Although new obligations under the AI Act, including transparency requirements, enter into force on August 2, this does not mean that the regulation will be fully applicable by that date. Some requirements have been postponed and, as with other complex regulations such as eIDAS or GDPR, rules and standards for the practical implementation of the requirements have to be developed over time. Additionally, AI is one of the fastest-developing technologies ever. This means that regulation has to evolve as quickly as possible to keep pace.
In our timeline, we show the dates that have already been determined as well as expected developments that have not yet been finalized.
|
When |
What |
Who |
|
June 13 2024 |
The AI Act was officially published in the Official Journal
|
|
|
August 1 2024 |
AI Act formally entered into force (transition periods apply) |
|
|
February 2 2025 |
Prohibited AI practices became illegal (e.g. certain social scoring, manipulative AI, some emotion recognition uses); AI literacy obligations begin. |
All providers and deployers |
|
August 2 2025 |
Rules for General-Purpose AI (GPAI) models became applicable; Member States designate AI authorities; AI Office governance becomes operational. |
Foundation model providers (OpenAI, Anthropic, Google, Meta, Mistral, etc.) |
|
Late 2025 – 2026 |
Codes of Practice, Commission guidelines, harmonized standards and technical guidance continue to be published |
Everyone implementing |
|
August 2 2026 |
Main enforcement phase begins. Transparency obligations apply (e.g. chatbot disclosure, certain AI-generated content disclosures), innovation measures begin, and regulators gain full enforcement powers |
Most organizations using |
|
December 2 2026 (AI Omnibus) |
Deadline for certain transparency solutions (such as machine-readable marking requirements) under the revised implementation schedule |
Providers of generative |
|
December 2 2027 (AI Omnibus) |
Stand-alone high-risk AI systems (e.g. hiring, education, credit scoring, recruitment, insurance) become subject to the main compliance obligations |
Providers of Annex III |
|
August 2 2028 (AI Omnibus) |
High-risk AI embedded in regulated products (medical devices, machinery, etc.) reaches its compliance deadline |
Manufacturers |
|
Expected timeframe |
Actions |
|
Q3–Q4 2026 |
First investigations and enforcement actions by national authorities |
|
2026–2027 |
Publication of additional harmonized standards from the European standardization organizations CEN and CENELEC |
|
2027–2028 |
First significant fines and court decisions that clarify ambiguous provisions |
|
2028 onwards |
Mature compliance ecosystem, certification schemes, established case law, and possible targeted amendments for AI agents |
The AI Act itself only defines the legal framework for fair AI use, compliance and related matters. How this will look at a technical level will be defined through standards, just like the ISO norms supporting other regulations. Companies affected by the AI Act will have to keep a close watch on the publication of these standards so they can react in time and adapt their systems accordingly.
The AI Act was drafted with generative AI in mind. However, in the meantime, the technology has evolved toward agentic AI. This raises new challenges for regulators. In a world where autonomous agents can browse the web, call APIs, write code, and purchase goods and services independently, responsibility plays a crucial role.
In the future, policymakers will have to set out rules that define whether agents can be contractually capable, to what degree human oversight should be mandatory, and how this can be controlled. Standards for logging and auditing will also have to be developed.
If you'd like to learn more about the latest deadline and find out what requirements must be met starting August 2, download our free white paper, "Meeting the EU AI Act's transparency requirements."