few questions and see which AP categories, which criteria and which AI Act duties come into view.
Frequently asked questions about the Dutch DPIA list and AI
Sources
Newsletter
Every Tuesday, the AI Act week ahead in 5 minutes
A practical briefing on deadlines, new guidance and enforcement, so you know what matters this week. No spam and you can unsubscribe in one click.
Practical and short · No spam · One-click unsubscribe