AI liability and operator duties in the Netherlands: 2026 guide.

An operator running an AI agent in the Netherlands faces a decentralised supervision model anchored by the Autoriteit Persoonsgegevens, coordinated by the Rijksinspectie Digitale Infrastructuur, and distributed across about eight sectoral authorities. Three layers of obligation are already in force: the Article 5 prohibited practices ban (since 2 February 2025), the Article 4 AI literacy duty (since 2 February 2025), and the GPAI model obligations under Chapter V (since 2 August 2025). Article 50 transparency obligations bite from 2 August 2026. High-risk Annex III obligations have a binding deadline of 2 August 2026, with a proposed deferral to 2 December 2027 that is not yet adopted. A Dutch draft Implementation Act was published 20 April 2026 and is not yet law. The EU AI Act applies directly regardless.

The Dutch supervisory architecture: decentralised by design

When the EU AI Act's Article 70 required Member States to designate national competent authorities by 2 August 2025, the Netherlands made a choice that distinguishes it from most of its EU peers: rather than creating a dedicated AI agency or assigning the entire portfolio to a single body, the Dutch government chose a decentralised model that mirrors the existing sectoral supervisory landscape.

At the centre of that architecture sits the Autoriteit Persoonsgegevens. The AP, which has operated as the Dutch data protection authority since 2001, was formally designated as the market surveillance authority for the majority of EU AI Act categories. Its mandate under the AI Act covers the Article 5 prohibited-practice prohibitions, the bulk of the Annex III high-risk categories (with carve-outs for specialised domains), and the Article 50 transparency obligations. The AP's existing GDPR enforcement powers and its institutional knowledge of automated decision-making made it the natural anchor for AI supervision in the Netherlands.

Coordinating across the decentralised structure is the Rijksinspectie Digitale Infrastructuur. The RDI is the Dutch inspection body for the digital and telecommunications infrastructure domain, with established roles under the Radio Equipment Directive, the European Electronic Communications Code, and the Digital Services Act. In the AI Act context, the RDI acts as the coordinating authority that aligns the work of the sectoral supervisors, serves as the contact point for the European AI Board under Article 70(2), and handles AI systems that fall within product safety or infrastructure domains.

The eight sectoral authorities

Beyond the AP and RDI, about eight additional supervisors hold specific AI Act competence for their regulated domains. This allocation is set out in the draft Implementation Act of 20 April 2026 and tracks the Annex III risk categories to the regulators who already supervise those sectors.

In financial services, the Autoriteit Financiele Markten (AFM) supervises AI used in investment advice, securities trading, and consumer credit decisions. De Nederlandsche Bank (DNB) covers AI used in banking, insurance underwriting, and payment services. Both apply their existing supervisory frameworks to AI, and both work within the expectations set by the EIOPA Opinion on Artificial Intelligence governance and risk management (EIOPA-BoS-25-360, published 6 August 2025), which provides interpretative guidance placing AI within Solvency II, IDD, DORA, and GDPR. The AFM has been active in algorithm supervision since publishing its Supervision of AI report in 2021, and its oversight of robo-advisers and credit-decision models is well established.

In healthcare, the Nederlandse Zorgautoriteit (NZa) oversees AI used in care market regulation and financing decisions. The Inspectie Gezondheidszorg en Jeugd (IGJ) handles AI in medical devices, clinical decision support, and patient safety contexts. AI systems qualifying as software as a medical device face a dual regulatory track: the EU Medical Device Regulation pathway and the AI Act high-risk requirements under Annex III category 6. These often run in parallel through the IGJ.

In education, the Inspectie van het Onderwijs supervises AI systems used in educational assessment, student performance management, and access decisions. This category engages Annex III point 4, which covers education and vocational training AI systems.

The College voor de Rechten van de Mens has a monitoring and advisory role for AI systems that affect fundamental rights, particularly in employment contexts, and can issue non-binding opinions that carry significant institutional weight in Dutch public discourse on algorithmic discrimination.

The Autoriteit Consument en Markt (ACM) holds residual supervision for AI in consumer markets and competition contexts, including AI-driven pricing and recommendation systems in retail and platform markets.

The draft Implementation Act: what it does and where it stands

On 20 April 2026, the Dutch Ministry of Economic Affairs published the draft Uitvoeringswet Verordening Kunstmatige Intelligentie for public consultation. The consultation closed in mid-May 2026. The draft performs three functions.

First, it formally designates the AP, RDI, and the eight sectoral authorities in law, giving each body explicit statutory authority to exercise the market surveillance, enforcement, and notification functions assigned to them under the EU AI Act. This matters because while the EU AI Act applies directly as an EU regulation, the national authority designations and the specific penalty-setting provisions require national implementing measures to be fully operative.

