The European Artificial Intelligence Board, the AI Board for short, is the body in which the Member States implement the EU AI Act together. It has one representative per Member State, the European Data Protection Supervisor takes part as an observer, and the Commission's AI Office provides the secretariat without voting (Article 65 AI Act). The Board advises and assists the Commission and the Member States so that the Regulation is applied consistently everywhere (Article 66 AI Act). It is not a supervisor: it imposes no fines, handles no complaints and takes no decisions that bind your organisation. What it says still counts. The Commission must involve it in its guidelines, takes utmost account of its opinion on codes of practice, and national authorities take its guidance into account when they advise organisations. For a municipality, a bank or an employer the Board is not the body that inspects you, but it is one of the sources of the reading your supervisor will apply to you.
Picture this: a municipality reads that the AI Board is preparing recommendations on AI literacy. Should it adapt its training plan, and who holds it to account if it does not?
In short
The AI Board is established by Article 65 AI Act and held its first official meeting on 10 September 2024. The governance chapter applies since 2 August 2025 (Article 113 AI Act). Each Member State designates one representative for three years, renewable once. One of them chairs the Board. The AI Act requires two standing sub-groups, for market surveillance and for notifying authorities; the Commission's register now lists twelve. It issues opinions and recommendations, among other things on guidelines, codes of practice, standards and whether Annex III or Article 5 should change. Those outputs are not law and create no new duties for you. Your duties are in the AI Act; supervision of them sits with the national market surveillance authority or, for a small group of systems, with the AI Office. For the GDPR there is a separate European body, the EDPB, and in the Netherlands the Dutch Data Protection Authority (Autoriteit Persoonsgegevens).
What the AI Board is
Article 65(1) establishes the Board as the European Artificial Intelligence Board; the Regulation then calls it the Board. The official Dutch text calls it the Europese raad voor artificiële intelligentie, or AI-board. According to the Commission's page on the AI Board it is an advisory body created by the AI Act. The Commission's register of expert groups lists it as a formal, permanent group under the Directorate-General for Communications Networks, Content and Technology (DG CNECT), with the AI Act as its creating act.
| Element | What the AI Act provides | Source |
|---|---|---|
| Members | One representative per Member State | Article 65(2) |
| Observer | The European Data Protection Supervisor (EDPS) | Article 65(2) |
| AI Office | Attends the meetings without voting, provides the secretariat, convenes meetings at the Chair's request and prepares the agenda | Article 65(2) and (8) |
| Guests | Other national and Union authorities, bodies or experts, case by case, where the issues are relevant to them | Article 65(2) |
| Term | Three years, renewable once | Article 65(3) |
| Requirements for representatives | Relevant competences and powers at home to contribute to the tasks in Article 66, a single contact point for the Board and, where appropriate, for stakeholders, and empowered to facilitate consistency between national competent authorities | Article 65(4) |
| Rules of procedure | Adopted by the representatives by a two-thirds majority; they cover, among other things, the selection and tasks of the Chair, voting arrangements and the sub-groups | Article 65(5) |
| Chair | One of the Member State representatives | Article 65(8) |
| Way of working | Objective and impartial | Article 65(7) |
The Commission adds to this. According to its AI Board page, the EEA-EFTA countries also take part as observers, alongside the European Data Protection Supervisor. At the first meeting on 10 September 2024, Norway, Liechtenstein and Iceland attended as observers, and the Board's organisation and the adoption of its rules of procedure were on the agenda. According to the register of expert groups, the rules of procedure were adopted at that meeting of 10 September 2024; they are available there as a download. The ninth meeting on 17 September 2026 took place under the chairmanship of the Irish Presidency of the Council of the EU; Moldova attended as an observer for the first time. For agendas, participant lists and minutes the AI Board page refers to the register of expert groups. That register lists the Dutch Ministry of Economic Affairs (still under the name Ministry of Economic Affairs and Climate Policy) as the Dutch member authority; the name of the representative is not in it.
Note Article 65(4)(b): the Board representative is the contact point for the Board, not necessarily the single point of contact for the public that each Member State designates under Article 70(2) AI Act. The EDPB makes the same point: the two roles can sit with different bodies (Statement 3/2024, paragraph 10).
Sub-groups
Article 65(6) requires two standing sub-groups: one for market surveillance authorities on market surveillance, and one for notifying authorities on notified bodies. The market surveillance sub-group should act as the administrative cooperation group (ADCO) for the AI Act within the meaning of the Market Surveillance Regulation, Regulation (EU) 2019/1020. That is where the authorities that will inspect you meet. The Board may set up other standing or temporary sub-groups, and representatives of the advisory forum may be invited to them as observers.
