AIRegistra

EU AI Act

Key deadlines

The EU AI Act applies in phases. Different obligations kick in on different dates between 2024 and 2027. This article maps the timeline and what mid-market companies should plan for at each milestone.

1 August 2024 — entered into force

Regulation (EU) 2024/1689 entered into force. Adoption phase. No operational obligations on companies yet, but the clock started for everyone.

What to do: nothing required. Companies that started inventories and policies early at this point have a smoother path through 2026.

2 February 2025 — prohibitions and AI literacy

Chapter I (general provisions) and Chapter II (prohibited AI practices) start applying. Article 4 (AI literacy) is in Chapter II.

What this means in practice:

  • Prohibited practices (Article 5) are illegal from this date. Most mid-market companies are nowhere near these (social scoring, real-time biometric ID for law enforcement, exploitative manipulation of vulnerable groups). If you are uncertain, talk to counsel.
  • Article 4 (AI literacy) technically starts applying. There is no enforcement mechanism with teeth until later, but the obligation exists. Many companies treat 2 August 2026 as the practical deadline because that is when penalties under Article 99 kick in.

What to do by this date: at minimum, an inventory of AI tools your company uses, and a written commitment to a training plan. Even an informal “we will run our first session by Q3 2025” plan with named owners is better than nothing on paper.

2 August 2025 — GPAI provider obligations

Chapter V (general-purpose AI models) starts applying for general-purpose AI providers. Notification of competent authorities, member-state-level governance bodies must be set up, codes of practice ready.

What this means for mid-market companies: if you are a deployer (most of you), nothing operational changes. If you are a provider of a general-purpose AI model — i.e., you have trained or fine-tuned a substantial model that is available for downstream use — additional obligations attach. Unusual at mid-market.

2 August 2026 — full applicability of most obligations

Chapters III (high-risk AI systems), IV (transparency), VII (governance), IX (post-market monitoring), X (codes of conduct) start applying. Article 99 penalties enforce.

This is the date everyone references. From 2 August 2026:

  • High-risk AI system obligations (Annex III) apply to providers and deployers of those systems. If you are a deployer of a system listed as high-risk, you have specific obligations: appropriate technical and organisational measures, human oversight, monitoring of operation, retention of automatically generated logs.
  • Transparency obligations (Article 50) apply to providers and deployers — for example, obligations to disclose AI-generated content (deep fakes, AI-generated text presented as factual).
  • Penalties under Article 99 are enforceable. Fines up to €35M or 7% of worldwide turnover for the most serious violations; lower tiers for less severe.
  • Article 4 (AI literacy) continues to apply. The combination of “Article 4 applies + Article 99 fines apply” is what makes 2 August 2026 the practical deadline for AI literacy training programmes.

What to do by this date:

  • AI inventory complete and live.
  • AI usage policy adopted.
  • Article 4 training programme run at least once for in-scope staff with evidence captured.
  • Role declaration done; readiness checklist worked through to a defensible state per role.
  • Snapshot PDF generated and stored as point-in-time evidence.

This is the v1 deliverable shape AIRegistra is built around.

2 August 2027 — high-risk obligations for embedded systems

Article 6(1) (high-risk AI systems that are safety components of products) starts applying for products subject to existing EU product-safety legislation. Toys, machinery, medical devices, etc.

What this means for mid-market companies: unless you build hardware products subject to EU product-safety regulation, this date is unlikely to apply to you. If you do build such products and they include AI components, this is a date to plan around with your product compliance team.

2 August 2030 — public-sector legacy systems

Article 111 — high-risk AI systems already placed on the market by public-sector entities before 2 August 2026 must be brought into compliance by this date. Long tail provision; rarely relevant to mid-market private-sector companies.

Summary table

DateWhat appliesWho is affected
1 Aug 2024Regulation enters into forceEveryone — clock starts
2 Feb 2025Prohibitions, AI literacy (Article 4)Providers and deployers
2 Aug 2025GPAI provider obligationsGPAI providers
2 Aug 2026Most obligations + penaltiesProviders, deployers, importers, distributors
2 Aug 2027High-risk safety-component obligationsProduct-safety-regulated industries
2 Aug 2030Legacy public-sector systemsPublic-sector deployers

Where to go next


This is general information, not legal advice. Application to your company depends on your specific facts; consult your own counsel. Dates summarised from Regulation (EU) 2024/1689 — refer to the official Official Journal text for canonical wording.

4 min read · Last updated

Found something out of date? Email hello@airegistra.com.