A technology company builds a generative AI system, trains it on licensed datasets, and deploys it commercially in the Netherlands. The system produces thousands of images, articles, and software modules each day. The company assumes it owns the output. Then a competitor copies the material wholesale – and when the company's Dutch counsel examines the position, the answer is far from clear.
AI-generated works occupy an unresolved space in Dutch intellectual property law. Dutch copyright legislation requires a human author as the originating creative force, meaning fully autonomous AI output does not automatically attract copyright protection. Businesses commercialising AI-generated content in the Netherlands must therefore structure their workflows, contracts, and IP claims carefully to avoid losing rights they assumed they held.
This analysis examines the doctrinal foundations, the gap between statute and practice, competing interpretations emerging from Dutch courts, cross-border implications under EU law, and the strategic choices available to businesses operating in this environment.
Doctrinal foundations: what Dutch copyright law requires
Dutch copyright legislation centres on a single foundational concept: the work must be an original creation reflecting the personal stamp of a human author. This requirement has been reinforced consistently by the Hoge Raad (Supreme Court of the Netherlands), which has held that originality in the Dutch sense demands creative choices made by a natural person. The work must carry the mark of individual personality.
When an AI system generates an image or a text, no natural person makes the specific creative choices embedded in that output. The developer writes code and defines training parameters. The operator configures prompts. The end user may enter instructions. But the selection, arrangement, and generation of the final expressive content is performed by the system itself. Dutch courts and most academic commentary treat this as an absence of human authorship at the point of creation – which, under current doctrine, means no copyright subsists in the output.
The position is not unique to the Netherlands. The Court of Justice of the European Union has grounded the EU copyright acquis in the same human-author principle, drawing on the fundamental rights dimension of authorship. Dutch law, as a civil law system with strong alignment to EU directives, follows this trajectory without deviation.
Dutch intellectual property legislation also covers related rights – neighbouring rights, database rights, and sui generis protections. The Databankenwet (Dutch Database Act), which implements the EU Database Directive, provides a separate protection track. A database producer who makes a substantial investment in obtaining, verifying, or presenting data may hold a database right even without originality in the copyright sense. For AI developers whose training datasets constitute structured, curated collections, this right may offer a partial alternative. However, the right protects the database itself, not AI-generated content derived from it.
A further consideration is the concept of geschriftenbescherming (protection for non-original writings), which historically gave thin protection to texts lacking the creativity threshold. Dutch legislation eliminated this category in 2015. Businesses that might have relied on this residual protection for AI outputs find no shelter there today.
The gap between statute and practice: where disputes are forming
The doctrinal position is relatively clear. The practical disputes are less so. Several distinct tension points are generating litigation and advisory complexity in the Netherlands.
The human contribution question arises in almost every commercial AI deployment. When a prompt engineer spends hours crafting detailed instructions to generate a specific image, does that human creative effort translate into authorship of the output? The Rechtbank (District Court) level has not yet delivered a definitive ruling on this precise question. Practitioners observe two lines of reasoning emerging. The first treats the prompt as a tool instruction – comparable to telling a printing press what to produce – with no creative ownership of the output attaching to the prompter. The second treats sufficiently detailed, iterative, and selective human guidance as constituting authorship, at least where the human exercises genuine creative control over the generative process.
Neither position is settled. The outcome turns on facts: the degree of human direction, the number of iterations, the selectivity of the final choice. Additionally. Whether the human can be said to have expressed a personal vision through the AI's operation. Businesses that document their prompting workflows systematically are in a stronger evidential position if the question reaches the Hoge Raad.
Training data and the reproduction question creates a second layer of complexity. Generating AI output typically requires training on large corpora of existing works. Under Dutch copyright legislation, reproducing a protected work without consent constitutes infringement. The text and data mining exceptions introduced by the EU Copyright in the Digital Single Market Directive have been transposed into Dutch law, but with conditions. Commercial AI training does not fall automatically within the research or non-commercial mining exception. Rights holders who have opted out of data mining for their content retain the ability to object. Dutch businesses building proprietary AI systems. typically structured as a BV (besloten vennootschap, private limited company) or an NV (naamloze vennootschap. Public company). must audit their training data provenance carefully before asserting ownership of derivative outputs.
