The Drum · Weekly AI news, freeSubscribe →
Timeline
The EU AI Act represents the world's first comprehensive legal framework governing artificial intelligence systems. Launched formally in April 2021 as a proposal by the European Commission, the legislation underwent nearly three years of negotiation between the European Parliament, Council of the European Union, and Commission before reaching political agreement in December 2023. The Act establishes a risk-based classification system that subjects high-risk AI applications to strict requirements including transparency, human oversight, and bias testing, while imposing lighter regulations on lower-risk systems. Implementation phases began in 2024, with full enforcement scheduled for 2026. This landmark regulation has reshaped global AI governance discussions and prompted other jurisdictions to develop their own frameworks.
The Story
The EU AI Act emerged from growing concern about AI's societal risks, particularly in high-stakes domains like criminal justice, employment, and healthcare. Early proposals focused narrowly on prohibiting certain AI uses, but negotiations broadened the scope to create a tiered system accommodating innovation while protecting fundamental rights. The final framework reflects compromise between those seeking strict prohibitions and those advocating lighter-touch regulation. Key tensions remain unresolved: defining what constitutes high-risk AI, determining enforcement mechanisms, and balancing innovation incentives against safety requirements. The Act's true impact will depend on how regulators interpret ambiguous provisions and whether compliance costs genuinely protect citizens or simply entrench incumbent firms.
2018
European Commission establishes AI High-Level Expert Group
The Commission convenes leading researchers and industry experts to assess AI's opportunities and risks for Europe. This group's work lays groundwork for future regulatory thinking.
2019-04
European Commission releases AI Ethics Guidelines
Non-binding guidelines emphasize trustworthiness, human agency, and fairness in AI systems. These principles later inform the regulatory framework.
2020-02
European Commission publishes White Paper on AI
The White Paper outlines a regulatory approach balancing innovation and safety. It proposes a risk-based framework that becomes central to the Act.
2021-04-21
European Commission proposes the AI Act
The formal legislative proposal introduces a risk-based classification system with four tiers: prohibited, high-risk, limited-risk, and minimal-risk AI. The proposal immediately sparks debate between innovation advocates and safety proponents.
2021-05
Tech industry launches coordinated lobbying campaign
Major AI companies and trade associations begin intensive efforts to weaken high-risk requirements. Concerns center on compliance costs and competitive disadvantage versus non-EU competitors.
2021-06
European Parliament's IMCO Committee begins detailed review
Parliamentary committees start line-by-line examination of the proposal. Hundreds of amendments are filed, reflecting diverse political priorities.
2021-09
Civil rights groups demand stronger prohibitions
Organizations including Amnesty International and Access Now call for banning facial recognition in public spaces. Their advocacy influences parliamentary positions.
2022-06
European Parliament votes on AI Act amendments
Parliament approves over 1,000 amendments, including provisions banning real-time facial recognition in public. The vote signals Parliament's preference for stricter rules than the Commission proposed.
2022-07
Council of the EU begins negotiations on AI Act
Member states enter formal negotiations, with significant disagreement over enforcement mechanisms and definitions of high-risk AI. France and Germany push for lighter requirements.
2022-09
Tripartite negotiations commence between institutions
The European Commission, Parliament, and Council begin intensive trilogue discussions. These negotiations will determine the Act's final form.
2022-11
ChatGPT launches, reshaping AI regulation debate
OpenAI's release of ChatGPT demonstrates AI's rapid capabilities and societal reach. Policymakers cite the chatbot as evidence that regulation cannot wait.
2022-12
Council presidency proposes compromise on high-risk definition
The Belgian presidency suggests narrowing the high-risk category to reduce compliance burden. This proposal becomes central to ongoing negotiations.
2023-01
European Parliament reaffirms facial recognition ban
Despite industry pressure, Parliament maintains its position on banning real-time facial recognition. This becomes a key sticking point in trilogue talks.
2023-02
Generative AI emerges as new regulatory concern
Rapid advancement of large language models prompts calls for additional provisions. Negotiators debate whether the Act adequately addresses foundation models.
2023-04
Spain assumes EU Council presidency
Spain takes over negotiations with renewed commitment to reaching agreement. The presidency prioritizes finalizing the Act before year-end.
2023-05
Trilogue negotiations intensify over prohibited practices
Negotiators debate which AI applications should be banned outright. Disagreement persists on social credit systems and emotion recognition in education.
2023-06
Commission proposes generative AI transparency requirements
New provisions require disclosure of training data sources and copyright compliance. These additions address concerns about foundation models not covered by original proposal.
2023-08
Breakthrough agreement reached on high-risk categories
Negotiators agree on specific high-risk applications including hiring systems, credit scoring, and law enforcement tools. The compromise expands the high-risk list beyond the Commission's original proposal.
2023-10
Final trilogue negotiations conclude on enforcement
Parties agree on fines structure: up to 6 percent of global revenue for violations. This penalty level reflects Parliament's insistence on meaningful deterrence.
2023-11
Generative AI provisions finalized in last-minute deal
Negotiators agree that large language models must comply with transparency requirements and copyright rules. Smaller open-source models receive lighter treatment.
2023-12-08
Political agreement reached on final AI Act text
After nearly three years of negotiation, the three EU institutions announce they have reached political agreement. The deal includes provisions on prohibited practices, high-risk systems, transparency requirements, and generative AI governance.
2024-01
European Parliament formally approves AI Act
Parliament votes 523 to 46 in favor of the final text. The overwhelming support signals broad consensus on the regulatory approach.
2024-02-02
Council of the EU formally adopts the AI Act
All member states approve the final legislation. The Act becomes official EU law and enters into force.
2024-05-24
AI Act officially enters into force
The regulation becomes binding law across the European Union. Most provisions take effect immediately, though full implementation extends to 2026.
2024-06
European Commission establishes AI Office
A dedicated office is created to oversee implementation and enforcement. The office begins developing detailed guidance for compliance.
2024-07
First compliance guidance published for high-risk systems
The Commission releases detailed requirements for documentation, testing, and human oversight. Companies begin assessing their AI systems against these standards.
2024-08
Tech companies announce compliance initiatives
Major AI developers commit to implementing Act requirements ahead of deadlines. Some establish dedicated compliance teams.
2024-09
First enforcement actions threatened against non-compliant systems
National regulators begin identifying AI systems that violate prohibited practices. Several companies receive formal notices.
2024-10
Global jurisdictions announce AI frameworks inspired by EU Act
The UK, Singapore, and other countries release AI regulations modeled on the EU approach. The Act becomes the de facto global standard.
2024-11
First major compliance deadline approaches for prohibited practices
Companies must cease deploying banned AI applications. Regulators prepare enforcement mechanisms.
2025-01
Prohibited AI practices ban takes full effect
Real-time facial recognition in public spaces and other banned applications become illegal. Enforcement begins across member states.
2025-06
High-risk AI compliance requirements become mandatory
Companies must implement full documentation, testing, and oversight systems for high-risk applications. Non-compliance triggers penalties.
2026-01
Full AI Act implementation deadline
All provisions of the Act become fully enforceable. The transition period ends and penalties reach maximum levels.
© 2026 Cedar & Bloom. All rights reserved.
