The EU AI Act enters into force
The European Union’s AI Act entered into force on 1 August 2024, twenty days after publication in the Official Journal. The law sorts AI systems into four risk tiers, bans some uses outright, and phases in over two years, with most high-risk rules applying from August 2026.
Why it mattered It was the first comprehensive binding AI law from a major regulatory bloc, and a text other governments have worked from while drafting rules of their own.
A regulation of this kind takes years to arrive. The European Commission published its proposal in April 2021, before general-purpose models reached the public. The European Parliament and the Council of the European Union then negotiated their own positions, the three institutions settled on a common text, and that text appeared in the Official Journal on 12 July 2024. Twenty days later, on 1 August, it became law in every member state.
The structure is the one proposed in 2021. Systems are sorted into four tiers according to what they are used for rather than how they are built. At the top sits a short list of uses banned outright, government social scoring among them. Below that come high-risk uses, permitted but carrying obligations on testing, documentation, and transparency before a system reaches the market. Lighter transparency duties cover a further group, and the rest of the field is left alone.
Entering into force is not the same as applying. The prohibitions bite from February 2025. Most of the high-risk obligations wait until August 2026. The law was therefore binding from the summer of 2024 and enforceable against most of the companies it names only two years later, a gap the union had used before when GDPR took effect after a two-year run-in.
The contrast with the United States was visible within two months. California’s legislature passed SB 1047, a narrower bill aimed only at the developers of the largest models, and the governor vetoed it on 29 September. The pledges sixteen companies had signed in Seoul in May were voluntary and carried no penalty for breaking them. At the close of 2024 the AI Act was the only comprehensive AI statute in force anywhere in a major market, which is why its four tiers turned up in the drafting done elsewhere.