For a tailored assessment of how Dutch AI regulation intersects with your IP strategy, contact us at our AI & Technology Law practice in the Netherlands or reach out to info@ferrazwhitmore.com.
Contractual ownership allocation is the third gap area. Dutch contract law, operating under the Burgerlijk Wetboek (Dutch Civil Code), gives parties substantial freedom to allocate IP ownership by agreement. AI platform terms of service routinely purport to assign output rights to users. But an assignment of a right that does not exist is legally incoherent. If Dutch law holds that no copyright subsists in purely AI-generated output, a contractual clause purporting to "assign all rights in generated content" cannot create rights that the law does not recognise. The recipient holds nothing – and may discover this only when enforcement is needed.
In practice, this means the contractual provisions that AI companies offer their business clients are worth less than they appear. A Dutch BV that builds a product on AI-generated content, registers at the KvK (Kamer van Koophandel. The Netherlands Chamber of Commerce). Additionally, launches commercially may find itself without enforceable IP when a competitor copies the core content assets.
Competing interpretations and the road to the Hoge Raad
Dutch courts have begun encountering AI-adjacent IP disputes, though most are resolved at first instance or settled before appellate review. Several recurring interpretive tensions are shaping the debate.
The first concerns the threshold of human creative contribution. Courts are applying a spectrum rather than a binary. A work produced entirely without human creative input sits clearly outside copyright. A work in which a human author uses AI as a tool – selecting from generated options, combining elements, adding original expression – may attract protection for the human-authored component. The difficulty is that most commercial AI workflows fall somewhere in the middle of this spectrum, and the threshold for "sufficient" human contribution is not yet defined with precision under Dutch law.
The second tension concerns algorithmic accountability and the question of who bears liability for infringing AI output. If an AI system reproduces substantial portions of a protected work in its output, the developer, the operator, and the user all have potential exposure. Dutch tort law and intellectual property legislation both support secondary liability theories. The question of which actor in the AI chain is primarily liable remains unresolved. The Rechtbank Amsterdam and the Rechtbank Den Haag – the two courts with the most concentrated IP caseload in the Netherlands – may reach divergent first-instance conclusions before the Hoge Raad provides harmonisation.
The third tension concerns software liability for AI-generated defective content. Where an AI system generates a product description, a legal summary, or a financial analysis that is materially wrong and causes commercial damage, Dutch product liability and tort law must determine who is responsible. The EU AI Act introduces a risk-based classification system that will reshape this analysis, but implementation is phased and national doctrine will continue to apply to conduct before the Act's obligations take full effect.
A client accustomed to common law precedent systems will find the Dutch civil law approach significantly different in one respect: courts apply statutory provisions to facts, and prior decisions carry persuasive rather than binding authority. This means that inconsistent first-instance decisions on AI authorship can coexist for years before the Hoge Raad resolves the contradiction. Businesses cannot rely on a single favourable District Court decision as establishing settled law.
Cross-border implications: EU law, the AI Act, and international strategy
The Netherlands sits at the centre of European digital commerce. Amsterdam hosts major technology infrastructure, and Dutch-registered entities are common holding and operating vehicles for cross-border AI businesses. The cross-border dimension therefore matters acutely.
EU copyright harmonisation is substantial but incomplete. The EU-level human authorship requirement means that an AI-generated work unprotected in the Netherlands will generally be unprotected across the EU. A Dutch business that fails to structure its AI content for copyright eligibility cannot simply claim protection by asserting rights in a more permissive jurisdiction. There is no EU equivalent of the UK's "computer-generated work" provision, which extends a limited copyright to the person who makes the arrangements necessary for a computer to generate the work. EU doctrine does not accommodate this shortcut.
The EU AI Act introduces a layered compliance obligation for systems deployed in the EU market. AI Act compliance requires providers and deployers of AI systems to classify their systems by risk level, maintain technical documentation, and – for high-risk systems – submit to conformity assessments. For AI systems that generate content used in regulated sectors (financial services, healthcare, legal advice), the AI Act obligations intersect directly with IP and liability questions. A Dutch technology company operating as a BV or NV and deploying AI-generated content in these sectors must address both the IP gap and the regulatory compliance obligation simultaneously.
