What is the AI Office?
The European AI Office is the European Commission body created to implement and enforce the EU AI Act, most importantly as the exclusive supervisor of general-purpose AI (GPAI) models across the Union. The architecture is deliberately split: national authorities police most AI systems in their own markets, while the AI Office polices the model layer, because a foundation model trained once is deployed everywhere and no single member state could sensibly supervise it.
What it actually does
- Facilitates the drawing up of the GPAI codes of practice and assesses them, and can request model documentation from providers;
- Designates models with systemic risk and runs the corresponding oversight tier;
- Can compel information, order model evaluations and demand corrective measures;
- Provides the secretariat for the AI Board (one representative per Member State, chaired by a Member State representative), attends its meetings without voting, and supports national regulators on cross-border cases.
The legal dimension: two layers of supervision
The AI Office supervises GPAI models and coordinates; member-state authorities enforce the Act against AI systems placed on their markets. One product can face both layers at once: the model inside it answers to the AI Office through its provider, while the system built around it usually answers to national regulators; under Regulation (EU) 2026/1744, however, the AI Office also supervises AI systems built on a general-purpose model by the same provider or group, and systems in designated very large online platforms and search engines, subject to sectoral exceptions. For anyone operating below the model layer, the practical instrument is the contract: the documentation your model provider owes you tracks what it owes the AI Office.
Turkish context
If a Turkish startup builds on top of a foundation model, the provider’s obligations to the AI Office (training-content summaries, model documentation) shape what the startup can pass through to its own customers and what its vendor contract must promise. If it fine-tunes a model substantially, it may inherit provider duties itself (see fine-tuning) and the AI Office is the regulator watching that boundary. Türkiye has no counterpart body: no AI-specific law is in force there, so the AI Office matters to Turkish companies exactly to the extent that their models or products reach the EU.
Do: mirror your model vendor’s AI Office documentation duties in your own contract, so that you receive what you must pass on. Don’t: fine-tune or rebrand a model substantially without checking whether you have crossed into provider territory.
Sources. Regulation (EU) 2024/1689 (AI Act).