A Brussels-based media company commissions an AI system to produce a series of commercial illustrations. The output arrives within seconds. The human team approves it without alteration and publishes. A competitor then reproduces the images. When the company's lawyers assess the options, a difficult question surfaces: does Belgian intellectual property legislation protect those images at all, and if so, who owns that protection? The answer is far from settled – and the commercial stakes are rising sharply.
Belgian intellectual property legislation conditions copyright protection on a work reflecting the personal intellectual creation of a human author. AI-generated outputs that lack sufficient evidence of human creative choice do not qualify for copyright under the current Belgian legal regime. The EU AI Act imposes conformity and transparency obligations on AI systems but does not resolve the ownership question – that remains an open doctrinal problem for Belgian courts and legislators.
This analysis examines the doctrinal foundations, competing interpretive positions, the gap between statute and practice. Cross-border implications for international clients operating in or through Belgium. Additionally, the strategic steps businesses should take now to protect their positions.
Doctrinal foundations: what Belgian IP law requires
Belgian intellectual property legislation has historically adopted the author's rights tradition common to continental civil law systems. Protection attaches to a work when it bears the imprint of the author's personality – expressed through free and creative choices in its conception or execution.
This requirement originates in the civil law doctrine of droit d'auteur (author's right), which differs meaningfully from the utilitarian copyright tradition familiar to common law practitioners. Under the common law approach, economic considerations and employer–employee relations can determine initial ownership without demanding evidence of individual creativity. Belgian law, by contrast, treats the human personality criterion as constitutive: without it, no protectable work exists.
The Court of Justice of the EU has reinforced this position in successive rulings on the interpretation of EU copyright directives. Those directives, transposed into Belgian law, require that a work be the author's own intellectual creation. Belgian courts apply this standard strictly. An AI system cannot exercise free creative choices in the legally relevant sense. It executes probabilistic outputs based on training data. The absence of a human decision-making subject at the moment of creation is the central doctrinal obstacle.
Three scenarios arise in practice, each with a different outcome under this framework.
In the first scenario, a human author uses an AI tool as a sophisticated instrument – selecting parameters, iterating outputs, and making substantive editorial choices that shape the final result. Belgian courts are likely to treat this analogously to a photographer selecting composition, lighting, and timing: the tool is instrumental, but the creative choices are human. Protection is available, and the human contributor is the author.
In the second scenario, a human provides a high-level prompt and accepts the AI's first output without meaningful curation. The human contribution is closer to commissioning than creating. Courts in Belgium have not yet definitively addressed this intermediate case, but the dominant scholarly view is that protection is precarious. The closer the process is to automated generation, the weaker the claim.
In the third scenario, an AI system generates content entirely autonomously – no specific human prompt, no editorial selection, no downstream curation. This output falls outside Belgian copyright protection entirely. No neighbouring right fills the gap. The content enters the public domain immediately upon creation.
Practitioners advising international businesses in Belgium note that this tripartite structure, while analytically useful, creates real documentation problems. Companies often cannot reconstruct, after the fact, exactly how much human creative input surrounded a given output. That evidentiary gap is itself a commercial risk.
Competing interpretations and the gap between statute and practice
The statutory position in Belgium is more settled than it may appear. The uncertainty lies not in the letter of intellectual property legislation but in its application to novel fact patterns – and in the divergence between what businesses assume and what courts actually decide.
One contested question concerns the treatment of AI-assisted works where the creative contribution is distributed across multiple human actors. A content platform might employ a prompt engineer, a creative director, and a post-production editor. Each contributes differently to a single AI-mediated output. Belgian law on co-authorship requires that each contributor independently satisfy the personal creation criterion. A contributor whose role is purely technical – running the AI system, adjusting parameters without creative intent – does not qualify as a co-author, even if their technical input was indispensable.
A second area of dispute concerns software liability and the ownership of AI models themselves. Belgian technology licensing agreements frequently attempt to assign all outputs to the licensee. The enforceability of such assignments depends on whether the outputs qualify as protected works in the first place. If they do not, the assignment clause purports to transfer something that does not legally exist as property. Businesses building licensing structures around AI-generated content should audit their agreements with this in mind.
A third interpretive problem arises in the context of training data. AI systems trained on third-party works may embed protected expression in their outputs. Belgian intellectual property legislation, as supplemented by EU directives on text and data mining, provides limited exceptions for certain research and data mining activities. Those exceptions are narrow. A commercial AI product trained on copyrighted Belgian works without authorisation exposes its operator to infringement claims from rights holders whose works formed part of the training corpus.
For international businesses, the gap between statute and practice is perhaps most visible in enforcement. Belgian courts have access to a well-developed civil procedure system. Interim injunctions are available from the president of the Enterprise Court and can be obtained with relative speed where urgency is established. However, the merits question – whether an AI-generated work is protected, and who owns it – requires a full analysis that most businesses have not yet undertaken for their AI-generated asset portfolios.
