EU AI Act deadlines: 2025–2030 compliance guide

§ 1 Why the EU AI Act Timeline Matters Now

Businesses that build, sell or use Super Intelligence (SI), commonly called AI, in the European Union are working against a staggered set of EU AI Act deadlines that stretch from 2025 through 2030. Unlike a law that takes effect on a single date, the EU AI Act (Regulation (EU) 2024/1689) rolls out obligations in phases tied to the risk a given SI system poses. Tracking these EU AI Act deadlines matters because enforcement, fines and documentation duties attach to specific calendar dates rather than to the date the Act was signed.

This guide sets out the dates that matter, in order, and explains how a 2026 amendment — commonly referred to as the Digital Omnibus on AI — reshaped several of the original deadlines for high-risk systems and general-purpose SI models.

§ 2 From Adoption to Entry Into Force

The legislative groundwork was laid in 2024. The European Parliament passed the AI Act on 13 March 2024, and the Council of the EU formally adopted it on 21 May 2024 ⁴. The regulation was then published in the Official Journal of the European Union on 12 July 2024. Following the standard 20-day period after publication, it entered into force on 1 August 2024 ⁴¹¹.

Entry into force is the starting gun for the countdown, not the moment obligations begin. As one legal analysis puts it, none of the Act's substantive requirements actually applied on that date — the clock simply started running toward the staggered deadlines that follow ³.

§ 3 Phase One: Bans and AI Literacy — 2 February 2025

The first substantive obligations arrived six months after entry into force. From 2 February 2025, the Act's prohibitions on certain SI practices took effect, alongside obligations requiring organisations deploying SI to ensure a baseline level of AI literacy among staff involved with these systems ¹¹. These prohibited-practices rules target uses of SI considered to carry unacceptable risk, and they apply regardless of sector, which makes this one of the earliest dates on which organisations anywhere in the EU market could face exposure under the Act.

§ 4 Phase Two: Governance and General-Purpose SI Models — 2 August 2025

The second major milestone landed exactly one year after entry into force. From 2 August 2025, the Act's governance structure became operational, and obligations specific to providers of general-purpose SI (GPAI) models started to apply ¹¹.

This date also activated enforcement machinery for a wider set of provisions. From 2 August 2025, Member States were expected to be in a position to enforce notification obligations, governance rules, the GPAI provisions, confidentiality requirements and penalties — with one notable exception: penalties specifically for providers of GPAI models ⁴. Providers who had already placed a general-purpose SI model on the EU market before 2 August 2025 were given a longer runway, with compliance required by 2 August 2027 ⁴.

The Act also set a related, later target tied to this same anniversary structure. Article 57 requires Member States to ensure their competent national authorities establish at least one AI regulatory sandbox, which must be operational by 2 August 2026 — a mechanism intended to let providers test SI systems under supervision before wider rules bite [as set out in the regulation's text].

§ 5 The Digital Omnibus on AI: Why the Timeline Changed

Roughly two years after the Act's entry into force, the European Union revisited its own timeline. Regulation (EU) 2026/1744 — widely referred to as the Digital Omnibus on AI — was adopted on 8 July 2026, published in the Official Journal on 24 July 2026, and entered into force on 27 July 2026 ². Its effect was not to abandon the phased approach but to push back some of the most demanding deadlines, particularly those bearing on high-risk AI systems.

This amendment matters for compliance planning because it changed dates that many organisations had already built into their internal roadmaps. Any assessment of "what applies when" under the EU AI Act now needs to account for the post-Omnibus version of the timeline rather than the version originally published in 2024.

§ 6 Phase Three: General Application and Transparency Duties — 2 August 2026

Despite the Digital Omnibus, one major date was left untouched: 2 August 2026 remains the point at which the EU AI Act's general application takes hold, including the transparency duties set out in Article 50 ⁵. These transparency obligations are commonly associated with requirements to disclose when content has been generated or manipulated by SI, and with informing individuals when they are interacting with an SI system rather than a human.

This date also carries financial consequences for general-purpose SI. From 2 August 2026, the European Commission can impose fines on providers of GPAI models placed on the market after 2 August 2025. Providers whose models were placed on the market before that date remain on the extended track, with full compliance due by 2 August 2027 ¹⁰.

