Belgium

Belgium’s Constitution does not contain any express provisions on regulating or restricting the use of AI. However, a number of constitutional norms bear directly on the development and deployment of AI systems:

• Equality of all Belgians, including the equality of all men and women: Article 10 of the Belgian Constitution (the “Constitution”). This provision is particularly relevant where AI systems are used in decision-making processes, as any output that treats individuals differently on the basis of gender or other characteristics protected under the equality principle may give rise to a constitutional challenge.
• Principle of non-discrimination: Article 11 of the Constitution. This principle is of central importance in the context of algorithmic bias, as AI systems trained on skewed or unrepresentative datasets risk producing discriminatory outcomes that could be challenged on constitutional grounds.
• Right to a fair trial, access to a court and judicial independence: Articles 12, 13, 14, 40 and 151 of the Constitution. These guarantees are engaged where AI tools are deployed within the justice system, for instance as decision-support instruments, since their use must not undermine the right to a fair hearing or compromise the independence and impartiality of the judiciary.
• Right to privacy: Article 22 of the Constitution. Given that AI systems frequently rely on the large-scale collection and processing of personal data, this constitutional right to privacy serves as a fundamental safeguard against disproportionate or unjustified intrusions into individuals’ private lives.

In a landmark ruling on the intersection of AI and judicial data, the Belgian Constitutional Court (Pierre Thiriar, Orde van Vlaamse balies, Grondwettelijk Hof nr. 149/2025, 13 November 2025) examined whether the legislature could lawfully restrict the development of AI-powered tools for processing data held in the Central Register (Centraal Register), a comprehensive database containing judicial decisions and procedural documents, to the judiciary alone, thereby excluding the legal profession and the private sector. The court upheld this restriction, finding that the legislature’s decision to confine AI development to the justice system, rather than opening it up to the legal profession, was constitutionally justified on the following grounds:

• The distinction between the judiciary and the legal profession rests on an objective criterion: the institutional role and public-interest mandate of the courts and serves the dual aims of supporting the judiciary in discharging its statutory duties more effectively and protecting the sensitive personal data contained in judicial files. The court recognised that these aims are of sufficient weight to justify differential treatment, particularly given the inherent risks that AI-driven data processing poses to the rights of individuals whose data appear in judicial records.
• By limiting AI development to the judiciary, the legislature ensures that such tools, which carry a well-documented risk of producing erroneous correlations, reinforcing biases and generating incorrect results, are built and deployed exclusively within a controlled, institutionally supervised environment. The court emphasised that the judiciary, unlike private actors, operates under strict procedural safeguards and public accountability mechanisms, making it better positioned to detect and mitigate the risks inherent in applying AI to sensitive judicial data.
• Applying a precautionary principle, the legislator reasonably concludes that AI development involving judicial personal data should not be left to the private sector for non-public purposes.

Belgium is part of numerous international human rights conventions, worth noting are the following conventions that relate to automatic processing and AI:

• Convention for the Protection of Individuals with regard to Automatic Processing of Personal Data (ETS No. 108); and
• Council of Europe Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law (CETS No. 225).

Patentability of an AI system

Under Belgian patent law, an invention must be

• new;
• involve an inventive step; and
• be industrially applicable in order to qualify for patent protection (Article XI.3, Belgian Code of Economic Law (BCEL)).

At first glance, this might seem to open the door for AI-related inventions. However, Article XI.4 of BCEL narrows the path considerably: mathematical methods and computer programs are not regarded as patentable inventions as such. That said, the exclusion is not absolute, where such methods or programs produce a “specific technical effect”, or where they function in combination with hardware and satisfy the standard patentability criteria, patent protection may still be within reach.

In practice, this means that a bare AI algorithm or model will not qualify for patent protection on its own. However, an AI system that delivers a concrete technical result (for example, by improving the performance of a physical device or optimising an industrial process) may well cross the threshold into patentable territory, provided it also meets the requirements of novelty, inventive step and industrial applicability.

Even where an AI system clears the patentability hurdle, a further practical challenge awaits: the patent application must sufficiently disclose how the system works. This is where the so-called “black-box” nature of many AI models, whose internal decision-making processes are notoriously difficult to explain or reproduce, can become a significant obstacle, potentially undermining an applicant’s ability to satisfy the disclosure requirement in practice.