The register of expert groups lists twelve active sub-groups. They include the administrative cooperation group (ADCO), a steering group on general-purpose AI and working groups on, among other things, AI in financial services, Annex III high-risk AI, prohibitions, law enforcement and security, regulatory sandboxes, standards, the interplay with other Union legislation and the review of the rules of procedure. We do not see a separate sub-group for notifying authorities in that list. On its AI Board page the Commission names two of them: one on the interplay between the AI Act and the Medical Devices and In Vitro Diagnostic Medical Devices Regulations (MDR and IVDR), and one on standards. The first produced, with the Medical Device Coordination Group, a document of frequently asked questions on the MDR, the IVDR and the AI Act (AIB 2025-1 and MDCG 2025-6, June 2025). That document itself says that the views in it are not legally binding and that only the Court of Justice can give binding interpretations of Union law. It predates the Digital Omnibus, so some of its dates are out of date. The high-risk requirements for systems under Annex I, for instance, now apply from 2 August 2028 (Article 113, third paragraph, point (c)(ii)), not 2 August 2027. The standards sub-group wrote three documents: on the progress of harmonised standards for high-risk systems, on possible elements of forthcoming standardisation requests under Article 40 AI Act, and on standardisation in other international fora.
What the AI Board does
Article 66 first sets the mandate: the Board advises and assists the Commission and the Member States to facilitate the consistent and effective application of the Regulation. It then lists what the Board may in particular do.
| Task | Article 66 |
|---|---|
| Contribute to coordination among national competent authorities and, with their agreement, support joint activities of market surveillance authorities (Article 74(11)) | Point (a) |
| Collect and share technical and regulatory expertise and best practices | Point (b) |
| Advise on implementation, in particular on enforcing the rules on general-purpose AI models | Point (c) |
| Contribute to harmonised administrative practices, including the derogation from conformity assessment (Article 46), regulatory sandboxes and testing in real world conditions (Articles 57, 59 and 60) | Point (d) |
| At the Commission's request or on its own initiative, issue recommendations and written opinions, including on codes of conduct, codes of practice and Commission guidelines, the evaluation of the Regulation, standards and common specifications, trends in AI value chains, and whether Annex III (Article 7) or Article 5 (Article 112) should be amended | Point (e)(i) to (vii) |
| Support the Commission in promoting AI literacy and public understanding of benefits, risks, safeguards, rights and obligations | Point (f) |
| Facilitate common criteria and a shared understanding of the Regulation's concepts, including through benchmarks | Point (g) |
| Cooperate with other Union bodies, including in product safety, cybersecurity, competition, financial services, consumer protection, data protection and fundamental rights | Point (h) |
| Cooperation with third countries and international organisations | Points (i) and (m) |
| Assist national competent authorities and the Commission in developing organisational and technical expertise, including by contributing to the assessment of training needs | Point (j) |
| Assist the AI Office in supporting national competent authorities in setting up regulatory sandboxes, and facilitate cooperation among sandboxes | Point (k) |
| Contribute to, and advise on, the development of guidance documents | Point (l) |
| Opinions to the Commission on qualified alerts about general-purpose AI models, and receiving Member State opinions on them | Points (n) and (o) |
Articles 65 and 66 were not amended by the Digital Omnibus, Regulation (EU) 2026/1744; the consolidated text of 27 July 2026 contains them unchanged.