For a broader view of how Belgian intellectual property legislation intersects with technology licensing and digital services disputes, see our dedicated practice page on intellectual property law in Belgium.
To discuss how Belgian IP doctrine applies to your AI-generated content portfolio, reach out to info@ferrazwhitmore.com for a tailored strategy session.
The EU AI Act and algorithmic accountability in the Belgian context
The EU AI Act, which entered into force in stages beginning in 2024, introduces a risk-classification system for AI applications. It imposes obligations around algorithmic accountability, transparency, conformity assessment, and human oversight. Belgian businesses deploying AI systems – whether as providers, deployers, or users – must assess where their systems fall within the Act's tiered structure.
The Act's direct relevance to intellectual property is frequently overstated. It does not grant rights in AI outputs. It does not designate an owner for AI-generated works. Its transparency obligations do, however, create an indirect connection to IP strategy that sophisticated operators are beginning to exploit.
Under the Act's transparency requirements, providers of certain AI systems must disclose that content is AI-generated and must maintain documentation of training data sources. That documentation obligation, while designed for consumer protection and algorithmic accountability, simultaneously creates an evidentiary record that is highly relevant to IP disputes. A business that maintains thorough AI Act compliance documentation – logging human decisions made during AI-assisted creative processes – builds a stronger foundation for copyright claims than one that treats AI outputs as fully automated.
The Autorité de protection des données (Belgian Data Protection Authority) and Belgian sector regulators are expected to play an increasing role in AI Act enforcement as national competent authority structures are established. The overlap between data protection legislation, AI Act compliance obligations, and intellectual property legislation creates a compliance matrix that international businesses operating digital services from or into Belgium must address coherently.
Software liability is a related dimension. Under Belgian civil and commercial legislation, a party that deploys an AI system causing harm – including infringement of third-party IP rights – may face liability both in contract and in tort. The EU Product Liability Directive's modernisation, which covers software and AI systems, is expected to sharpen this exposure further. Businesses should review their technology licensing agreements to ensure that indemnification provisions adequately reflect the AI-specific risk profile.
One practical implication is that AI Act compliance documentation serves a dual purpose: it satisfies regulatory requirements and simultaneously supports IP ownership claims and limits liability exposure. Treating compliance as an isolated regulatory exercise, rather than as part of a broader IP and liability strategy, leaves significant value unrealised.
Cross-border dimensions for European clients
Belgium occupies a singular position within the EU's legal geography. It hosts EU institutions, has a sophisticated multilingual legal system, and its courts regularly handle disputes with cross-border dimensions. For international businesses, this creates both opportunities and complications in AI-related IP matters.
Copyright protection in the EU is governed by a combination of national legislation and harmonised EU directives. The harmonisation is substantial but incomplete. Member states retain discretion in implementing certain provisions – including exceptions and limitations that affect how AI-generated works are treated. A work that might attract some form of protection in one member state may receive no protection in Belgium, or vice versa.
This divergence matters for businesses with multi-jurisdictional content strategies. A company that generates AI content in Belgium for distribution across the EU cannot assume that Belgian IP analysis translates directly to other member states. The interaction between Belgian intellectual property legislation and the laws of other civil law systems – France, the Netherlands, Germany – requires jurisdiction-specific review for each material market.
Enforcement of IP rights across borders adds a further layer of complexity. Belgian courts have jurisdiction to grant pan-European injunctions in certain circumstances, particularly where the defendant is domiciled in Belgium or where Belgian intellectual property legislation is directly implicated. However, enforcement of Belgian judgments in non-EU jurisdictions requires recognition proceedings that vary by country.
For businesses comparing AI-generated IP treatment across civil law jurisdictions, our analysis of AI-generated works and intellectual property in Portugal provides a useful comparative reference point. Particularly given the similarities in the underlying civil law approach to author's rights.
The cross-border training data problem deserves particular attention. Many AI systems used by Belgian businesses are trained on datasets assembled globally. Where that training incorporates works protected under Belgian intellectual property legislation without authorisation, the liability can attach to the Belgian operator even if the AI system itself was developed abroad. Belgian courts apply national and EU IP rules to the act of deployment, not only to the act of training.
A practical scenario illustrates this. A Belgian marketing agency uses a generative AI platform developed by a US company. The platform was trained on images scraped from European websites without licence. The agency deploys the platform to create campaign visuals for a Belgian client. Belgian rights holders whose images were used in training bring a claim. The agency, as a deployer in Belgium, faces exposure under Belgian intellectual property legislation even though it had no involvement in assembling the training data. Technology licensing agreements that do not include clear representations and indemnities about training data provenance leave the deployer unprotected in precisely this scenario.