Second, it sets out the Dutch penalties framework consistent with Article 99 of the EU AI Act. The Act sets maximum fines at EUR 35 million or 7 per cent of global annual turnover for breaches of Article 5 prohibited practices, EUR 15 million or 3 per cent for breaches of other obligations, and EUR 7.5 million or 1 per cent for supplying incorrect or misleading information. The draft Implementation Act assigns enforcement of these penalties to the designated competent authority for the relevant category.

Third, the draft clarifies the process for coordinating enforcement actions when a single AI deployment engages more than one sectoral supervisor. Given the breadth of the Dutch distribution of competences, cross-authority coordination is a foreseeable operational need. The draft establishes a coordination mechanism through the RDI.

As of June 2026, the draft had not been enacted. It must be reviewed by the Raad van State (Council of State), pass both chambers of parliament (the Tweede Kamer and the Eerste Kamer), and be published in the Staatsblad to become law. Operators should not wait for the Implementation Act before preparing their AI governance programmes. The EU AI Act applies directly, and the AP has stated that it considers itself already competent to receive reports and handle prohibited-practice complaints under Article 5.

What is already in force: the three active obligation layers

Article 5: prohibited AI practices (in force since 2 February 2025)

Article 5 of the EU AI Act prohibits a set of AI practices that the legislature deemed incompatible with fundamental rights. These prohibitions apply immediately and unconditionally to any operator deploying AI in the Netherlands, regardless of whether the draft Implementation Act is in force.

The prohibited practices include: deploying AI systems that use subliminal techniques operating below a person's perception to materially distort their behaviour in a way that causes or is likely to cause harm; exploiting vulnerabilities due to age, physical or mental disability, or social situation to materially distort behaviour; real-time remote biometric identification of natural persons in publicly accessible spaces for law enforcement purposes, subject to narrow judicial-authorisation exceptions; retrospective deployment of biometric identification databases for law enforcement; AI-based assessment of the risk of an individual committing a criminal offence solely on the basis of profiling; social scoring by public authorities; and AI systems that infer emotions in workplaces and educational institutions, except for safety or medical reasons.

Dutch operators should note that the AP has flagged algorithmic hiring, facial recognition in retail, and emotion inference in workplace monitoring as areas of specific supervisory interest. The AP published guidance on Algorithm Supervision in the context of the GDPR and has indicated it will use its AI Act authority to examine Article 5 compliance on a proactive basis.

Article 4: AI literacy (in force since 2 February 2025)

Under Article 4, both providers and deployers must take measures to ensure a sufficient level of AI literacy among their staff and others who operate AI systems on their behalf. The literacy requirement is calibrated to the role of the person and the nature of the AI system: a legal team approving an AI-generated contract needs different competencies than a warehouse worker interacting with an algorithmic task-allocation system.

The Digital Omnibus provisional agreement of 7 May 2026 proposes to soften the wording of Article 4 from a duty to ensure a specific level of literacy to a duty to support the development of AI literacy. This proposed softening has not been adopted. The obligation as enacted applies in full.

In the Dutch context, the AI literacy obligation intersects with existing training requirements under the Wet bescherming persoonsgegevens (GDPR transposition) and sector-specific requirements from the AFM and DNB on model governance. Operators who already have a GDPR training programme should map their AI literacy obligations to it rather than treating Article 4 as a separate compliance stream.

GPAI model obligations: Chapter V (in force since 2 August 2025)

From 2 August 2025, providers of general-purpose AI models became subject to the obligations in Chapter V of the EU AI Act, including documentation requirements, obligations to comply with copyright law, and transparency requirements for model capabilities and limitations. The European AI Office, which coordinates GPAI supervision at the EU level, began publishing its GPAI Code of Practice guidelines in the second half of 2025.

For deployers in the Netherlands, the practical consequence is that any AI system built on a GPAI model must be deployed in a way that is consistent with the model provider's usage policies and documentation. If the intended application diverges from the provider's described intended purposes, the deployer may be reclassified as a provider for that application and inherit the full provider obligation set. Dutch operators building AI agents on top of foundation models from non-EU providers should review the technical documentation those providers are required to make available and assess whether the specific use case aligns with the documented intended purpose.

Article 50: transparency obligations from 2 August 2026

Article 50 of the EU AI Act requires operators of chatbots and conversational AI systems to notify users that they are interacting with an AI system, unless this is obvious from the context. This obligation applies from 2 August 2026 and has not been deferred by the Digital Omnibus proposal.