Technology licensing structures are one strategic response. A business that cannot hold copyright in AI-generated output can still hold copyright in the software, the model architecture, and the training infrastructure – provided those involve sufficient human creativity. Licensing the software rather than purporting to assign output rights may create a more defensible commercial position. Dutch contract law supports detailed technology licensing arrangements, and Dutch notarial practice (involving a notaris) is available for agreements requiring formal authentication.
Cross-border enforcement presents further complexity. A Dutch business seeking to enforce IP rights in AI-generated content against a German or French competitor must confront the same human-authorship question under those jurisdictions' analogous laws. The harmonisation of EU copyright doctrine means the analysis is similar across member states, but not identical. Jurisdictions that have adopted broader interpretations of authorship – or where national courts have not yet engaged with the question – may offer different enforcement prospects. An integrated European IP strategy for AI content must map these divergences.
For businesses with significant IP portfolios across multiple EU jurisdictions, our analysis of intellectual property law in the Netherlands provides a framework for assessing protection strategies in the Dutch context specifically.
The digital services dimension adds another layer. Platforms distributing AI-generated content to Dutch or EU users must comply with the Digital Services Act, the Digital Markets Act, and sector-specific regulations. The intersection of these obligations with the IP uncertainty around AI output creates compounded compliance risk. A platform that cannot assert copyright in its AI-generated content may also face reduced ability to control downstream copying under platform liability rules.
To explore legal options for structuring AI-generated IP assets across European jurisdictions, schedule a consultation at info@ferrazwhitmore.com.
Strategic recommendations for businesses operating in this environment
The doctrinal uncertainty does not mean businesses are without options. Several strategies can meaningfully reduce exposure and improve the prospects of enforceability.
Document human creative involvement systematically. For every AI-generated work that the business intends to commercialise or enforce, maintain contemporaneous records of the human creative contribution. This includes prompt design, iterative refinement, selection from multiple outputs, and any post-generation editing. The stronger and more documented the human contribution, the better the prospects of satisfying the Dutch originality threshold. Version histories, prompt logs, and creative direction documents are all relevant evidence.
Audit training data provenance before asserting downstream rights. If the AI system was trained on third-party data without clear licensing or under a mining exception. The right to generate and commercialise output is itself at risk. A Dutch business – whether a BV or a larger NV – should conduct a training data audit as part of any serious IP due diligence. This is especially important in M&A contexts where AI assets are part of the deal value.
Structure contracts to reflect actual rights, not aspirational ones. AI platform agreements that purport to assign non-existent rights give false comfort. Dutch counsel should review all AI output agreements to ensure that what is being allocated is legally capable of being held and transferred. Where copyright cannot be established, contracts should address trade secret protection, confidentiality obligations, and database rights as alternative protective layers.
Integrate AI Act compliance from the outset. The EU AI Act's obligations are phased, but the classification of AI systems and the documentation requirements apply progressively. Dutch businesses deploying AI in commercial products should not treat AI Act compliance as a future project. The Act's risk-based approach to algorithmic accountability will increasingly determine who bears liability for AI output errors – and that determination affects IP, contract, and tort positions simultaneously.
Consider trade secret protection as a parallel track. Where copyright is unavailable, Dutch trade secret law – implementing the EU Trade Secrets Directive – may protect AI model weights, training methodologies, and proprietary datasets. Trade secret protection requires the business to take reasonable steps to maintain secrecy. For AI businesses whose competitive advantage lies in their models rather than their outputs, this may be the more durable protection strategy.
Monitor the Hoge Raad and the CJEU. The legal position will shift. The Hoge Raad's treatment of human authorship in AI-assisted works, and any CJEU preliminary reference on the same question, will reshape the analysis. Businesses should build legal monitoring into their IP management cycle and be prepared to adjust strategies as doctrine develops. Engaging a lawyer Netherlands practitioners with deep experience in both IP and technology regulation is essential for staying ahead of these shifts.