Where else the AI Act gives the Board a role
Outside Articles 65 and 66, the AI Act mentions the Board in a series of places. These are the provisions that matter for organisations:
| Provision | Role of the Board | What it means for you |
|---|---|---|
| Article 4(3) | Adopts recommendations, taking into account European competence frameworks, to help the Commission and the Member States promote AI literacy, including by setting common objectives (inserted by Regulation (EU) 2026/1744) | A reference point for your measures, not a new duty for you |
| Article 6(5) | The Commission provides guidelines on high-risk classification, with practical examples, after consulting the Board; the deadline was 2 February 2026 | Check with the Commission which version exists before relying on it |
| Article 50(7) and Article 56(6) | The Commission assesses, taking utmost account of the Board's opinion, whether codes of practice are adequate. That arrangement applies since 27 July 2026 (Regulation (EU) 2026/1744); before, the AI Office and the Board assessed together (the former Article 56(6)) | Following a code found adequate gives you a recognised way to demonstrate compliance |
| Article 56(2) and (4) | The AI Office and the Board aim to ensure that the general-purpose AI code of practice covers at least Articles 53 and 55, with clear objectives | Affects the model provider and, through its documentation, you |
| Article 57(14) to (16) | National authorities, the EDPS and the AI Office cooperate within the framework of the Board on regulatory sandboxes (paragraph 14; the EDPS and the AI Office were added by Regulation (EU) 2026/1744). National authorities inform the AI Office and the Board of a new sandbox (paragraph 15) and report on it annually to both (paragraph 16) | Relevant if you want to test in a sandbox |
| Article 70(8) | National competent authorities take into account the guidance and advice of the Board and, where appropriate, the Commission when they provide guidance and advice | The Board's direct line to your supervisor; the Board's role was already there, Regulation (EU) 2026/1744 added small mid-caps |
| Article 72(3) | The Commission adopts guidance, including a template, for the post-market monitoring plan by 2 September 2027, taking utmost account of the Board's opinion. That is new since 27 July 2026: an implementing act was foreseen here before | For providers of high-risk systems |
| Article 75(2) | Market surveillance authorities that consider a general-purpose AI system with a high-risk use non-compliant inform the Board | The Board sees such files but does not decide them |
| Articles 90(2) and 92(1) | The AI Office informs the Board after a qualified alert and consults it before a model evaluation | Only for providers of general-purpose AI models |
| Article 96(1) | When issuing implementation guidelines, the Commission involves the Board and pays particular attention to SMEs, small mid-caps, local public authorities and the sectors most affected (involving the Board and the small mid-caps were added by Regulation (EU) 2026/1744) | Guidelines on, among others, Article 26 (deployers), Article 5 and Article 50 go through the Board |
| Article 112(9) | In evaluating the Regulation, the Commission takes into account the positions and findings of the Board | Changes to Annex III or Article 5 come partly through the Board |
Where the Board has no role is just as telling. The Union safeguard procedure in Article 81 AI Act, in which the Commission assesses whether a national measure against an AI system is justified, does not mention the Board; the Commission decides. A complaint about an AI system goes to a market surveillance authority (Article 85 AI Act), not to the Board. And fines are imposed by national authorities, or by the Commission through the AI Office; on fines imposed on providers of general-purpose AI models the Board only receives information (Article 101(4)).
What the Board has done so far
On its AI Board page the Commission lists nine meetings: 10 September 2024, 10 December 2024, 24 March 2025, 30 June 2025, 24 October 2025, 4 December 2025, 20 March 2026, 11 June 2026 and 17 September 2026. In its news item on the ninth meeting the Commission expected the tenth meeting on 18 November 2026.
As documents endorsed by the Board, the page lists:
- The adequacy assessment of the General-Purpose AI Code of Practice (AI Board Adequacy Assessment). On 1 August 2025 the Commission and the Board confirmed that the code is an adequate voluntary tool for providers of such models to demonstrate compliance. That assessment still fell under the former Article 56(6), under which the AI Office and the Board assessed together. More in our article on the AI Office.
- The adequacy assessment of the Code of Practice on Transparency of AI-Generated Content. The Commission concluded on 8 July 2026 that the code adequately covers Article 50(2), (4) and (5) (Commission); the Board confirmed this on 9 July 2026. According to the AI Board page, the code is therefore an adequate voluntary tool for providers and deployers of generative AI for the marking and labelling obligations.
- The document on the MDR, the IVDR and the AI Act (MDCG 2025-6) and the three documents of the standards sub-group.
At the ninth meeting, according to the Commission's report, the Commission gave an update on, among other things, the development of recommendations to support AI literacy; in our reading that is the work under Article 4(3). Cooperation on market surveillance and the governance of pre-market conformity assessment were also on the agenda. As of 17 November 2026 those recommendations are not on the AI Board page.
