← Regulators
AI Act EU 2024/1689 EU-wide

Who is the European AI Office: the EU's enforcer for GPT-class AI models

The European AI Office is the body that polices general-purpose AI — the GPT/Claude/Gemini-class models — directly from Brussels. What it does, the structural detail most coverage skips (it is not an independent agency — it is a department inside the European Commission), who runs it, and where it stops and national authorities take over — every fact sourced.

Published May 17, 2026 · 8 min read · 5 sources cited

Search for “the EU’s AI regulator” and you land on the European AI Office — then notice it is not an EU agency in some city with its own legal personality. It is a department inside the European Commission. That is not a contradiction; it is the design, and it changes who you actually deal with. This is a profile of what the AI Office is, what it does under the AI Act, and where its writ stops. Every fact is sourced.

Quick facts

  • Full name: European Artificial Intelligence Office (the “AI Office”)
  • Established: by Commission Decision of 24 January 2024 establishing the European Artificial Intelligence Office (C/2024/1459), published in the Official Journal on 14 February 2024; under its Article 9 it “shall enter into force on 21 February 2024”
  • Legal status: not a standalone EU agency — it is “part of the administrative structure of the Directorate-General for Communication Networks, Content and Technology” (DG CNECT — the Commission’s digital-policy department), inside the European Commission (C/2024/1459, Article 1). The decision states explicitly it “is not a European Office within the meaning of the Financial Regulation”
  • Based in: Brussels, as part of the Commission / DG CNECT (unit CNECT.A) (EU Whoiswho)
  • Statutory anchor: Article 64 of the AI Act (Regulation (EU) 2024/1689) — the Office was set up ahead of the AI Act and the Regulation then built its enforcement role around it
  • Director: Lucilla Sioli — Director, Artificial Intelligence Office (CNECT.A); an EU official since 1997, previously Director for AI and Digital Industry (EU Whoiswho)
  • Staff: more than 125, across 6 units and 2 advisors (European Commission)
  • Website: digital-strategy.ec.europa.eu/en/policies/ai-office

What the AI Office is — and what it is not

The AI Office is the European Commission’s central body for the AI Act — and, uniquely, the direct EU-level enforcer for general-purpose AI models (GPAI — the large foundation models that power many products, e.g. GPT, Claude, Gemini).

The detail most coverage skips: the AI Office is not an independent authority, and not a separate agency at all. A data-protection regulator like France’s CNIL or Spain’s AEPD is, by EU law, structurally independent of government. Even Spain’s AESIA is at least a distinct state agency. The AI Office is neither — its founding decision places it “as part of the administrative structure” of DG CNECT and notes it “is not a European Office within the meaning of the Financial Regulation” (C/2024/1459, Article 1 and footnote 6). It is staffed by Commission officials redeployed within DG CNECT (C/2024/1459, Article 8). When the AI Office acts against a GPAI provider, it is the European Commission acting — that is the structural fact to hold onto.

What the AI Office actually does

The Commission set the Office up before the AI Act was even adopted, “as a matter of urgency … following the political agreement of the co-legislators achieved on 8 December 2023” (C/2024/1459, recital 8), so implementation could start immediately. Its mandate, per Articles 2–7 of the founding decision, breaks into:

  1. Implementing and enforcing the AI Act, especially for GPAI models — developing the tools, methodologies and benchmarks to evaluate model capabilities, and investigating possible infringements (C/2024/1459, Article 3).
  2. Running the governance plumbing — it provides the Secretariat for the AI Board (the body of all 27 national AI authorities that coordinates enforcement across the EU) and administrative support to the scientific panel and advisory forum.
  3. Codes of practice and guidance — encouraging and facilitating the codes of practice (voluntary rulebooks a provider can sign to demonstrate it meets the AI Act’s GPAI obligations), and helping the Commission prepare guidelines, implementing and delegated acts.
  4. Coordination, not duplication — where an AI system also falls under the Digital Services Act or Digital Markets Act (the EU’s platform and big-tech-gatekeeper laws the Commission already enforces), the Office keeps that enforcement aligned, and it must not duplicate other EU bodies’ work (C/2024/1459, recital 7 and Article 3).
  5. International work — promoting the EU’s trustworthy-AI approach with third countries and international organisations.

