Section 2

The EU AI Act

Regulation (EU) 2024/1689 — the world's first comprehensive legal framework for artificial intelligence, in force since 1 August 2024.

Read the Act ↗ EU AI Policy Hub ↗
Entry into force

1 August 2024

The Regulation entered into force 20 days after publication in the Official Journal of the EU.

Full application

2 August 2026

Most provisions apply 24 months after entry into force, with earlier deadlines for prohibited practices (Feb 2025) and GPAI (Aug 2025).

Scope

Extra-territorial

Applies to any provider or deployer placing AI systems on the EU market or affecting persons in the EU — regardless of where the provider is based.

Fines

Up to €35 million / 7% turnover

Maximum penalties for violations involving prohibited AI practices. Fines scale with risk tier and organisation size.

What is the EU AI Act?

The EU AI Act (Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024) is the world's first comprehensive horizontal regulation of artificial intelligence. It was published in the Official Journal of the European Union on 12 July 2024.

The Act takes a risk-based approach, classifying AI systems into four tiers: Unacceptable Risk (prohibited), High Risk (heavy obligations), Limited Risk (transparency obligations), and Minimal Risk (voluntary codes of practice).

In addition to regulating AI systems, the Act introduces a new horizontal framework for General-Purpose AI (GPAI) models — a category that captures the large foundation models (GPT-4, Claude, Gemini, Llama, etc.) that underpin most modern AI products.

Key institutions

  • EU AI Office — European Commission body responsible for supervising GPAI models.
  • National Competent Authorities (NCAs) — member-state regulators for high-risk AI systems.
  • European AI Board — coordination body across NCAs.
  • Advisory Forum — technical and scientific expert body advising the Commission.

Risk Classification

Unacceptable Risk — Prohibited

AI systems that pose a clear threat to fundamental rights: social scoring by public authorities, real-time biometric surveillance in public spaces, manipulation using subliminal techniques, exploitation of vulnerable groups.

High Risk

AI systems in eight regulated sectors: biometrics, critical infrastructure, education, employment, essential services, law enforcement, migration, and administration of justice. Subject to conformity assessment, registration in the EU database, and ongoing monitoring.

Limited Risk

AI systems that interact with humans (chatbots, deepfakes) must disclose their AI nature. Providers must ensure users are informed they are interacting with an AI system.

Minimal Risk

All other AI systems — spam filters, AI in games, recommendation systems. No mandatory obligations but encouraged to adopt voluntary codes of practice.

GPAI Models & the EU AI Act Equivalent of Model Cards

The Act does not use the term "model card" but creates functionally equivalent documentation requirements for foundation models.

What the Act requires (Articles 53 & 55)

Providers of GPAI models must draw up and keep up to date technical documentation covering training methodology, evaluation results, energy consumption, copyright compliance, and safety testing. This must be made available to the EU AI Office and, in summary, to downstream providers.

Technical Documentation (Article 53)

All GPAI model providers must produce technical documentation including:

  • General description of the model and its intended uses
  • Training data description including provenance and filtering
  • Training methodologies and techniques
  • Model architecture and parameter count
  • Evaluation results on recognised benchmarks
  • Energy consumption of training and inference
  • Known limitations, bias, and safety characteristics

This is functionally equivalent to a model card, but with additional emphasis on transparency about training data and energy consumption.

Systemic Risk — Enhanced Obligations (Article 55)

GPAI models with systemic risk (defined as models trained using more than 10^25 FLOPs) face additional requirements:

  • Adversarial testing (red-teaming) across the model supply chain
  • Incident reporting to the EU AI Office
  • Cybersecurity protection measures
  • Energy efficiency reporting

Models currently meeting the systemic risk threshold include: GPT-4 series, Claude 3 Opus+, Gemini Ultra/Pro, and Llama 3 405B. This list will evolve as compute thresholds are reviewed.

These requirements go beyond traditional model cards by adding post-deployment monitoring and incident reporting obligations.

Model cards vs. EU AI Act technical documentation — key differences

Industry model cards are voluntary, vary in depth, and have no standardised format. The EU AI Act creates a mandatory, standardised technical documentation requirement with regulatory oversight — but aligns closely with best-practice model card content. Providers who already publish comprehensive system cards (e.g. OpenAI, Anthropic) are better positioned for compliance.

Key Dates & Timeline

Phased application of the EU AI Act obligations.

Date Milestone What applies
12 July 2024 Published in Official Journal Regulation enters the statute book
1 August 2024 Entry into force Regulation legally in force across the EU
2 February 2025 Chapter I & II apply Prohibited AI practices banned; definitions and scope active
2 May 2025 GPAI Codes of Practice First GPAI Code of Practice expected to be finalised
2 August 2025 GPAI provisions apply Technical documentation and transparency rules for GPAI models
2 August 2026 Full application High-risk AI obligations, conformity assessments, national enforcement
2 August 2027 Legacy systems High-risk AI systems already on the market must comply

Official Resources

Direct links to official EU and Commission documents.

Official Text

Regulation (EU) 2024/1689

The full text of the EU AI Act as published in the Official Journal of the European Union.

EU AI Office

AI Office Website

The official hub for the EU AI Office, including guidance documents, GPAI model registration, and consultations.

Guidance

GPAI Code of Practice

The first draft Code of Practice for GPAI model providers, developed through a multi-stakeholder process led by the AI Office.

Implementation

AI Act Implementation

Commission resources on implementing the AI Act, including delegated acts, standards mandates, and the EU AI database.