The Board next to the advisory forum, the scientific panel and the AI Office
Four European bodies around the AI Act are often confused.
| Body | Who sits on it | What it does | Source |
|---|---|---|---|
| AI Board | One representative per Member State; EDPS as observer | Advises and assists the Commission and the Member States, coordinates between national authorities | Articles 65 and 66 |
| Advisory forum | Stakeholders from industry, start-ups, SMEs, civil society and academia, appointed by the Commission; the Fundamental Rights Agency, ENISA, CEN, CENELEC and ETSI are permanent members | Provides technical expertise and advises the Board and the Commission, at their request also through opinions, recommendations and written contributions | Article 67 AI Act |
| Scientific panel | Independent experts selected by the Commission; the Commission sets their number in consultation with the Board | Advises and supports the AI Office, can issue a qualified alert about a general-purpose AI model | Article 68 AI Act and Article 90 |
| AI Office | A function of the European Commission | Enforces the rules for general-purpose AI models and is exclusively competent for the systems in Article 75(1); secretariat of the Board | Article 64 AI Act and Article 75 |
The difference in one sentence: the advisory forum brings in the outside world, the panel brings in science, the AI Office enforces, and the Board is where the Member States reach agreement among themselves and with the Commission. For when the AI Office is your supervisor, see the EU AI Office: what it does and when it supervises your organisation. An overview of the whole system is in EU AI Act enforcement: the key agencies.
What the Board's outputs mean legally
A recommendation or opinion of the Board is not law. Nowhere does the AI Act give the Board the power to take decisions that bind operators; it advises and assists. The Commission's guidelines themselves are not law either. That does not make them optional in practice. Board outputs reach you along three routes:
- Through the Commission. It must involve the Board in guidelines (Article 96(1)) and since 27 July 2026 takes utmost account of its opinion on codes of practice and on the post-market monitoring template (Articles 50(7), 56(6) and 72(3)).
- Through your national supervisor. It takes the Board's guidance and advice into account when it gives guidance and advice (Article 70(8)). What the Board recommends therefore shows up in your own supervisor's guidance.
- Through the law itself. In evaluating the Regulation the Commission takes the Board's positions into account (Article 112(9)), and the Board advises on amending Annex III and Article 5 (Article 66(e)(vii)).
In our reading you should therefore treat a Board output as an authoritative interpretation that your supervisor is likely to follow: if you depart from it, record why. Your duties themselves remain those in the AI Act.
The GDPR side: the Board is not the EDPB
For data protection there is a different European body: the European Data Protection Board (EDPB). The differences are large.
| AI Board | EDPB | |
|---|---|---|
| Legal basis | Article 65 AI Act | Article 68 GDPR |
| Status | A body established by the Regulation, supported by the AI Office | A Union body with legal personality |
| Members | One representative per Member State | The head of one data protection authority per Member State and the EDPS |
| Commission | The AI Office attends without voting and provides the secretariat | Takes part without voting rights (Article 68(5) GDPR) |
| Binding decisions | None | Yes, in disputes between supervisory authorities (Article 65 GDPR) |
| Guidelines | Advises on the Commission's guidelines | Issues its own guidelines, recommendations and best practices (Article 70 GDPR) |
The two connect along a few lines. The Board cooperates, as appropriate, with other Union bodies, expressly including in the field of data protection (Article 66(h) AI Act). The EDPS sits on the Board as an observer and, within the framework of the Board, coordinates with national authorities and the AI Office on regulatory sandboxes (Article 57(14) AI Act). The EDPB and the EDPS delivered a joint opinion on the proposal that became the Digital Omnibus on 20 January 2026 (recital 47 of Regulation (EU) 2026/1744).
In Statement 3/2024 of 16 July 2024 the EDPB recommended that Member States designate data protection authorities as market surveillance authorities for the high-risk systems in Article 74(8) AI Act, such as systems for law enforcement, migration and the administration of justice (Annex III points 6, 7 and 8), and that they consider doing so for other Annex III systems that affect individuals' rights, except where the AI Act already requires a particular designation, such as in the financial sector (paragraph 9). The EDPB also noted that the AI Act sets up no clear coordination between the AI Office and the data protection authorities or the EDPB (paragraph 13), and asked the Commission to establish cooperation with them (paragraph 15).
In the Netherlands no market surveillance authority for the AI Act has been designated yet. On 20 April 2026 the Dutch government proposed that existing supervisors supervise AI within their own domain, with the Dutch Data Protection Authority as supervisor where no clear supervisor exists, and the Dutch DPA and the Dutch Authority for Digital Infrastructure (RDI) in a coordinating role (Rijksoverheid). For the GDPR the Dutch DPA is already your supervisor. See also AI Act supervision in the Netherlands.
Three examples
The examples are made up.