The AI Office’s role under the AI Act

The AI Act (Regulation (EU) 2024/1689, the EU’s horizontal AI law) makes the AI Office the EU-level home for the part of the regime that does not sit with national regulators: general-purpose AI models. The GPAI rules (Chapter V) became applicable on 2 August 2025, and from that point providers of GPAI models answer to the AI Office in Brussels — not to 27 separate national authorities — for those obligations. For the GPAI rulebook, the timelines and the penalty ceilings, see the AI Act pillar.

Everything else in the AI Act — high-risk systems in banking, HR, healthcare, education; prohibited practices; transparency duties — is enforced by national competent authorities in each Member State (Spain’s AESIA, and the equivalent body in every other country). The AI Office supports and coordinates them; it does not replace them (C/2024/1459, recital 7).

The AI Office vs national authorities — who enforces what

This split is constantly blurred. The clean version:

  • AI Office (Commission, Brussels) — direct supervision and enforcement of GPAI models (the foundation models). It evaluates them, can request information and corrective measures, and prepares the Commission’s enforcement decisions.
  • National competent authorities — one or more per Member State (e.g. AESIA in Spain). They handle high-risk AI systems, prohibited practices and transparency on the ground, and impose those fines.
  • The AI Board — the 27 national authorities sitting together, with the AI Office as its Secretariat, to keep interpretation consistent so a provider cannot shop for the softest Member State.

So “who enforces the AI Act?” — the honest answer is both, by design: the EU level (AI Office) for the foundation models that cross every border, the national level for systems deployed in a specific market.

Leadership

The AI Office is run by a Director within DG CNECT, under the Commission’s internal hierarchy — it does not have an independent governing board the way an EU agency or a DPA does.

  • Director: Lucilla Sioli. Listed in the official EU Whoiswho directory as Director of the Artificial Intelligence Office (CNECT.A). She has been a European Commission official since 1997 and was previously Director for AI and Digital Industry; she was named to lead the AI Office when the Commission announced its set-up in May 2024 (EU Whoiswho).

Because the Office is a Commission department, its ultimate accountability runs up through DG CNECT and the responsible Commissioner, not to an independent board.

What this means for you

  • If you build or provide a general-purpose AI model: the AI Office is your direct EU-level supervisor under the AI Act. It can evaluate your model, ask for information and corrective action, and feed the Commission’s enforcement decisions — you are answerable to Brussels, not to 27 national regulators, for the GPAI obligations.
  • If you deploy AI in a single market (HR tool, credit scoring, a chatbot): your enforcer is the national competent authority, not the AI Office. The Office sets the central direction and coordinates; it is not who fines you for a high-risk deployment.
  • If you are weighing whether to sign a code of practice: that instrument is shaped and facilitated by the AI Office — it is the route it offers to demonstrate GPAI compliance.
  • If you are a journalist or researcher: “the EU AI Office” is a Commission department inside DG CNECT, created by Commission Decision C/2024/1459 (24 January 2024, in force 21 February 2024), not an independent agency; its statutory anchor is Article 64 of the AI Act; it is led by Director Lucilla Sioli.

TL;DR

The European AI Office is the European Commission’s central body for the AI Act and the direct EU-level enforcer for general-purpose AI models (the GPT/Claude/Gemini-class foundation models). It was created by Commission Decision C/2024/1459 of 24 January 2024 (in force 21 February 2024), ahead of the AI Act, and is not an independent agency — it sits inside DG CNECT, staffed by redeployed Commission officials, and the decision states it “is not a European Office within the meaning of the Financial Regulation”. Its statutory anchor is Article 64 of the AI Act; GPAI rules became applicable on 2 August 2025. National authorities (Spain’s AESIA and equivalents) enforce high-risk and prohibited AI; the AI Office runs the GPAI tier and the AI Board Secretariat. Director: Lucilla Sioli.

Sources