The comparative perspective is also worth holding in view. Our parallel analysis of AI-generated works and intellectual property in Portugal illustrates how the same EU doctrinal foundation plays out differently across civil law systems, offering useful reference points for businesses operating across European markets.
The Ferraz & Whitmore perspective: two traditions, one analysis
The AI-generated IP question sits at the intersection of two legal traditions that Ferraz & Whitmore knows well. From a civil law perspective, the Dutch doctrinal position reflects the deep-rooted European conception of authorship as an expression of human personality. Copyright is not primarily a commercial tool in this tradition – it is a reflection of creative individuality. The absence of human authorship is therefore not a procedural gap to be filled by contract. It is a foundational legal absence.
From a common law perspective, the contrast is instructive. English law's "computer-generated work" provision. which has no EU counterpart. reflects a more instrumentalist view: where human arrangement and effort produce a work. Some protection should attach, regardless of whether a natural person made each creative choice. The EU has not adopted this approach, and the Netherlands follows the EU line.
For international businesses accustomed to common law IP systems, this difference carries real commercial consequences. An AI content business that structures its rights strategy on common law assumptions will find those assumptions do not hold in the Netherlands or elsewhere in the EU. The gap between the legal systems is not merely academic – it determines who can enforce what, against whom, and with what remedies.
The practical consequence is that businesses operating across both legal traditions must maintain dual-track IP strategies: one that optimises for common law enforceability in English-speaking jurisdictions. Additionally. One that structures for civil law protection across the EU. These strategies can coexist, but they require deliberate design. Ferraz & Whitmore's dual-tradition position is directly relevant to this challenge.
Frequently asked questions
Q: Can a Dutch company claim copyright in content generated entirely by an AI system without human creative input?
A: Under current Dutch copyright law, no. Dutch copyright legislation requires that a work originates from a human author's personal creative choices. Where an AI system generates content autonomously, the human-authorship requirement is not satisfied, and no copyright subsists in the output. The company may hold rights in the underlying software or training data, but not in the AI-generated content itself.
Q: How long is it likely to take before Dutch courts establish a clear position on AI-generated IP?
A: A definitive position from the Hoge Raad could take several years to emerge. First-instance decisions from the Rechtbank Amsterdam and Rechtbank Den Haag are accumulating, but most AI-adjacent IP disputes are currently settled before reaching appellate review. A preliminary reference to the Court of Justice of the European Union could accelerate harmonisation across the EU, but that process itself typically takes two to three years from referral to judgment. Businesses should plan on operating under uncertainty for the medium term.
Q: Does the EU AI Act change the intellectual property position for AI-generated works in the Netherlands?
A: The EU AI Act is primarily a regulatory instrument governing AI system deployment, not an intellectual property law. It does not directly resolve the authorship question. However, its requirements for transparency, documentation, and algorithmic accountability create an evidentiary record that may become relevant in IP and liability disputes. Compliance with the AI Act may also indirectly support arguments about human creative control in AI workflows – a factor courts may weigh when assessing whether the human contribution threshold is met.
About Ferraz & Whitmore
Ferraz & Whitmore is an international law firm based in Lisbon, advising business clients across 46 jurisdictions. Our practice in AI and technology law covers IP structuring, AI Act compliance, algorithmic accountability, technology licensing, and digital services regulation across European and international markets. We work with technology companies, institutional investors, and in-house legal teams navigating the intersection of intellectual property law and emerging AI regulation. As an international law firm in the Netherlands and across the EU, we combine Portuguese civil law expertise with English common law tradition to deliver integrated advice for clients operating across multiple legal systems. The firm's IP and technology practice includes practitioners with experience before Dutch courts and EU regulatory bodies, supporting both enforcement strategies and proactive IP structuring for AI-driven businesses. To discuss how Dutch and EU intellectual property law applies to your AI operations, contact us at info@ferrazwhitmore.com.
Disclaimer: This publication is provided for informational purposes only and does not constitute legal advice. The information herein should not be relied upon as a substitute for professional legal counsel tailored to your specific circumstances. Ferraz & Whitmore assumes no liability for actions taken or not taken based on the contents of this material. For advice regarding your particular situation, please contact info@ferrazwhitmore.com.