Municipality. A Dutch municipality wants to procure a system that assesses whether residents are entitled to social assistance. That is a high-risk system under Annex III point 5(a). From 2 December 2027 its duties as a deployer apply (Article 26 AI Act) and it carries out a fundamental rights impact assessment beforehand (FRIA, Article 27 AI Act). If the system was placed on the market or put into service earlier, those duties apply only after a significant change in design, and for systems intended for use by public authorities by 2 August 2030 at the latest (Article 111(2)). Where the Board comes in: the Commission drafts the guidelines on the Article 26 duties with the Board's involvement, and must pay particular attention to local public authorities (Article 96(1)). Once those guidelines exist, they belong in the tender requirements and in the FRIA process: record how the municipality follows them and, where it departs, why. The template for the FRIA questionnaire comes not from the Board but from the AI Office (Article 27(5)). The FRIA may refer to the DPIA the municipality carries out under the GDPR, or include parts of it (Article 27(4)); the Dutch DPA supervises the DPIA. And the AI literacy recommendations from the opening? They support the Commission and the Member States (Article 4(3)). The municipality's own duty remains to take measures that support the AI literacy of its staff (Article 4(1)). Use the recommendations as a yardstick for that plan; it is the national supervisor, not the Board, that will hold it to account.
Bank. A bank uses a system that assesses the creditworthiness of individuals (Annex III point 5(b)). From 2 December 2027 it carries out a FRIA. If the system was placed on the market or put into service before 2 December 2027, the high-risk duties apply only once its design changes significantly from that date (Article 111(2)); for private parties that provision sets no final date like the 2 August 2030 for public authorities. For financial institutions the market surveillance authority is the national financial supervisor, insofar as the use is directly connected with the financial service (Article 74(6)), unless the Member State designates another authority (Article 74(7)). Market surveillance authorities meet in the Board's market surveillance sub-group (Article 65(6)), and the register lists a separate Board working group on AI in financial services. For the bank, what the Board says about standards matters most: it advises on technical specifications, harmonised standards and common specifications (Article 66(e)(iii) and (iv)), and the Commission consults it on standardisation requests (Article 40(2)). Ask your vendor which standards it applies; a harmonised standard gives a presumption of conformity only for the requirements it covers. If the Board later advises on amending Annex III (Article 66(e)(vii)), the list of high-risk uses may change for the bank; the amendment itself goes through the Commission (Article 7 AI Act). See from scoring algorithm to transparent credit decision.
Employer. An employer uses a recruitment platform that analyses applications and ranks candidates (Annex III point 4(a)). Emotion recognition in the workplace is already prohibited, except for medical or safety reasons (Article 5(1)(f) AI Act). The Commission's guidelines on prohibited practices fall under Article 96(1)(b), and the Board also advises on the review of Article 5 (Article 66(e)(vii)). The register lists Board working groups on prohibitions and on Annex III high-risk AI for this. From 2 December 2027 the duties of Article 26 apply, including human oversight and informing workers' representatives and affected workers in advance (Article 26(2) and (7)). If the system was placed on the market or put into service before 2 December 2027, the high-risk duties apply only once its design changes significantly from that date (Article 111(2)); for private parties that provision sets no final date like the 2 August 2030 for public authorities. For the AI literacy of recruiters and managers the same applies as for the municipality: the Board's recommendations are a yardstick, the duty is in Article 4(1). Under the GDPR the employer remains responsible for the DPIA and, where candidates are rejected on the basis of solely automated processing, for the rules of Article 22 GDPR; a score can also be such a decision where it plays a determining role in the rejection (CJEU, SCHUFA, C-634/21); in our reading the same applies to a ranking. See AI in recruitment and selection.
What to do now
- Know who your supervisor is. Not the Board, but the national market surveillance authority, or for a small group of systems the AI Office. For the GDPR, your data protection authority.
- Follow the Board's outputs through the Commission. The AI Board page lists the meetings and endorsed documents. Watch in particular for the AI literacy recommendations (Article 4(3)) and the Article 96 guidelines on Article 26.
- Record how you follow guidelines and recommendations. Per AI system in your register: which guidelines apply, which code of practice or standard your vendor follows, and where you depart and why.
- Ask your vendor about standards and codes. Which harmonised standards and which code of practice found adequate it applies, and whether it updates its documentation when the Commission adopts new guidelines.
- Start from your case. If you are unsure whether your system is high-risk, or which duties you have as a deployer, start with the case check.
Frequently asked questions about the European AI Board
Sources
Newsletter
Every Tuesday, the AI Act week ahead in 5 minutes
A practical briefing on deadlines, new guidance and enforcement, so you know what matters this week. No spam and you can unsubscribe in one click.
Practical and short · No spam · One-click unsubscribe