On 13 March 2024 the European Parliament adopted the AI Act by 523 votes to 46, with 49 abstentions. The margin means the substantive content is settled; what remains are procedural steps.
From vote to application
After Parliament, the Act required formal approval by the Council, signature by the institutions and publication in the Official Journal. It entered into force on 1 August 2024.
Entry into force is not the same as full application. The Act applies in stages: prohibitions first, general-purpose AI obligations next, and the full high-risk regime later. Companies should not treat the transition periods as a reason to postpone preparation.
What the staging means in practice
- Prohibitions arrive first. Social scoring, exploitation of vulnerabilities and untargeted facial scraping face the shortest transition.
- General-purpose AI is a separate layer. Model providers carry documentation, information and copyright-related duties; deployers carry different ones.
- High-risk obligations are the heaviest. Risk management, data governance, technical documentation, record-keeping, transparency, human oversight, accuracy and cybersecurity require a systematic evidence chain.
Three judgments for Chinese companies
First, scope follows users, not registration. Offering services to users in the EU may bring the Act into play.
Second, identify the role. Provider, deployer, importer and distributor carry different duties, and one company may occupy several roles across products.
Third, treat compliance as documentation engineering. The technical file and records required by the Act are essentially a checkable account of model capability, limits and monitoring; retrofitting is difficult.
For EU market routes see EU AI Act; for data transfer see cross-border data; for parallel jurisdictions see multi-jurisdiction compliance.
This article is based on the European Parliament’s vote of 13 March 2024 and on public information. It is for reference only and does not constitute legal advice.