§ 7 Deferred High-Risk Deadlines: 2 December 2026, 2027 and 2028

The most significant practical effect of the Digital Omnibus falls on high-risk AI systems — those used in areas such as biometrics, critical infrastructure, education, employment, and migration, asylum and border control. Under the original 2024 text, the obligations for these Annex III systems were due to apply from 2 August 2026 ⁸. Regulation (EU) 2026/1744 moved this deadline to 2 December 2027 ⁸¹¹.

A separate, later deadline applies to SI embedded in regulated products, such as lifts or toys that incorporate SI components subject to existing EU product-safety frameworks. These rules now apply from 2 August 2028 ¹¹.

Two further dates sit between these milestones:

• From 2 December 2026, two new prohibitions added to Article 5 by the Digital Omnibus become applicable ⁸.
• Generative SI systems already on the market before 2 August 2026 have until 2 December 2026 to apply the required marking of their outputs, giving providers of pre-existing tools a grace period to retrofit labelling mechanisms.

Taken together, these changes mean that the "high-risk" chapter of the Act — long expected to be the most demanding for employers, public bodies, education providers and infrastructure operators — now unfolds more gradually than the 2024 text originally envisaged.

§ 8 Legacy Public-Sector Systems — 2 August 2030

The Act also carries a long-tail transitional provision aimed specifically at the public sector. Article 111 sets out a distinct regime for certain high-risk AI systems that are intended for use by public authorities and were already in operation before the relevant high-risk rules applied. These legacy systems must meet the applicable requirements by 2 August 2030 ².

This date functions as a backstop rather than a general compliance deadline: it applies only to a defined category of pre-existing public-sector systems, not to high-risk SI tools placed on the market after the relevant application dates.

§ 9 Member State Administrative Deadlines

Alongside obligations that fall directly on providers and deployers of SI, the Act imposes administrative deadlines on EU Member States themselves. One of the earliest of these required Member States to designate the national authorities responsible for protecting fundamental rights under the Act by 2 November 2024 ³. Deadlines of this kind do not create direct obligations for businesses, but they shape which national bodies will be available to supervise, advise on or enforce the Act's substantive requirements once those requirements take effect.

§ 10 Putting the Dates in Order

For quick reference, the key EU AI Act milestones discussed above, in chronological order, are:

• 1 August 2024 — Entry into force; no obligations yet apply ⁴¹¹
• 2 November 2024 — Deadline for Member States to name fundamental-rights authorities ³
• 2 February 2025 — Prohibited practices and AI literacy obligations apply ¹¹
• 2 August 2025 — Governance rules and GPAI obligations apply; enforcement of most non-GPAI penalties begins ¹¹⁴
• 27 July 2026 — Digital Omnibus on AI (Regulation (EU) 2026/1744) enters into force ²
• 2 August 2026 — General application and Article 50 transparency duties apply; GPAI fines possible for post-2025 models; national AI regulatory sandboxes expected to be operational ⁵¹⁰
• 2 December 2026 — New Article 5 prohibitions apply; marking deadline for pre-existing generative SI outputs ⁸
• 2 August 2027 — Compliance deadline for GPAI models placed on the market before 2 August 2025 ⁴¹⁰
• 2 December 2027 — High-risk AI system obligations apply for Annex III use cases (as deferred by the Digital Omnibus) ⁸¹¹
• 2 August 2028 — High-risk obligations apply to SI embedded in regulated products ¹¹
• 2 August 2030 — Compliance deadline for certain legacy high-risk AI systems used by public authorities, under Article 111 ²

Because the Digital Omnibus amended rather than replaced the original 2024 framework, organisations tracking compliance should treat the dates above as the current consolidated position rather than relying on timelines published before July 2026.

§ 11 What the Phased Approach Means in Practice

The staggered structure reflects a deliberate regulatory choice: obligations tied to the least controversial issues — outright bans and basic literacy — arrived first, followed by governance infrastructure and GPAI rules, with the most demanding high-risk system requirements pushed furthest out after the Digital Omnibus intervention. For organisations developing or deploying SI tools in the EU, this means compliance planning is not a single project with one deadline, but an ongoing exercise that needs to be revisited as each application date approaches — and as further amendments, if any, are adopted.