For Dutch operators, Article 50 arrives in an environment where the AP has already taken enforcement actions on chatbot transparency under GDPR Article 13 (transparency at point of data collection). The new AI Act obligation reinforces and extends those requirements. A customer-service chatbot that already carries a GDPR transparency notice must now also carry an AI-identity disclosure that is distinct from the data-processing notice: the user must be informed they are talking to an AI, separately from being informed their data is being processed.

The synthetic-media provisions in Article 50 are particularly relevant for Dutch media, advertising, and financial-services operators. AI-generated audio, video, and text designed to resemble real persons or events must carry a machine-readable marking that identifies the content as AI-generated. A proposed grace period to 2 December 2026 applies for AI-generated content from systems already on the market before 2 August 2026, but this grace period is part of the Digital Omnibus proposal that has not yet been adopted into law.

High-risk AI obligations: the contested timeline

The most debated question for Dutch operators planning their AI compliance programmes concerns the high-risk Annex III obligations. The original EU AI Act requires compliance by 2 August 2026 for standalone Annex III systems (Article 6(2)) and by 2 August 2027 for AI embedded in regulated products covered by Annex I (Article 6(1)).

On 7 May 2026, the Council of the EU, the European Parliament, and the European Commission reached a provisional agreement in trilogue on the Digital Omnibus on AI package, which was proposed by the Commission in November 2025. The provisional agreement proposes to move the Annex III standalone deadline to 2 December 2027 and the Annex I embedded deadline to 2 August 2028. The agreement also replaced the Commission's originally proposed conditional stop-the-clock mechanism with these fixed dates.

The provisional agreement is not law. It has not been formally adopted by the Council and Parliament. It has not been published in the Official Journal. Until that publication occurs, the original 2 August 2026 and 2 August 2027 dates are the legally binding deadlines. Formal adoption is expected before 2 August 2026 but is not guaranteed.

Dutch operators who are in scope for Annex III high-risk compliance should prepare on the basis of the original deadline and treat the proposed deferral as a contingency. The Annex III categories most likely to affect Dutch businesses are: employment and workers management systems (Annex III point 4), access to private services including credit assessment (Annex III point 5), and fundamental rights-affecting classification systems used by public authorities (Annex III point 6). The high-risk obligations include maintaining a risk management system under Article 9, ensuring data governance under Article 10, producing technical documentation under Article 11, maintaining logs under Article 12, ensuring transparency to deployers under Article 13, enabling human oversight under Article 14, and achieving appropriate accuracy, robustness, and cybersecurity under Article 15.

Deployer obligations under Article 26: what Dutch businesses must implement

Article 26 of the EU AI Act sets out the obligations that apply to deployers of high-risk AI systems. These apply to any Dutch business that puts a high-risk AI system to use in a professional context, regardless of whether the AI system was developed in the Netherlands. The deployer does not need to have built the system. Using it is sufficient to engage the Article 26 obligations.

The core Article 26 obligations for Dutch deployers are as follows. They must use the system in accordance with the provider's instructions for use. They must assign a human oversight function and ensure that the persons carrying out that function have the competence, authority, and time to intervene if the system behaves unexpectedly. They must monitor the system's operation and log relevant events. They must notify the provider and the AP of any serious incidents as defined in Article 3(49) of the AI Act. They must comply with the registration obligations in the EU database for high-risk AI under Article 71. They must carry out a fundamental rights impact assessment (FRIA) before deploying certain Annex III systems, particularly those used by public authorities or in access-to-services contexts. They must provide affected individuals with information about AI-assisted decisions that affect them, under Article 86.

A deployer who modifies a high-risk AI system beyond what the provider's instructions for use contemplate, or who puts the system to a use outside its intended purpose, may be reclassified as a provider under Article 25 and face the full provider obligation set, including conformity assessment procedures.

The Algoritmeregister: a Dutch-specific obligation for public authorities

The Dutch government launched the Algoritmeregister (Algorithm Register) in 2022 as a transparency instrument for public sector AI use. Public authorities in the Netherlands are expected to publish entries for impactful algorithmic systems in the register, describing the system's purpose, the data it uses, the decisions it informs, and the oversight mechanisms in place.

The Algoritmeregister goes beyond the EU AI Act's registration requirements in two respects. It applies to a wider range of algorithmic systems than the EU AI Act's Annex III high-risk categories, including decision-support tools that do not meet the strict EU high-risk threshold. It also applies to systems currently in use by public authorities, not only to new deployments after 2 August 2026. Private sector operators working under government contracts to deploy AI systems used in public administration should understand whether their system falls within scope of the register, as disclosure requirements may flow down through procurement contracts.

