On 21 May 2024 the Council of the EU formally approved the Artificial Intelligence Act. The Act was published in the Official Journal on 12 July 2024 and entered into force on 1 August 2024. Council approval was the final stage of the legislative process. The compliance pressure comes afterwards, from obligations that apply in stages according to risk level.
What approval means procedurally
After Council approval, the Act completed the parliamentary and Council stages and moved to publication and entry into force. Drafts, political agreements and the final text may differ in wording, so companies should work from the Official Journal version and compare citations and descriptions against any documents prepared earlier. A clause-by-clause check at this point is cheaper than discovering a mismatch during a supervisory inquiry.
The risk-tier structure
The Act allocates obligations by risk: prohibited practices, high-risk systems, general-purpose AI models, transparency duties and market surveillance each carry different requirements. Prohibitions and some transparency duties apply first; the full set for high-risk systems applies later. Placing a product in the correct tier is the first compliance step. A wrong tier classification throws off the technical documentation and conformity work that follows.
Role determines the scope of obligations
Providers, deployers, importers and distributors have different duties. A company may play different roles across different products, and may become a provider through white-labelling, substantial modification or placing on the market. A wrong role assessment misaligns the entire obligation list. Groups should decide early which legal entity carries technical documentation, quality management and market-surveillance cooperation.
The timetable for market entry
There is still time before the main obligations apply, but product design, technical documentation, quality management and conformity arrangements need to start early. For general-purpose AI providers, transparency and training-data summaries are particularly time-consuming. Embedding compliance milestones in the product roadmap is more manageable than assembling material before a deadline. Companies serving both Chinese and EU markets also need to reconcile two documentation systems so that the same function is not described inconsistently.
Overlay with existing rules
The AI Act is not a stand-alone framework. Personal-data processing remains subject to the GDPR, product safety, consumer protection and platform liability have their own rules, and some high-risk products must still meet existing conformity requirements. Treating the AI Act as a substitute misses parallel obligations. Compliance planning should cross-check three lines: product, data and sector, rather than produce a single AI checklist. Hardware and software lines have different starting points, and a group must also decide which entity bears provider obligations so that subsidiaries do not prepare inconsistent positions.
For role assessment and obligations in the EU market, see our EU AI Act service and multi-jurisdiction service; for cross-border data flows, see GDPR cross-border service.
This article is based on the Artificial Intelligence Act approved by the Council of the EU on 21 May 2024, published in the Official Journal on 12 July 2024 and in force from 1 August 2024, and on public information. It is for reference only and does not constitute legal advice.