Dutch AI supervision takes shape as regulator establishes new directorate

Dutch AI supervision takes shape as the regulator establishes a dedicated directorate to coordinate oversight, prepare market surveillance and support responsible innovation.

The Dutch Data Protection Authority (AP) has established a new Directorate for AI and Algorithm Supervision and Coordination as the Netherlands prepares to implement the EU AI Act. The directorate brings the AP’s existing coordination of AI and algorithm oversight together with preparations for future market surveillance, the national AI regulatory sandbox and a joint AI Coordination Centre with the Radiocommunications Agency Netherlands (RDI).

The announcement gives organisations a clearer indication of how Dutch supervision is developing, but it does not mean that the AP already holds all AI Act market-surveillance powers. The authority is the proposed regulator for significant parts of the Dutch regime and will receive those powers only when the national AI Act Implementation Act enters into force. In the meantime, existing GDPR obligations continue to apply whenever AI systems process personal data.

The new directorate combines strategic risk analysis with technical market supervision.

The directorate is divided into two departments. The AI and Algorithm Risk Analysis and Coordination department will examine the effects of AI and algorithms on people, fundamental rights and society. Its work includes identifying emerging risks, conducting research and coordinating with other regulators and experts. The AP highlights impact assessments, algorithm registers and audit frameworks as areas of particular interest, including the testing of risk-selection tools for discrimination and accuracy.

The AI Market Surveillance department is preparing for the AP’s proposed enforcement role under the AI Act. It is building the technical expertise needed to investigate and test AI systems and will also provide information and support for responsible innovation. National authorities will cooperate with the European AI Office on AI safety, including the testing and evaluation of frontier models and systems.

Formal designation remains unfinished even as substantive AI rules take effect.

The Netherlands is still working on its national implementation legislation. The AP notes that the deadline for designating national competent authorities was 2 August 2025 and that discussions about its designation, powers and funding are continuing. This institutional delay does not pause the AI Act itself.

Some obligations and prohibitions already apply, including rules concerning prohibited AI practices and AI literacy. Transparency requirements for certain AI systems and AI-generated or manipulated content have also taken effect. From 2 December 2026, the AP says a further prohibition will apply to AI that virtually undresses people without their consent. Organisations should therefore distinguish between uncertainty about the final supervisory structure and the legal obligations that already govern their activities.

Existing GDPR supervision will shape the emerging AI enforcement approach.

The AP is not starting from a blank page. It has supervised the use of personal data in algorithms and AI systems for years under the GDPR and has coordinated Dutch oversight of AI and algorithms since 2023. That experience is likely to influence how the authority examines purpose limitation, data minimisation, transparency, automated decision-making, security and accountability in AI deployments.

For organisations, this makes an isolated AI Act compliance project less useful than an integrated governance approach. An AI system can fall outside a high-risk category under the AI Act and still present significant GDPR issues. Conversely, a system that does not process personal data may still require controls under the AI Act because of its purpose, capabilities or effects on fundamental rights.

Risk signals may determine where supervisory attention is directed.

The AP describes the two departments as mutually reinforcing. Risks identified through social analysis and coordination can help determine where market supervision should focus, while technical understanding of AI systems can improve the authority’s assessment of emerging harms.

The announcement identifies systems that influence, select or assess people as areas of concern. It also refers to risks affecting children and other vulnerable groups, manipulation and deception, and recent incidents in which AI agents gained unauthorised access to private and public online environments. These examples point towards scrutiny of both system design and the organisational safeguards surrounding access, autonomy and human intervention.

The planned regulatory sandbox will support supervised development.

The AP and RDI are developing an AI regulatory sandbox in which organisations can work with supervisors on innovative systems in a controlled environment and bring them into line with the AI Act. The sandbox is intended to support responsible innovation and provide practical clarity, rather than serving only as an enforcement mechanism.

Participation will not replace an organisation’s own compliance work. Businesses considering a sandbox should arrive with a defined use case, mapped data and model flows, an initial classification, identified risks and clear questions for the supervisors. Records created during development should be capable of supporting later technical documentation, risk management and accountability.

DPOs should connect AI inventories with privacy and fundamental rights assessments.

The developing supervisory model increases the value of a reliable AI inventory. Organisations should know which systems they provide, deploy or substantially modify, which teams use them and whether they influence decisions about individuals. The inventory should include embedded AI features in ordinary business software, not only systems procured as standalone AI products.

Where personal data are processed, the organisation should determine whether a DPIA is required and how it interacts with any wider fundamental rights assessment. The AP has already described the AI Act’s fundamental rights impact assessment as complementary to a DPIA. The two assessments have different scopes, but their evidence, stakeholder input and mitigation measures should be coordinated to avoid contradictory or incomplete records.

Governance should reflect shared responsibility across the organisation.

AI compliance cannot sit with the DPO alone. Product, procurement, security, legal, compliance, human resources and operational teams may each hold essential information about a system’s purpose, data, performance and real-world effects. The DPO should retain the independence required for data protection advice and monitoring while contributing to a broader governance structure with clear decision-making authority.

Organisations should also review supplier evidence critically. Documentation should explain the system’s intended purpose, training and input data where relevant, performance limitations, human oversight measures, logging, security and routes for incident escalation. Contractual assurances are useful only when they match the system as deployed and can be tested against actual use.

Businesses should prepare for supervision before institutional details are finalised.

The AP’s new directorate shows that supervisory capability is being built while the Dutch implementation framework is still under discussion. Waiting for the final allocation of powers would leave organisations with less time to correct undocumented AI use, weak classification or inconsistent controls.

A proportionate preparation programme should prioritise the systems capable of materially affecting people. For each one, the organisation should confirm its role, legal basis where personal data are involved, applicable AI Act category, transparency duties, human oversight and monitoring arrangements. Governance should also provide a route for employees and affected individuals to raise concerns and for the organisation to investigate unexpected outcomes.

Discover how Aphaia helps organisations prepare for coordinated AI supervision.

Aphaia helps organisations map AI use, align AI Act and GDPR responsibilities, assess privacy and fundamental rights risks and establish governance that can withstand regulatory scrutiny. Practical policies, DPIAs, supplier reviews and accountability records can help organisations demonstrate that responsible AI is built into procurement, development and deployment.

Discover how Aphaia can help you build a compliant and future-ready data protection framework, including with the performance of DPIAs. Our new DPO+ and Super DPO services now come with enhanced Superpowers, including AI governance, US state privacy laws expertise and digital services regulation support. Contact Aphaia today.

Share the Post:
Related Posts