Ownership of a patent

The question of who owns a patent is governed by Article XI.3 of the BCEL, which vests the right to a patent in the inventor or in his successor in title. This immediately raises an important question in the AI context: can an AI system itself be named as the inventor? The answer under Belgian law, which is fully aligned with the European Patent Convention on this point, is a clear no. Inventorship is reserved exclusively for natural persons, reflecting the longstanding principle that only a human being can conceive of an invention in the legal sense.

That said, the involvement of AI in the inventive process does not automatically disqualify an invention from patent protection. Where a natural person has made a genuine and qualifying inventive contribution, for instance, by defining the problem, designing the methodology or selecting and interpreting the AI’s output, the resulting invention may well be patentable in that person’s name. In such cases, the AI system is treated as a sophisticated tool deployed within the technical process, much like any other instrument used in research and development, rather than as a co-inventor entitled to recognition in its own right.

Under Belgian law, copyright protection is available for works that are the product of a creative activity, expressed in a concrete form and bearing the hallmark of originality. At the heart of this framework lies a distinctly human-centred requirement: only works that reflect the personal stamp of a human author, shaped by free and creative choices, can qualify for protection. This raises an immediate and increasingly pressing question in the age of generative AI: can content produced autonomously by an AI system attract copyright protection? Under the current framework, the answer is almost certainly no. Where, however, a human author has played a meaningful creative role in shaping the output, for example through deliberate prompting, selection or editing, the resulting work may still qualify as an AI-assisted creation eligible for copyright protection, provided the human contribution meets the originality threshold.

This human-centred approach is further reinforced by Article XI.170 of the BCEL, which expressly provides that an author must be a natural person. The implication is clear: an AI system, no matter how sophisticated its output, cannot be recognised as the author of a work under Belgian law.

Unlike patents or copyrights, trade secrets do not constitute a standalone category of intellectual property under Belgian law. Instead, they are protected as a distinct legal interest, designed to sit alongside and to complement the more traditional IP rights.

For information to qualify as a trade secret, three cumulative conditions must be satisfied under Article I.17/1 of BCEL. The information must be:

• secret (meaning it is not generally known among, or readily accessible to, persons within the circles that normally deal with that kind of information);
• of commercial value precisely because it is secret; and
• the subject of reasonable steps, under the circumstances, by the person lawfully in control of it, to preserve its secrecy.

The legal framework underpinning this protection is set out in Article XI.332/1 et seq. of the BCEL. In the AI context, trade secret protection is particularly relevant: proprietary algorithms, training datasets, model architectures and fine-tuning methodologies may all qualify for protection under this regime provided that the holder can demonstrate that each of the three conditions outlined above is met.

The Belgian courts have not yet had an occasion to rule on issues such as AI inventorship, copyright in AI-generated output, or database rights in AI training data.

With respect to personal data, the provisions of the GDPR apply in Belgium as supplemented by the Belgian Act of 30 July 2018 concerning the protection of natural persons with regard to the processing of personal data.

Further, we must note the sui generis right for databases. The sui generis right applies to databases that are the result of a qualitatively or quantitatively substantial investment. The elements that may be collected include text, sound, images, figures, facts or data, which are relevant in light of training data for AI systems. In principle the database holder has the right to decide who may use its database.

As noted above, the GDPR applies directly in Belgium and is enforced by the Belgian Data Protection Authority (DPA). Although the GDPR does not expressly address AI systems as such, a number of its provisions are of direct relevance to AI development and deployment:

• Article 5: the core data protection principles, including purpose limitation, data minimisation, accuracy, storage limitation, integrity and confidentiality, all apply to the training and deployment of AI systems.
• Article 22: individuals have the right not to be subject to decisions based solely on automated processing, including profiling, where such decisions produce legal or similarly significant effects, subject to limited exceptions.
• Article 35: a data protection impact assessment is mandatory where AI-driven processing is “likely to result in a high risk” to individuals’ rights and freedoms, including in cases of systematic automated decision-making or large-scale processing of biometric data.

