Writing

Norway AI Act Consultation: What the KI-Omnibus Hearing Changes Before 11 December

11 min read

A printed consultation document on a bench in a winter-lit government corridor, with snow falling outside tall windows
Norway’s hearing on the EU AI omnibus runs until 11 December 2026 – a short memo with long consequences for anyone shipping AI into Norway. Illustration: AI-generated.

Norway’s AI Act consultation opened on 30 October 2026. The Ministry of Digitalisation and Public Governance wants views on Regulation (EU) 2026/1744, the AI omnibus, by 11 December 2026 and says no extension should be expected. It says the omnibus needs no changes to the draft KI-lov. My thesis: that makes Norway predictable, and predictability is worth more than the delay costs.

This is the follow-up to my October breakdown of what the AI omnibus moved and what already applies in the EU. This one covers the Norwegian hearing: what the memo says, which numbers are being over-read, and what I intend to submit.

Key facts per November 2026

  • What: a national consultation on the AI omnibus, published by the Ministry of Digitalisation and Public Governance (DFD) on 30 October 2026, reference 26/3045.
  • Deadline: 11 December 2026. The ministry writes that an extension “cannot be expected”. Anyone can respond, and responses are published.
  • Draft law: the ministry sees “in principle” no need to change the KI-lov provisions it sent out on 30 June 2025. Norway still plans to incorporate the AI Act by reference.
  • EU dates restated: Annex III high-risk rules from 2 December 2027, Annex I from 2 August 2028, at least one national AI sandbox by 2 August 2027.
  • New tier: small mid-caps (SMCs), with fewer than 750 employees and turnover up to EUR 150 million or a balance sheet up to EUR 129 million, get lighter obligations alongside SMEs.
  • Formal status: the AI Act is not yet part of the EEA Agreement and the KI-lov has not been passed. GDPR and existing Norwegian law already apply to AI.

What is Norway’s KI-omnibus consultation, and what is the deadline?

The KI-omnibus consultation is a supplementary hearing, not a new draft law. According to the consultation page on regjeringen.no (30 October 2026), it supplements the 30 June 2025 hearing on implementing the AI Act in Norwegian law. You respond through a form on that page, and every response becomes public under the Freedom of Information Act.

The consultation list includes Datatilsynet, Nkom, Finanstilsynet, NHO, Abelia, SINTEF and Simula, and four health bodies: the Norwegian Medical Products Agency, the Directorate of Health, Norsk helsenett and the Norwegian Board of Health Supervision.

In its press release of 30 October 2026, the ministry says it received “about 150” responses to the 2025 KI-lov hearing and is “currently working on the bill”. Digitalisation minister Torgeir Micaelsen said, in my translation, that the simplifications “should make the rules easier to follow and give better opportunities to test new solutions”, particularly for small and medium-sized businesses.

What does the memo change, and what does it leave alone?

I read the five-page consultation memo in full. Its most important sentence is a negative one: “the risk-based approach, the original prohibitions and the rules on general-purpose AI models are unchanged.” The omnibus adjusts timing and burden. It does not change the architecture.

What moves

  • High-risk deadlines. Annex III systems move from 2 August 2026 to 2 December 2027. Annex I systems, meaning AI embedded in regulated products, move from 2 August 2027 to 2 August 2028.
  • AI literacy (Article 4). Providers and deployers must now take measures to support AI literacy. The memo says operators “do not need to guarantee a specific level of competence in each individual”.
  • Bias testing (Article 10). The legal basis for processing special categories of personal data to detect and correct bias is widened. The memo describes it as permitted only where “strictly necessary” and subject to specific conditions.
  • Supervision. The EU AI Office gets wider powers over certain systems, including those built on general-purpose models and those inside very large online platforms or search engines.

What is added

Two new prohibitions: AI systems that generate or manipulate intimate content without consent, and systems that produce child sexual abuse material. According to the Commission’s AI Act Service Desk timeline, these apply in the EU from 2 December 2026.

The footnote everyone will quote

Footnote 1 of the memo reads, in my translation: “AI has been used as a support tool in the work on the consultation memo. AI-generated content has been quality-assured and edited.” That is not embarrassing. It is the spirit of the transparency Article 50 asks of AI-generated content, and I would like the same line in every tender response I review.

Does the EU AI Act apply in Norway yet?

Not formally, per November 2026. The EFTA EEA-Lex entry for the AI Act still lists Regulation (EU) 2024/1689 as “under scrutiny for incorporation into the EEA Agreement”. The KI-lov is still a bill in preparation. In August, then-minister Karianne Tung said existing law – data protection, the Penal Code, copyright – already applies, and that the consequence is “first and foremost that the introduction of the AI-specific rules … is postponed in Norway”. That was in the ministry’s release of 4 August 2026, which also said she aimed to put the bill to the Storting in spring 2027.

So 2 December 2027 is an EU date: the memo restates it, but no Norwegian source says it applies in Norway that day. That depends on EEA incorporation, the EEA adaptations Tung said were still being negotiated, and the KI-lov entering into force. A Norwegian vendor selling into Sweden or Germany is covered by the EU rules there regardless; that follows from the regulation’s scope.

What does the SMC category mean for Norwegian companies?

An SMC, or small mid-cap, is a company larger than an SME but within the thresholds above. The memo lists where SMCs get relief: technical documentation, quality management systems, sandbox access and penalties.

The memo then makes a striking claim. It cites SSB statistics showing that “only 987 businesses (0.2 %) have at least 250 employees”, and concludes that “the vast majority of businesses in Norway” will follow the simplified SME and SMC requirements. The SSB table (updated 12 January 2026) counts 656,492 active establishments on 1 January 2026, of which 448,658 had no employees.

