AI Act timeline: which deadline applies when?
All deadlines from 2025 to 2030
Prohibited practices and Article 4 have applied since 2 February 2025, the GPAI rules since 2 August 2025 and Article 50 since 2 August 2026. The next date is 2 December 2026. The high-risk duties follow on 2 December 2027 (Annex III) and 2 August 2028 (Annex I); existing public-authority systems by 2 August 2030 (Article 111(2)).
Every AI Act deadline in one table
The dates after the Digital Omnibus (Regulation (EU) 2026/1744). For a system already running, check the transition rule in Article 111.
| Obligation | Date | Who | Status |
|---|---|---|---|
| Prohibited practices (Art. 5) | 2 February 2025 | Everyone | Applies |
| AI literacy (Art. 4) | 2 February 2025; duty to take measures since 27 July 2026 | Providers and deployers | Applies |
| GPAI model obligations (Chapter V) | 2 August 2025 | Providers of GPAI models | Applies |
| GPAI enforcement and fines (Art. 101) | 2 August 2026 | Providers of GPAI models | Applies |
| Transparency (Art. 50) | 2 August 2026 | Providers (paras 1 and 2) and deployers (paras 3 and 4) | Applies |
| Marking for existing systems (Art. 50(2), Art. 111(4)) | 2 December 2026 | Providers of systems from before 2 August 2026 | Coming up |
| New ban on intimate deepfakes and child sexual abuse material (Art. 5(1)(ba) and (bb)) | 2 December 2026 | Providers and deployers | Coming up |
| Existing GPAI models (Art. 111(3)) | 2 August 2027 | Providers of models from before 2 August 2025 | Later |
| High risk Annex III, including FRIA (Art. 27) | 2 December 2027 | Providers and deployers | Later |
| High risk Annex I (products) | 2 August 2028 | Product manufacturers and providers | Later |
| Existing high-risk systems for public authorities (Art. 111(2)) | By 2 August 2030, earlier if significantly changed | Providers and deployers | Later |
Planning by period
Now
- Update the AI register, including systems that have been running for years
- Remove or block prohibited practices
- Record AI literacy measures per role
- Make Article 50 transparency work in chatbots, content processes and publications
Until 2 December 2026
- Providers: machine-readable marking working for existing systems
- Check generative systems for technical safeguards against the new ban
- Record per system the applicable date under Regulation (EU) 2026/1744
2027
- Build the high-risk files: risk management, data governance, logging, technical documentation and human oversight
- Carry out a FRIA where Article 27 requires one
- Record procurement and supplier arrangements
2028 and later
- Fold the AI Act into the CE route for Annex I products
- Public sector: bring existing high-risk systems into line by 2 August 2030 at the latest
- Make AI changes part of change management
Priorities for October 2026
Based on current status and upcoming deadlines
Complete timeline 2024 to 2030
All important milestones
Want these dates in your own calendar? View the AI Act calendar
The regulation is officially in force. Start of the phased implementation period.
- •Regulation published in Official Journal
- •Transition period begins
- •AI Office established
The prohibited AI practices in Article 5 (such as social scoring and manipulative AI) apply. Article 4 requires providers and deployers to take AI literacy measures.
- •Social scoring prohibited
- •Manipulative AI practices prohibited
- •Article 4: AI literacy measures
- •Since 27 July 2026: no individual level required
Original deadline for Code of Practice GPAI (delayed to July 2025).
- •Deadline postponed
- •Multi-stakeholder process continues
The European Commission publishes the definitive Code of Practice for general-purpose AI models with rules for transparency, copyright and safety.
- •4 chapters: Governance, Transparency, Copyright, Safety
- •Signed by, among others, Google, Microsoft, OpenAI, Anthropic and Amazon
- •A means to demonstrate compliance
Rules for General Purpose AI (ChatGPT, Claude, Gemini, etc.) are in effect. Providers must maintain technical documentation, respect copyright and publish training data summaries.
- •Transparency obligations for GPAI providers
- •Model card and documentation mandatory
- •Copyright compliance required
- •Systemic risk models (>10²⁵ FLOPs): additional obligations
The Commission missed this deadline for the Article 6 guidelines. Draft guidelines followed on 19 May 2026.
- •When a system is high-risk
- •Explanation of the Article 6(3) exceptions
- •Practical examples
Regulation (EU) 2026/1744 (published 24 July 2026) amends the AI Act: new high-risk dates, Article 4 as a duty to take measures and a new ban.
- •Annex III to 2 December 2027
- •Annex I to 2 August 2028
- •Article 50 not postponed
The AI Act applies generally. Article 50 (transparency) applies and the Commission enforces the GPAI rules. In the Netherlands the implementing act still has to designate the supervisors. The high-risk obligations follow later.
- •Supervision and enforcement structure active
- •Transparency obligations for limited-risk AI
- •GPAI enforcement
- •National market surveillance authorities in scope
The ban on AI for non-consensual intimate deepfakes and child sexual abuse material applies (Article 5(1)(ba) and (bb)). Existing systems for synthetic content must have the machine-readable marking of Article 50(2) (Article 111(4)).
- •Technical safeguards against misuse are decisive (Article 5(1a))
- •Only systems from before 2 August 2026 had extra time
Providers of GPAI models already on the market before 2 August 2025 must comply by now (Article 111(3)).
- •Documentation and copyright policy in place
- •New models had to comply from 2 August 2025
Obligations for certain high-risk areas such as biometrics, critical infrastructure, education, employment, migration and border control apply.
- •Risk management and technical documentation
- •Human oversight and monitoring
- •Registration and evidence where applicable
Rules for high-risk AI integrated into regulated products, such as lifts, toys, machinery and medical devices, apply.
- •Annex I product integrations
- •Conformity assessment in product context
- •Align documentation and CE route
High-risk systems already in use before the application date and intended for use by public authorities must comply by now at the latest (Article 111(2)). Other existing systems are caught only once significantly changed.
- •Includes the FRIA for existing public-authority systems
- •Earlier if significantly changed
- •Articles 5 and 50 already apply
Frequently Asked Questions
Answers to the most common questions about the EU AI Act
Article 4 already applies. Train your team now
LearnWize makes AI literacy demonstrable with role-based training, assessment and evidence reporting.
Ready for your first AI Act route?
Discover which EU AI Act gap your organization should solve first.