Importantly, the AI Act (Regulation (EU) 2024/1689) does not override or otherwise affect the GDPR or related EU privacy and data-protection instruments. EU data protection law continues to apply in full to any personal data processed in connection with obligations arising under the AI Act.

In addition, the DPA has published an information brochure exploring the interplay between the GDPR and the AI Act in the context of AI system development. The brochure highlights a number of core principles that must be observed when processing personal data within AI systems (www.autoriteprotectiondonnees.be/publications/artificial-intelligence-systems-and-the-gdpr---a-data-protection-perspective.pdf):

• lawfulness, fairness and transparency of processing;
• purpose limitation and data minimisation;
• accuracy and up-to-dateness of data;
• security of processing; and
• demonstrating accountability.

The DPA has also signed the Joint Statement on Generated Imagery and the Protection of Privacy, dated 23 February 2026 (www.autoriteprotectiondonnees.be/publications/joint-statement-on-ai-generated-imagery-and-the-protection-of-privacy.pdf). The statement sets out a series of expectations for organisations that develop or deploy AI content generation systems, built around four guiding principles:

• implementing robust safeguards;
• ensuring meaningful transparency about AI systems;
• providing effective and accessible mechanisms to respond to individuals’ requests; and
• addressing specific risks to children.

Lastly, the Belgian DPA published an information brochure in April 2026, the first in a series (see www.dataprotectionauthority.be/publications/the-impact-of-artificial-intelligence-on-privacy.pdf). The aim of this brochure is to help individuals understand how AI systems may affect their privacy and the protection of their personal data, given that such systems are increasingly present in daily life and are often trained on personal data from sources such as social media, chatbots, and online purchases. It also provides practical recommendations to help individuals maintain control over their data.

Belgium has transposed the Open Data Directive into federal, regional, and community law. The national legislation requires public sector bodies to make their data available for re-use, subject to certain exceptions. Data.Gov.be is the federal online website where datasets can be retrieved.

Further noteworthy legislation includes the Data Governance Act (Regulation (EU) 2022/868) which aims to increase data availability and promote data sharing between sectors and EU countries and the Data Act, which is applicable directly in Belgium and sets out a horizontal set of rules on data access and use.

Lastly, it should be noted that Belgian copyright provides certain exceptions with respect to the reuse of copyrighted material, for example the reproduction of lawfully accessible works for the purposes of text and data mining, provided that the use of those works has not been expressly reserved in an appropriate manner by the author (Article XI.190 BCEL).

The DPA has published its recommendation with respect to the processing of biometric data. This includes valuable recommendations with respect to the facial recognition data (www.gegevensbeschermingsautoriteit.be/publications/aanbeveling-nr.01-2021-van-1-december-2021.pdf).

The Belgian DPA has not yet issued specific guidance on AI-based voice and facial recognition data in the private sector context, but any such deployment must comply with the GDPR, the AI Act prohibitions and high-risk obligations, and the Belgian Act of 21 March 2007 regulating the installation and use of surveillance cameras.

As noted in Section 1.1 above, the principle of non-discrimination is firmly anchored in the Constitution. This constitutional guarantee is reinforced at the legislative level by the Belgian Act of 10 May 2007 combating certain forms of discrimination, which sets out detailed anti-discrimination rules backed by criminal sanctions. The Act prohibits both direct and indirect discrimination.

A direct distinction on the basis of one or more of the following protected criteria amounts to direct discrimination, unless it can be objectively justified by a legitimate aim pursued through appropriate and necessary means such as age, sexual orientation, civil status, birth, wealth, belief or philosophy of life, political conviction, trade union conviction, language, state of health, disability, a physical or genetic characteristic, or social origin or status.

An indirect distinction based on any of these criteria likewise constitutes indirect discrimination, unless:

• the apparently neutral provision, criterion or practice giving rise to the indirect distinction is objectively justified by a legitimate aim and the means of achieving that aim are appropriate and necessary; or
• in the case of an indirect distinction on the basis of disability, it is demonstrated that no reasonable adjustments can be made.

Many EU instruments are relevant with respect to cybersecurity and resilience. For instance, the Cyber Resilience Act (Regulation (EU) 2024/2847) (CRA) and NIS2 Directive (Directive (EU) 2022/2555). Although both instruments do not specifically regulate AI, both laws will have an impact on the AI industry.