You will see this reported as “99.8 per cent of Norwegian companies”. That is a subtraction, not an SSB figure, and it over-reads the data twice. First, SSB defines a bedrift as a locally delimited functional unit, which means an establishment, not an enterprise or a group. A subsidiary of a large group does not count as an SME under EU rules, however small the site. Second, the SMC ceiling is 750 employees, so 250 is not the relevant boundary. The direction of the claim is right: nearly every Norwegian AI deployer will be on a lighter track. The precision is not.

Health shows the same shape: SSB counts 60,739 establishments in health and social services, only 209 of them with 250 or more staff. Most sites that will run clinical AI are small, whatever group they belong to. They need guidance more than lighter penalties.

When does Norway’s AI regulatory sandbox open?

The legal deadline is 2 August 2027. That is when the omnibus requires each member state to have at least one AI regulatory sandbox in operation, a year later than the original 2 August 2026. Norway has chosen its model. According to Nkom (24 March 2026), the sandbox will be run in formal cooperation between Digdir, Datatilsynet and Nkom, placed organisationally in Digdir as part of KI Norge.

Nkom is clear about what the sandbox is not: neither a rule-free zone nor a formal pre-approval. Participants get limited protection from administrative fines under Article 57(12), not from liability for damage. On request they also get written evidence and an exit report that providers can use in conformity assessment. That exit report is the valuable part: it turns a regulatory conversation into an audit artefact.

Don’t confuse this with Datatilsynet’s existing privacy sandbox, a separate guidance service. I found no published opening date or application process for the AI Act sandbox.

Supervision follows the same pattern. Nkom confirmed in March 2026 that it will be the coordinating market surveillance authority and national contact point, with Norsk Akkreditering as accreditation body. It also said the national organisation “as a whole” was not yet decided.

Why is predictability worth more than the delay?

This is my opinion. Norway is behind its own plan: the sandbox deadline slipped a year, the supervisory map is incomplete, and the second KI-lov hearing promised for this autumn became a five-page supplement. You can read that as drift.

I read it differently. For a vendor, the worst state is a national text that might diverge from the EU text, so you build and certify twice. The memo removes that risk. Norway will incorporate the regulation by reference, the ministry proposes no national rule changes, and the EU dates are the dates. I can now design a clinical system for the EEA to one rulebook and one calendar. That is worth more than a sandbox twelve months earlier.

The counter-argument is real. Small Norwegian health-tech firms lose a year of structured regulatory dialogue, and the Annex III extension delays protections for patients as well as obligations for vendors. A delay that buys clarity is worth having. A delay that buys nothing is not, and the next six months will show which this is.

What I intend to submit

I plan to respond before 11 December in my own name, drawing on what we have learned building clinical AI at MediVox and Eir Tec. Three points:

  1. Prioritise health in the first sandbox cohort. Health dominates the consultation list, and most health establishments are small. If KI Norge opens with limited capacity, Annex III health use cases should go first, so the exit reports exist before December 2027 rather than after. See also what the AI Act means for healthcare AI teams.
  2. Clarify who supervises GPAI-based systems. The omnibus gives the AI Office wider powers over systems built on general-purpose models. I expect most clinical assistants sold in Norway to be exactly that. I will ask the bill to state where Nkom’s role ends and Brussels’ begins (background: GPAI enforcement from August 2026).
  3. Guidance on Article 10 bias testing with health data. The widened legal basis only helps if Datatilsynet and the health authorities say jointly what “strictly necessary” means in practice. The memo does not say GDPR Article 9 conditions are automatically met. Without joint guidance, careful teams will keep doing bias tests on synthetic data that does not represent Norwegian patients.

What I would take from this

  • Design to the EU text and the EU calendar. The omnibus regulation is now the Norwegian plan as well, just later.
  • Prediction: the KI-lov bill reaches the Storting in the first half of 2027 with no national deviations from the AI Act’s substantive rules.
  • Prediction: KI Norge’s AI Act sandbox will not take applications before spring 2027, and its first cohort will include at least one health project.
  • Prediction: the published responses will cluster on two themes, sector supervision and Article 10 health data, not on the deadlines.

Mapping a health or public-sector system to these dates is the kind of review work I describe on my about page.

Frequently asked questions

Does the EU AI Act apply in Norway, and when?

Not formally per November 2026. It applies once it is incorporated into the EEA Agreement and the KI-lov enters into force. The ministry said in August 2026 that it aimed to put the bill to the Storting in spring 2027. GDPR and existing law already apply to AI.

What is Norway’s KI-omnibus consultation and what is the deadline?

It is a supplementary hearing on Regulation (EU) 2026/1744, opened by the Ministry of Digitalisation and Public Governance on 30 October 2026. Responses are due by 11 December 2026, with no extension expected, and all responses are published.

When does Norway’s AI regulatory sandbox open?

The omnibus requires at least one national sandbox in operation by 2 August 2027. Norway’s will be run jointly by Digdir, Datatilsynet and Nkom under KI Norge. No opening date has been published.

What does the SMC category mean for Norwegian companies?

Companies with fewer than 750 employees and turnover up to EUR 150 million or a balance sheet up to EUR 129 million get lighter rules on documentation, quality management, sandbox access and penalties. The ministry expects the vast majority of Norwegian businesses to be on SME or SMC tracks.

Does the omnibus change Norway’s draft KI-lov?

The ministry says it requires, in principle, no changes to the provisions consulted on in June 2025. The bill will still be updated, and responses from the 2025 hearing may shape it.

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