The EU AI Act enters into force

What the act is

The European Artificial Intelligence Act is a European Union law that sets rules for artificial intelligence systems. It entered into force on 1 August 2024^[1]. In EU law, entry into force is the formal step by which a measure becomes law; individual provisions can still begin to apply on later dates. The act's purpose is to replace differing national approaches with a uniform framework that applies across EU countries^[1].

From proposal to agreement

The Commission said the act was proposed in April 2021 and agreed by the European Parliament and the Council in December 2023^[1]. Under the EU's ordinary legislative procedure, the Commission drafts a proposal, and the Parliament and the Council then negotiate and must both agree before it can be adopted — which is why a proposal and the final agreement can be separated by a period of negotiation.

A framework built on risk tiers

The act follows a risk-based approach with four tiers: minimal risk, specific transparency risk, high risk and unacceptable risk^[1]. The logic of such an approach is that obligations scale with the potential for harm: a system that poses little risk carries few requirements, while uses that could affect people's rights or safety carry more.

At the transparency tier, systems such as chatbots must clearly inform users that they are interacting with a machine, and certain AI-generated content must be labelled as such^[1]. Someone talking to a customer-service bot, for example, is entitled to know that the replies are produced by software rather than a person.

At the top tier, AI systems that allow social scoring by governments or companies are banned as an unacceptable risk^[1]. Social scoring means rating people on their behaviour or personal characteristics and using those ratings to treat them differently. Placing such systems in the highest tier makes them prohibited outright rather than merely subject to conditions.

General-purpose AI models

The act also reaches general-purpose AI models — models trained on broad data that can be adapted to many different tasks, as distinct from systems built for a single narrow use. The Commission said the provisions on general-purpose AI models would enter into application 12 months after entry into force^[1]. It expected to finalise the Code of Practice for general-purpose AI models by April 2025^[1]. A code of practice is a practical document that sets out how organisations can meet the requirements the rules impose.

Supervision of this part of the framework sits with the AI Office set up under the Commission, which will supervise implementation and enforcement of the AI Act rules on general-purpose AI^[1]. Supervision and enforcement means monitoring whether the rules are being followed and having the means to act when they are not.

Why it matters

For organisations that build or deploy AI in Europe, the act means the same set of rules applies in every EU country rather than varying from one member state to the next^[1]. Two obligations are stated in concrete terms: users must be told when they are interacting with a machine, and certain AI-generated content must be labelled^[1]. Social scoring by governments or companies is prohibited as an unacceptable risk^[1].

Because the provisions on general-purpose AI models enter into application 12 months after entry into force^[1], and because the Code of Practice supporting them was expected to be finished by April 2025^[1], the framework's coverage of AI arrives in stages rather than all at once. The AI Office's role in supervising implementation and enforcement of the general-purpose AI rules^[1] is the part of the structure that carries those stages from text into practice.