• The CRA imposes new requirements for products with digital elements, i.e. hardware and software products.
• The NIS 2 Directive as transposed in the Belgian Act of 26 April 2024 establishing a framework for the cybersecurity of networks and information systems of general interest for public security (the “NIS2 Law”). When an entity is subject to the NIS2 Law, that entity must, among other things, ensure its supply chain security. AI providers could be part of their supply chain, and the necessary measures must be taken in this respect.

The Belgian legal framework for AI is currently governed by EU legislation (such as the AI Act). No specific Belgian antitrust legislation targeting AI has been adopted to date. The Belgian Competition Authority applies general competition law principles to AI-related conduct such as unfair practices.

Competition is regulated by Book IV of BCEL and Article 101 of the Treaty on the Functioning of the European Union. The Belgian Competition Authority has jurisdiction to investigate anti-competitive agreements, abuse of dominance, and merger control.

At this moment, the AI Act is the primary instrument that regulates artificial intelligence in Belgium. However, AI agents are not specifically addressed in the AI Act.

Further, under Belgian law, the AI agents have not been recognised as having legal personality, contractual capacity or the ability to act as legal principals. Only natural persons and legal persons (companies, associations, public bodies) have legal personality under Belgian law.

Currently in Belgium, no sector has adopted binding rules specifically targeting agentic AI as a distinct category.

There are currently no specific liability rules for AI systems that would be applicable in Belgium only. Therefore, in case a claimant would suffer damages caused by an AI system, they will need to rely on the general product liability rules as set forth in Book 6 of the Belgian Civil Code (BCC).

Pursuant to Article 6.41 of the BCC the producer is liable for damage caused by its product. However, a critical limitation under current Belgian law is the definition of “product” in Article 6.42, which is defined as “any tangible movable property, even if it forms a component of another movable or immovable property, or if it has become immovable by destination”. This could mean that standalone AI software not embedded in a physical product may not fall within the scope of the current regime.

However, changes are expected in this respect, considering the Revised EU Product Liability Directive (Directive (EU) 2024/2853) that must be transposed into Belgian law by 9 December 2026 clearly mentions software as a product.

As mentioned above, there is currently legal uncertainty whether or not standalone AI software not embedded in a physical product may fall within the scope of the current product liability regime. In cases where it would not be a standalone product, the Belgian product liability regime will be applicable. Under this regime, a producer is strictly liable for damage caused by its product. However, the injured party must prove the damage, the defect, and the causal link (Article 6.41 et seq. of the BCC). It should be noted, however, that a producer may escape liability by proving, among other grounds, that it was impossible, given the state of scientific and technical knowledge at the time of placing the product on the market, to discover the existence of the defect.

Belgium has a well-developed general product safety and recall framework in Book IX of the BCEL. However, it is mainly designed for tangible products. For standalone AI systems, Belgian law mainly relies on the European Framework which is directly applicable and enforced through Belgian market surveillance mechanisms. Belgium has not adopted any standalone national legislation specifically addressing AI prod

There are currently no Belgian civil court judgments awarding damages for harm caused by an AI system as a product or under general tort law.

With respect to the upcoming AI regulation in Belgium, the revised EU Product Liability Directive (Directive (EU) 2024/2853) that must be transposed into Belgian law by 9 December 2026, is noteworthy. The Directive expands the scope of a product to software and clearly addresses in its recitals that a developer or producer of software, including AI system providers, should be treated as a manufacturer. Further, the Product Liability Directive introduces a more claimant-friendly strict product liability regime by introducing a presumption of defectiveness of a product in certain circumstances.

As mentioned in Section 1, the Belgian AI strategy (AI convergence plan) (see www.bosa.belgium.be/en/themes/digital-administration/digital-strategy-and-policy/national-convergence-plan-development) has been approved by the Council of Ministers on 28 October 2022, which stresses the importance of a reliable, safe and trustworthy approach to AI. It seeks to achieve this by putting in place a legal and regulatory framework that fosters innovation while safeguarding citizens’ fundamental rights and freedoms.