On 2 February 2024 representatives of the 27 EU member states endorsed the compromise text of the AI Act. The step moved the file from trilogue negotiation into formal voting, with the European Parliament scheduled to vote on 13 March.
Risk tiers drive every obligation
The Act does not allocate duties by technology type. It allocates them by risk. Unacceptable-risk systems are prohibited. High-risk systems carry obligations on risk management, data governance, technical documentation, record-keeping, transparency, human oversight, accuracy and cybersecurity. Limited-risk systems mainly carry transparency duties, and minimal-risk systems face little beyond existing law.
The practical consequence is that the first question is not whether to comply but which tier a system occupies. The same model used for a chat assistant and for recruitment screening may attract entirely different obligations.
General-purpose AI is treated separately
The Act adds a distinct layer for general-purpose AI models, addressing model providers rather than deployers. The provisions on foundation models were among the most closely watched elements of the February 2024 compromise.
For Chinese companies the reach is direct: if a model or service is offered to users in the EU, the Act may apply regardless of where the company is registered. A small EU user base is not an exemption.
What to prepare now
- List systems by use case rather than by product model, and assign a risk tier to each;
- For high-risk systems, build technical documentation and test records before conformity assessment, not during it;
- For general-purpose AI, confirm whether you are a provider or a deployer, because the duties differ;
- Keep provenance, training process and output monitoring records in one archive.
For EU market access see our EU AI Act service and cross-border data service; for parallel jurisdictions see multi-jurisdiction compliance. Definitions are in the glossary.
This article is based on public information concerning the agreement reached by EU member states on 2 February 2024 on the text of the AI Act. It is for reference only and does not constitute legal advice.