Why the Act exists

Lesson 1 of 5 in The EU AI Act at a Glance: Purpose, Scope, and the Risk Pyramid.

On 1 August 2024, Regulation (EU) 2024/1689 — the EU AI Act — entered into force: the world’s first comprehensive, horizontal law for artificial intelligence. Not a sector rule, not a voluntary framework — a directly binding regulation across 27 Member States, with fines that reach 7% of worldwide turnover.

To read it correctly, you need to know what kind of law it is. The AI Act is, at its core, product-safety legislation. It descends from the EU’s New Legislative Framework — the same machinery that governs toys, lifts, and medical devices. In that world, a product must meet essential requirements, prove it through conformity assessment, and carry a CE marking before it can be sold anywhere in the Union. The AI Act takes that fifty-year-old machine and points it at algorithms.

Article 1 announces a hybrid that runs through everything else: product safety plus fundamental rights. Toys don’t discriminate; hiring algorithms do. So the Act grafts rights protections — non-discrimination, privacy, human dignity, access to remedies — onto a framework built for physical hazards. When you later meet oddities like a fundamental rights impact assessment sitting next to CE marking, this hybrid parentage is why.

The third motive is the quietest and the most legally decisive: single-market harmonisation. By 2021, Member States were drafting their own AI rules, threatening 27 incompatible regimes. A single EU-wide regulation pre-empts that fragmentation — which is also why the Act is a regulation (directly applicable everywhere, no national transposition) rather than a directive. One market, one rulebook, one CE mark.

Key terms: EU AI Act, New Legislative Framework, CE marking, fundamental rights, regulation vs directive

The legislative history matters mostly for one reason: generative AI arrived mid-draft. The Commission proposed the Act in April 2021 — before ChatGPT existed — with a clean, use-based risk pyramid. By the time Parliament and Council struck their trilogue deal in December 2023, foundation models had broken the assumption that every AI system has a knowable purpose. The negotiators bolted on an entire chapter for general-purpose AI models, and the Act you study today carries both architectures: risk tiers for systems, a parallel track for models. Keep that dual structure in view — it is the next lesson.

Interactive checkpoint quiz (2 questions) — open this page in a browser to take it.