To explore how Belgian and EU intellectual property rules apply to your cross-border AI deployment strategy, contact us at info@ferrazwhitmore.com.
Strategic recommendations and outlook
Businesses that treat AI-generated content as an undifferentiated asset class are systematically undervaluing some outputs and overestimating the protection available for others. A structured approach to IP strategy in Belgium requires four operational steps.
First, audit existing AI-generated content. Map all AI-generated or AI-assisted works currently in use or held as assets. For each category, reconstruct the degree of human creative input involved. Works with strong human creative contributions – documented through version histories, prompt records, and editorial decision logs – should be treated as protectable and registered or otherwise documented accordingly. Works generated with minimal human input should be flagged as unprotected and excluded from IP-dependent commercial arrangements.
Second, restructure technology licensing agreements. Standard software licensing terms drafted before generative AI became commercially prevalent almost never address the IP ownership question adequately. Agreements should specify who owns outputs, what warranties the licensor provides about training data provenance, and who bears liability for third-party IP infringement arising from use of the system. Belgian contract law provides significant flexibility in structuring these allocations, but the parties must do so explicitly.
Third, build AI Act compliance into IP workflow. The documentation and transparency obligations under AI Act compliance regimes should be designed to serve dual purposes. Compliance records that capture human decision points in AI-assisted creative processes are simultaneously the evidentiary foundation for copyright claims. This convergence is not accidental – it reflects a broader EU regulatory policy of requiring algorithmic accountability as a condition of operating AI systems commercially.
Fourth, monitor legislative developments actively. The Belgian legislature is considering amendments to intellectual property legislation that would address AI-generated works more directly. EU-level initiatives on the copyright status of AI outputs are also advancing. Businesses that engage with these processes – either through industry bodies or through direct advocacy – have the opportunity to shape rules that will govern their commercial models for years. Those that wait for settled law before acting risk having to restructure their asset base under a less favourable regime.
The outlook for AI-generated IP in Belgium is one of constructive uncertainty. Courts are beginning to encounter these cases. Doctrine is developing. The EU AI Act compliance infrastructure is being built. International businesses that establish sound practices now. documenting human creative contributions, auditing technology licensing agreements. Additionally. Aligning AI Act compliance with IP strategy. will hold a stronger position than those who defer until the law is fully settled.
The lost opportunity is not abstract. Businesses that fail to document human creative input forgo copyright protection that would otherwise be available. They expose themselves to infringement claims they could have avoided with properly structured agreements. And they miss the chance to build AI-generated content portfolios that can be licensed, sold, or used as collateral – assets that depend entirely on the underlying IP position being sound.
Our dedicated advisory service on AI and technology law in Belgium covers the full spectrum of these issues, from AI Act compliance assessments to IP ownership structuring and cross-border enforcement strategy.
Frequently asked questions
Q: Can an AI system hold copyright over a work it generates in Belgium?
A: No. Belgian intellectual property legislation reserves copyright protection for works that reflect the personality of a human author. An AI system has no legal personality and cannot hold rights. The practical question is whether sufficient human creative input surrounds the AI's output to qualify the human contributor as the legal author.
Q: How long does it typically take for Belgian courts to resolve an IP dispute involving AI-generated content?
A: Litigation timelines in Belgium vary considerably. A first-instance decision before the Brussels Enterprise Court may take between one and three years. Where urgent interim relief is sought, proceedings before the president of that court can produce a ruling within weeks. Cross-border dimensions involving EU law tend to extend timelines further.
Q: Is it a misconception that the EU AI Act directly grants or resolves IP rights for AI outputs?
A: Yes, that is a common misconception. The EU AI Act is a risk-classification and conformity regulation focused on algorithmic accountability and transparency obligations. It does not amend copyright or neighbouring rights legislation. IP ownership for AI-generated works remains governed by Belgian intellectual property law and, where applicable, the EU Software Directive transposed into Belgian law.
About Ferraz & Whitmore
Ferraz & Whitmore is an international law firm based in Lisbon, advising business clients across 46 jurisdictions. Our team combines Portuguese civil law expertise with English common law tradition to deliver cross-border legal solutions in AI and technology law, intellectual property protection, and technology licensing. We advise technology companies, institutional investors, and in-house legal teams on AI Act compliance, algorithmic accountability obligations, software liability, and IP ownership structuring across European and international markets. As a law firm in Belgium advising on digital services and AI regulation, we bring both doctrinal depth and practical cross-border experience to each matter. Our IP and technology practice covers matters before Belgian courts, EU institutions, and arbitral bodies. Engaging a lawyer in Belgium with multi-jurisdictional experience is essential where AI-generated works cross national borders. To discuss your situation, 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.