Summary: The Act’s institutional architecture: the AI Office and AI Board, advised by the Scientific Panel of Independent Experts and the Advisory Forum. Both advisory bodies were appointed on 1 June 2026.
Sources: raw/Laws/The Scientific Panel of Independent Experts...md; raw/Laws/The Advisory Forum...md; raw/Laws/Enforcement of Chapter V under the EU AI Act...md; raw/AI-regulations/Regulation - EU - 2024_1689.md
Last updated: 2026-07-28
The structure
| Body | Legal basis | Role |
|---|---|---|
| AI Office | Commission | Supervises and enforces GPAI rules; facilitates codes of practice |
| AI Board | Art 65 | Member State coordination |
| Scientific Panel | Art 68 | Independent technical expertise; qualified alerts |
| Advisory Forum | Art 67 | Stakeholder technical expertise |
| National competent authorities / MSAs | Ch VII | High-risk system supervision |
Sources disagree on how many advisory bodies there are
The Scientific Panel page describes the Panel as “one of the three advisory bodies, alongside the AI Board (established under Article 65) and Advisory Forum (established under Article 67)” (source: The Scientific Panel of Independent Experts…md).
The Advisory Forum page describes the Forum as “one of the two advisory bodies, in addition to the Scientific Panel of independent experts (established under Article 68)” (source: The Advisory Forum…md).
Both are from the same publisher (Future of Life Institute’s AI Act site). The discrepancy is whether the AI Board counts as an advisory body: it is a Member State coordination body under Art 65 that also advises the Commission, so both readings are defensible. Treat “the two advisory bodies” as meaning the Panel and the Forum specifically, and the AI Board as a separate governance organ that happens to have an advisory function.
The Scientific Panel
Established by Art 68(1), with detailed arrangements in Implementing Regulation (EU) 2025/454. Members appointed 1 June 2026 (source: The Scientific Panel of Independent Experts…md).
Composition — “up to 60 independent experts serving two-year renewable terms, selected to ensure fair gender balance and geographical representation. There is at least one expert per EU Member State and EFTA/EEA country (maximum three per country), and at least 80% of experts come from EU/EFTA/EEA nations.”
Tasks (Art 68(3)(a)) include:
- “alerting the AI Office to possible risks posed by GPAI models, in accordance with Article 90”;
- “contributing to the development of tools and methodologies for evaluating capabilities of general-purpose AI models and systems, including through benchmarks”;
- “providing advice on the classification of general-purpose AI models with systemic risk”;
- advice on classification of GPAI models and systems generally.
Powers: qualified alerts and information requests.
Qualified alerts — the mechanism that matters
The Panel “can request information and formally alert the AI Office to possible Union-level systemic risks from GPAI models” (source: The Scientific Panel…md). Two consequences flow from an alert:
- Designation. Art 51(1)(b) allows the Commission to classify a model as having systemic risk “following a qualified alert from the scientific panel” — a route to designation that does not depend on the 10²⁵ FLOP compute presumption at all (source:
raw/AI-regulations/Regulation - EU - 2024_1689.md). See GPAI and systemic risk. - Enforcement. The Panel is one of three routes by which enforcement against a GPAI provider can be triggered other than the Commission acting on its own initiative (source:
raw/Laws/Enforcement of Chapter V...md). See enforcement and penalties.
For AI security researchers this is the most relevant institution in the Act. It is a standing body of technical experts with a formal channel to trigger regulatory action on the basis of capability findings — including capability findings about a model that sits below the compute threshold.
The Panel’s benchmark-development task also connects to the measurement gap in Art 15(2) and the evaluation requirements in Code of Practice Appendix 3.
The Advisory Forum
Established under Art 67. Advises the Commission and the AI Board, providing technical expertise. Appointed 1 June 2026 with 174 members selected from more than 700 applications (source: The Advisory Forum…md).
Composition — “a balanced selection of stakeholders drawn from different groups, including industry, start-ups, SMEs, civil society, and academia.”
Five permanent members:
- Fundamental Rights Agency (FRA)
- European Union Agency for Cybersecurity (ENISA)
- European Committee for Standardisation (CEN)
- European Committee for Electrotechnical Standardisation (CENELEC)
- European Telecommunications Standards Institute (ETSI)
ENISA’s permanent seat is the security-relevant detail. The three standards bodies’ presence reflects that the Act’s substance is largely delegated to harmonised standards — the presumption of conformity under Arts 40–41 and 55(2) runs through CEN/CENELEC/ETSI deliverables.
SME participation in the Forum is one of the Act’s named SME provisions — see AI Act and SMEs.
The enforcement asymmetry these bodies operate under
GPAI obligations have applied since 2 August 2025; Commission supervision and enforcement powers arrive 2 August 2026 (source: Enforcement of Chapter V…md). Both advisory bodies were appointed 1 June 2026 — two months before enforcement powers commence.
Review of the architecture
By 2 August 2028 the Commission must “evaluate the functioning of the AI Office, whether the AI Office has been given sufficient powers and competences to fulfil its tasks, and whether it would be relevant and needed… to upgrade the AI Office and its enforcement competences and to increase its resources.” The 2029 and 2031 reviews both revisit “the possible need for a Union agency” (source: raw/AI-regulations/Regulation - EU - 2024_1689.md). See timeline.