Writing

EU AI Act Delay to 2027? It Didn’t Pause — It Moved. What Applies in October 2026

10 min read

Station departure board being reset at night, with one northern platform still waiting – the AI Act’s shifted timetable
The AI Act’s timetable was reset, not cancelled – and Norway’s platform is still waiting for the EEA train. Illustration: AI-generated.

The EU AI Act was not paused – it was amended. Regulation (EU) 2026/1744, the Digital Omnibus on AI, has been in force since 27 July 2026 and moves only the high-risk rules: Annex III use cases now apply from 2 December 2027, AI in regulated products such as medical devices from 2 August 2028. Prohibitions, GPAI duties and Article 50 transparency already apply.

Founders and CTOs keep asking me: “Is the AI Act delayed?” The honest answer is “parts of it, by a fixed number of months, in a law that is already in force.” A delay you can plan against is not a pause you can wait out – and the omnibus also adds a new ban that starts in six weeks. Here is the dated before-and-after, per October 2026, from someone building clinical AI that sits close to the high-risk line.

  • Digital Omnibus on AI = Regulation (EU) 2026/1744, signed 8 July 2026, published in the Official Journal on 24 July, in force since 27 July 2026.
  • Annex III high-risk (employment, education, biometrics, critical infrastructure, migration and more): 2 August 2026 → 2 December 2027.
  • Annex I high-risk (AI in products such as medical devices, lifts, toys): 2 August 2027 → 2 August 2028.
  • Not moved: prohibitions and AI literacy (since 2 February 2025), GPAI obligations (since 2 August 2025), general application and enforcement (since 2 August 2026).
  • New on 2 December 2026: a ninth prohibition on AI that generates non-consensual intimate imagery or child sexual abuse material, and the end of the grace period for machine-readable labelling by existing generative systems.
  • Norway: the AI Act is not yet part of the EEA Agreement; the government aims to put its KI-lov before the Storting in spring 2027.

Is the EU AI Act delayed? What the omnibus actually moved

Only one block moved: Chapter III, Sections 1–3 – the high-risk classification rules, requirements and obligations. According to the amended Article 113 in Regulation (EU) 2026/1744 on EUR-Lex (Official Journal, 24 July 2026), those sections now apply on two new dates.

Annex III: 2 August 2026 → 2 December 2027

Annex III covers high-risk use cases rather than products. The European Commission’s AI Act page (accessed October 2026) lists biometrics, critical infrastructure, education, employment, migration, asylum and border control among them. Under the original AI Act, Regulation (EU) 2024/1689, these obligations would have applied from 2 August 2026. They now apply from 2 December 2027 – a shift of 16 months.

Annex I: 2 August 2027 → 2 August 2028

Annex I covers AI in products already regulated by EU harmonisation law – the Commission cites “lifts or toys”, and the omnibus names the Medical Device Regulation (EU) 2017/745 and the IVD Regulation (EU) 2017/746 among the Section A acts. Note that this date did not move from 2026: the original Article 113(c) already set 2 August 2027. The omnibus adds one year.

Why the dates moved

The recitals of Regulation (EU) 2026/1744 point to “the delayed preparation of standards” and the delayed establishment of national authorities, and say that keeping 2 August 2026 would “risk a significant increase in implementation costs”. The rules did not move because Brussels lost its nerve; they moved because the harmonised standards were not ready.

What is the Digital Omnibus on AI?

The Digital Omnibus on AI is an EU regulation that amends the AI Act, the EASA Regulation (EU) 2018/1139 and the Machinery Regulation (EU) 2023/1230 to simplify how the AI Act is implemented. The Commission proposed it on 19 November 2025 as part of its digital simplification package; according to the Commission, political agreement was reached on 7 May 2026, and the regulation entered into force on 27 July 2026, the third day after publication.

Beyond the dates, three changes are worth knowing:

  • AI literacy is softened, not removed. Article 4 now requires providers and deployers to “take measures to support the development of AI literacy” of their staff, and explicitly “does not require providers or deployers to guarantee any specific level of AI literacy of any individual”.
  • Bias-detection data. A new Article 4a gives a legal basis under the AI Act to process special categories of personal data for bias detection and correction – it does not replace your GDPR analysis.
  • A stronger AI Office. Its exclusive competence now covers AI systems built on general-purpose models also where model and system come from different providers “that form part of the same undertaking”. Member States must also have a national AI regulatory sandbox “operational by 2 August 2027”.

Which AI Act obligations did not move?

Most of what people worry about day to day is untouched. The Commission’s own summary is blunt: “The AI Act entered into force on 1 August 2024 and became applicable on 2 August 2026, with some exceptions.” Per October 2026, these already apply:

  • Prohibitions 1–8 – since 2 February 2025.
  • General-purpose AI obligations – since 2 August 2025, with the AI Office’s power to fine GPAI providers live since 2 August 2026. More in what GPAI enforcement from 2 August means in practice.
  • Article 50 transparency (disclosing AI interaction, marking synthetic content) – since 2 August 2026.
  • Enforcement. “From 2 August 2026, the AI Office and authorities of the Member States are responsible for implementing, supervising and enforcing the AI Act,” according to the Commission.

The fine ceilings are unchanged. Under Article 99 of the AI Act, breaching a prohibition can cost up to EUR 35 million or 7 % of worldwide annual turnover, whichever is higher; breaching other operator obligations, including Article 50, up to EUR 15 million or 3 %.

What starts on 2 December 2026?

Two things land six weeks after this post.