The revised Product Liability Directive: what Dutch operators should prepare for

Directive (EU) 2024/2853 on product liability was published in the Official Journal on 18 November 2024 and entered into force on 8 December 2024. It must be transposed into Dutch national law by 9 December 2026. It applies to products placed on the market or put into service after 9 December 2026.

The revised Directive expressly includes software, including AI systems, within the definition of product for liability purposes. This means that from December 2026, the developer or provider of an AI agent that causes damage to a user or a third party may be subject to strict product liability in the Netherlands, without the need for the claimant to prove negligence. The Directive introduces rebuttable presumptions of defectiveness that are particularly significant for AI systems: where an AI system is technically complex and disclosure of evidence is in the hands of the defendant, courts may presume defectiveness unless the defendant disproves it.

The Directive also expands the categories of compensable damage to include data loss and medically recognised psychological harm, both of which are common consequences of AI agent failures in customer-facing and healthcare contexts.

Dutch operators who deploy AI agents for third parties or consumers should begin reviewing their indemnification provisions in commercial contracts and their liability caps in terms of service. Contracts that were drafted before the revised PLD was adopted may not adequately allocate the new strict-liability exposure.

AI liability and insurance in the Dutch market

The Dutch insurance market is one of the most developed in Europe, with strong penetration of professional indemnity, directors and officers, and cyber policies among large and mid-sized businesses. However, the standard forms of these policies were written before AI agents became common operational tools, and most carry exclusions or gaps that leave AI-generated errors uninsured.

The practical insurance issues Dutch operators face in 2026 are four. First, general liability and professional indemnity policies typically exclude or limit coverage for losses arising from the outputs of automated systems. Operators should review their policy wordings for exclusions referencing automated decision-making, algorithms, or software-generated outputs. Second, cyber policies in the Netherlands generally cover data breaches and ransomware but do not contain affirmative coverage for third-party losses caused by an AI agent that provides incorrect advice, executes an unauthorised transaction, or generates misleading content. Third, the revised Product Liability Directive exposure from December 2026 onward creates a new category of strict product liability that existing products liability policies, where they exist, may not have priced for AI systems. Fourth, the AP's dual enforcement role means that an AI incident may simultaneously produce a regulatory enforcement fine and a civil damages claim, each of which needs separate cover.

The emerging affirmative AI insurance market provides early reference points. The ElevenLabs AIUC-1 policy, announced in February 2026 and backed by Munich Re, is the first commercial AI agent insurance product structured specifically for AI agent operators. The Munich Re aiSure product provides performance or parametric cover settling on measurable AI performance data. Armilla, a Lloyd's coverholder managing general agent, offers AI liability and performance cover for AI-based systems. These products are available in certain markets, and Dutch operators should confirm availability and applicability with a specialist broker. The product-specific reviews at agentinsured.eu provide structured analysis of current carrier positions.

What operators in the Netherlands should do by category and timeline

The practical preparation sequence for a Dutch business deploying AI agents has four phases corresponding to the regulatory timeline.

Now, before 2 August 2026: confirm that no deployed AI system engages the Article 5 prohibited practices; document your AI literacy programme under Article 4; identify all AI systems built on GPAI models and confirm that the provider has met its Chapter V obligations; review your customer-facing AI systems for Article 50 compliance and prepare user-disclosure mechanisms; audit your current insurance policies for AI exclusions and gaps; and if you operate AI systems that may fall within Annex III high-risk categories, begin the self-classification exercise and prepare a preliminary risk management framework under Article 9.

By 2 August 2026 (Article 50 deadline, binding): deploy AI-identity disclosures on all chatbots and conversational AI systems; implement machine-readable markings on synthetic media; ensure that emotion-recognition and biometric-categorisation systems carry the required user notices.

By the high-risk compliance deadline, currently 2 August 2026 but proposed under the Omnibus for deferral to 2 December 2027 (treat the original date as binding unless the Omnibus is adopted): complete Annex III self-classification; implement the Article 9 risk management system; produce Article 11 technical documentation; configure logging under Article 12; deploy human oversight mechanisms under Article 14; register in the EU high-risk AI database under Article 71; complete the Article 27 fundamental rights impact assessment where required.

By 9 December 2026 (revised PLD transposition deadline): review and update commercial contracts and terms of service to reflect the incoming strict product liability exposure; engage your insurance broker on product liability cover for AI systems deployed for consumers or third parties.

Frequently asked questions