Input to the Standing Committee on Local Government and Public Administration: AI-Norway

We appreciate the opportunity to provide input to the Standing Committee on Local Government and Public Administration. Our comments concern Chapter 1540, The Norwegian Digitalisation Agency, item 28, AI Norway, under the Ministry of Digitalisation and Public Governance:

• Chapter 1540, The Norwegian Digitalisation Agency, item 28, AI Norway

‍

Langsikt supports the government's proposal to allocate 30 million NOK to test the safety of AI models and to task AI Norway with working on AI safety. AI Norway is also intended to serve as the point of contact for international AI safety organizations, such as the state-run AI Safety Institutes established by the UK, Canada, and several other countries. In last year's consultation, we called for a Norwegian AI safety institute, and we view this proposal as an important first step.

This year, we would like to recommend three points for the committee to consider, which we believe will ensure that AI Norway can fulfill its new responsibilities and contribute to safer AI development and implementation.

1. Mandate and independence: "Committee members from ... assume that the 30 million NOK increase in item 28 will be used for testing AI models and other safety-related work. These members consider it crucial that this safety work is granted professional independence and the expertise required to commission and quality-assure tests of AI models."
2. Collaboration with research communities: "These members note that AI Norway's testing work should be conducted in collaboration with research communities, and wish to emphasize the importance of providing such partners with multi-year agreements, professional independence, and access to computing power."
3. Scaling up: We propose the following request: "The Storting requests that the government present a plan in the 2028 national budget to scale up AI Norway's safety work, so that Norway can contribute professionally to the international network of AI safety institutes."

Rationale

To ensure that AI Norway succeeds in this work, we believe the following three points are important:

Independent and earmarked. The government justifies this initiative by citing the need for independent assessments of AI models. At the same time, according to the budget proposal, AI Norway is expected to act as a driving force and promote the efficient and responsible use of AI. An entity that is also tasked with assessing risk requires a clearly defined mandate to maintain credibility. Furthermore, item 28, which is increasing from 10 to 45 million NOK (of which 30 million is new funding, while approximately 5 million is reallocated from another item), must also cover AI Norway's other activities and the regulatory AI sandbox—a scheme where businesses receive assistance to ensure that new AI solutions comply with regulations. The Storting approves the appropriation for this item as a whole. A formal note would clarify that the 30 million NOK is intended specifically for safety work.

Expertise and collaboration with research communities. The government intends for AI Norway to collaborate with research communities on testing. Norway has strong academic environments in AI safety, including at the University of Oslo, SINTEF/NTNU, and Simula. We therefore expect that AI Norway will largely commission tests from these research communities rather than conducting everything in-house. This only works if AI Norway has its own technical experts with experience in testing. They must be able to determine which models to test, what the tests should measure—for example, how models perform in Norwegian and within the Norwegian Labour and Welfare Administration (NAV), the health service, and municipal case management—and whether the results are satisfactory. Without such expertise, the research communities will effectively be the ones deciding what is tested. Experience from the UK AI Safety Institute (UK AISI) shows that such experts require more competitive salaries and greater professional freedom than is typical in the public sector. Shared positions with academia and secondments could make it easier to recruit them. For their part, research communities need multi-year agreements and access to computing power—the data processing capacity required to run the tests—to build expertise over time. Singapore has a similar model: the state provides funding, holds responsibility for policy, and represents the country internationally, while a university prioritizes and conducts the tests. We recommend a middle-ground solution where AI Norway defines the assignments, while the research communities retain professional freedom in how the tests are carried out.

Level and scaling. In our assessment, 30 million NOK is sufficient for coordination and advisory services, but insufficient for conducting independent tests. We estimate that approximately 50–70 million NOK annually is required, including 5–10 million for computing power, for the safety work to deliver high-quality tests and contribute to the international network of AI safety institutes. For comparison, Canada allocated 50 million Canadian dollars over five years to its institute in 2024, and the UK's institute has an annual budget of approximately 66 million pounds (around 900 million NOK). Therefore, we are requesting a plan for scaling up in the 2028 budget.

About us

‍Langsikt is an independent, non-profit think tank dedicated to long-term policy. We bridge the gap between research and policy development to ensure that society's resources are used where they can do the most good. Our focus is on major global challenges such as artificial intelligence, pandemics, and extreme poverty. Read more in our concept note for a Norwegian AI safety body (September 7, 2026): [https://www.langsikt.no/publikasjoner/konseptnotat-for-et-norsk-ki-sikkerhetsorgan]

‍

‍

Lotte Skolem | Executive Director at Langsikt | lotte@langsikt.no

Anders Eidesvik | AI Advisor at Langsikt | anders@langsikt.no

‍