A ninth prohibition

New points (ba) and (bb) in Article 5(1) ban AI systems that generate or manipulate realistic images, video or audio of an identifiable person’s intimate parts or sexual activity without that person’s explicit consent, and systems that generate child sexual abuse material. The Commission calls it “Prohibition 9”, aimed at “AI ‘nudification’ apps”. The scope is narrower than headlines suggest: it covers systems intended for this, and systems where such output is a “reasonably foreseeable and reproducible outcome” without adequate technical safeguards. A general image model with working guardrails is not automatically caught – one without them may be.

The Article 50(2) grace period ends

Under the new Article 111(4), generative systems placed on the market before 2 August 2026 must comply with Article 50(2) – machine-readable marking of output – by 2 December 2026. Newer systems never had a grace period. When I wrote about provenance becoming law in the EU and the US on 2 August, the full shape of this transition was not yet visible – keep this nuance in mind.

Where does healthcare AI land?

This is the part I live with. At MediVox we build clinical speech and documentation AI, and the first compliance question is always which annex you are in – the omnibus now splits them by eight months.

  • AI that is, or is part of, a medical device requiring a notified body under the MDR or IVDR is high-risk via Annex I. The AI Act’s recital 50 names “medical devices, in vitro diagnostic medical devices”. Date: 2 August 2028.
  • AI used in listed high-risk use cases is high-risk via Annex III. The AI Act explicitly mentions “emergency healthcare patient triage systems”. Date: 2 December 2027.
  • Everything else – an ambient scribe that is not a medical device, a patient-facing chatbot – is not high-risk by default, but Article 50 transparency and the AI literacy duty already apply.

My earlier post on what healthcare AI teams actually have to do assumed the old dates. The engineering advice – logging, traceability, documented human oversight – stands; the calendar just gives more time to build it properly.

Does the AI Act apply in Norway?

Not yet as Norwegian law. The AI Act is marked “Text with EEA relevance”, but an EU regulation only binds Norway once the EEA Joint Committee has incorporated it into the EEA Agreement and, where national law must change, the Storting has approved it. In a press release of 4 August 2026, the government wrote that it “is working to incorporate the AI regulation into the EEA Agreement and implement it in Norwegian law” (my translation).

Why Norway is behind its own plan

Norway originally planned to move in step with the EU. In the 2026 budget proposal of 15 October 2025, the KI-lov was “planned to enter into force at the same time as in the rest of the EU in August 2026” (my translation), with NOK 30 million proposed, NOK 20 million of it for supervision, guidance and a sandbox at Nkom, Datatilsynet and Digdir.

In August 2026, then-minister Karianne Tung said the government would send the law out for a second public consultation this autumn because of the omnibus, and that her ambition was to put it before the Storting in spring 2027. She attributed the delay to the omnibus and to negotiating “necessary EEA adaptations”. As of mid-October 2026 I could not find the new consultation published. Spring 2027 is an ambition, not a legal date.

Who will supervise it?

According to Nkom (24 March 2026), the government has decided that the Norwegian Communications Authority will be the coordinating market surveillance authority and national point of contact for the AI Act, with Norsk Akkreditering accrediting conformity assessment bodies. The rest of the national set-up is not yet decided.

The practical point for Norwegian builders: if you sell into the EU, EU dates apply to that market regardless of where your company sits. And at home, the government stresses that GDPR, the Penal Code and the Copyright Act already apply to AI.

What I would take from this

Treat the omnibus as a new timetable, not a signal that the AI Act is negotiable. Three things I would do now: label existing generative output before 2 December 2026; decide, in writing, which annex each product falls under; and use the extra 16 months (Annex III) or 12 months (Annex I) to build the audit trail and oversight that the high-risk rules will demand anyway.

Three predictions to hold me to:

  • The 2 December 2027 Annex III date will hold – no further general postponement of it will be adopted before that date (about 70 % confident).
  • The AI Act will not be applicable as Norwegian law before 1 July 2027 (about 80 % confident).
  • At least one EU or national authority will publicly announce an enforcement action under the new nudification ban by 30 June 2027 (about 55 % confident).

Frequently asked questions

Is the EU AI Act delayed to 2027?

Partly. Regulation (EU) 2026/1744 moved the Annex III high-risk rules from 2 August 2026 to 2 December 2027, and Annex I product rules from 2 August 2027 to 2 August 2028. Most other rules already apply.

What is the Digital Omnibus on AI?

It is Regulation (EU) 2026/1744, an EU law in force since 27 July 2026 that amends the AI Act to simplify implementation. It shifts the high-risk dates, softens the AI literacy duty and adds a new prohibition.

Which AI Act obligations did not move?

The prohibitions, AI literacy, the GPAI obligations, Article 50 transparency and the fine ceilings all kept their dates. Enforcement by the AI Office and national authorities has applied since 2 August 2026.

Does the EU AI Act apply in Norway?

Not yet as Norwegian law: per October 2026 it has not been incorporated into the EEA Agreement. The government aims to put its KI-lov before the Storting in spring 2027, with Nkom as coordinating supervisor.

When do the rules apply to AI medical devices?

AI that is a medical device needing a notified body falls under Annex I and has until 2 August 2028. Healthcare AI in Annex III use cases, such as emergency triage, has until 2 December 2027.

Portrait of Håkon Berntsen

Written by

Håkon Berntsen

CTO and co-founder of Skjld Labs AS. Builds production AI agents, Microsoft 365 systems and the courses around them, from Sandefjord.

About LinkedIn GitHub