From 2027, all Norwegian government agencies must work for greater efficiency and quality with particular emphasis on the responsible use of AI in central government. The requirement follows from circular D-3/2026 and the letters of allocation for 2027–2029. Each agency sets its level of ambition with the employee representatives and reports on measures and results in its annual report.
The Norwegian Ministry of Digitalisation and Public Governance (Digitaliserings- og forvaltningsdepartementet, DFD) adopted the common directives (fellesføringer) for the letters of allocation for 2027–2029 on 1 October 2026 in circular D-3/2026. One of the directives requires central government agencies to work to increase efficiency and quality with particular emphasis on the responsible use of AI. Each agency must itself set a reasoned level of ambition in cooperation with the employee representatives and give an account of measures and results in its annual report. The government presented the circular in a press release of 9 October 2026.
The directive raises two questions in particular. The first concerns the co-determination rights of the employee representatives when the level of ambition is set, and the second concerns how the requirement of human oversight relates to the General Data Protection Regulation, the new Public Administration Act and the AI Act. Both questions should be clarified before the letters of allocation for 2027 are issued.
What is the common directive on AI in central government?
A common directive is an assignment that the ministries include in the letter of allocation to every central government agency. The government has limited the number of such directives to a maximum of three and assumes that they will apply over several years. Two common directives have been adopted for the period 2027–2029, one of which concerns inclusion and diversity in recruitment. Under the other, central government agencies are to work to increase efficiency and quality in central government with particular emphasis on the responsible use of AI.
The directive applies both to subordinate agencies and to the ministries themselves. The ministries must include it in the letters of allocation and follow up on its implementation by the agencies. The circular links the use of AI to safeguarding the rule of law, data protection, transparency, inclusion and democratic control. In the press release, the Minister of Digitalisation and Public Governance, Torgeir Micaelsen, goes further and states that decisions in central government are to be taken on a professional basis and be subject to human oversight.
In legal terms, the common directive is an instruction within the governance dialogue between ministry and agency, and it confers no rights on citizens. It nevertheless gives the ministry a basis for requiring an explanation where the agency has done little, and it gives the employee representatives an express role in the agency’s work on AI that they did not previously have.
Who sets the level of ambition for AI?
The level of ambition is set by each agency in cooperation with the employee representatives, and it is to be adapted to the nature and size of the agency. The circular therefore does not envisage a common minimum level for central government as a whole.
Employers shall, ‘in cooperation with the employee representatives, themselves define an appropriate and reasoned level of ambition for their work with AI in accordance with the nature and size of the agency’ (circular D-3/2026, our translation).
The circular does not specify what ‘in cooperation with’ entails, and it does not state whether the level of ambition is to be set through discussion or negotiation. In our assessment, setting it is in any event a matter for discussion under the Basic Agreement for the Civil Service (hovedavtalen i staten). Under Basic Agreement for the Civil Service § 18(1)(d), business plans are subject to discussion, and § 18(2) allows the parties to demand discussion of other matters that affect the working situation of employees. An AI initiative that changes workflows and the allocation of tasks falls, in our assessment, within the scope of these provisions.
Basic Agreement for the Civil Service § 1(9) points in the same direction. The provision states that ICT and AI are to provide better services and serve as a tool for renewal and efficiency, and that employee representatives are to be involved in the development process. The Basic Agreement has been extended to the end of 2026 by PM-2025-7, and it is not yet known whether a new Basic Agreement from 2027 will amend these provisions.
The employer takes the final decision on the level of ambition. Because the choice must be justified in the annual report, a process in which the employee representatives have not been heard will nevertheless weaken that justification in relation to the ministry. The same premise underlies the AI agreement in working life of 2 October 2026, under which the state and the main organisations agreed that participation and co-determination are to be actively used when AI is introduced. The common directive gives concrete form to this commitment for the state in its capacity as employer.
What must be reported, and when?
Agencies report on their work with AI in Chapter IV of the annual report, which concerns governance and control. The report must cover the measures implemented and concrete results in the form of savings in time and cost or improved quality of services, and the agency must justify its chosen level of ambition and explain how it has worked to achieve it. No new statistical forms are introduced, since developments are to be monitored through the annual survey by Statistics Norway (SSB) on digitalisation and ICT in the public sector, which the agencies already complete. The deadlines in the reporting cycle are set out in the table below.
| Who | What | Deadline |
|---|---|---|
| The agency | Annual report describing the work with AI | 15 March |
| DFD | Summary to the ministries | No later than early June |
| The ministry | Assessment of its subordinate agencies and its own activities | 15 August |
| Responsible ministry | Evaluation of the common directive | 2029 |
What does human oversight of decisions mean?
The circular does not regulate how human oversight is to be exercised in practice. The requirements follow from legislation that applies independently of the common directive, and they therefore cannot be departed from when the level of ambition is set. The legal framework is reviewed as a whole on the topic page on artificial intelligence.
Under GDPR Art. 22, the data subject has, as a starting point, the right not to be subject to a decision based solely on automated processing which produces legal effects concerning him or her or similarly significantly affects him or her. A caseworker who merely approves the system’s proposal without a genuine assessment does not make the decision any less automated. Where the threshold lies is discussed in the article on automated decisions and the Uber fine.
For the public administration, the question will be regulated by the new Public Administration Act (forvaltningsloven 2025). Under Public Administration Act 2025 § 11(3), an administrative body may not take automated decisions covered by GDPR Art. 22 without a specific legal basis. § 12(2) gives the data subject the right to an explanation and to manual review of the decision, and § 13 requires the legal content of automated case-processing systems to be documented and, as a general rule, published. The Act has been passed, but the King has not yet set the date of its entry into force, and until then the Public Administration Act of 1967 applies together with the GDPR.
The AI Act requires high-risk systems to be designed so that they can be effectively overseen by natural persons (AI Act Art. 14), and requires deployers to assign human oversight to persons who have the necessary competence, training and authority (AI Act Art. 26(2)). Public bodies must also assess the impact on fundamental rights before certain high-risk systems are put into use (AI Act Art. 27). Although the Regulation has not yet been implemented in Norwegian law, an agency that builds up its use of AI towards 2029 should comply with it already, because systems procured today are likely to remain in use when the Regulation applies in Norway.
How does this relate to AI policies for employees?
The common directive specifies what the agency is to achieve with AI, whereas internal policies regulate how individual employees use the tools, and the two levels must be consistent. If the ambition is for caseworkers to use a language model to draft decisions, the policy must state who reviews the draft, which information may be entered, and how it is documented that a human has actually assessed the case.
How far the employer may go in monitoring individual use is discussed in the article on an AI policy for employees. Logging who has used AI in which cases may be necessary to document human oversight. Where the log can be used to monitor individual employees, it is a control measure that must be discussed with the employee representatives under Working Environment Act § 9-2. The introduction of new systems for planning and performing work also triggers requirements for information, training and participation under Working Environment Act § 4-2.
What should central government agencies do now?
- Map where the agency currently uses AI and which processes lead to decisions about individuals.
- Place the matter on the agenda with the employee representatives before the letter of allocation is issued, so that the level of ambition can be set early in 2027.
- Choose objectives that can be measured in time, cost or quality, since this is what the annual report must document.
- Clarify, for each AI solution that affects individuals, whether the decision is automated within the meaning of GDPR Art. 22, and who has the competence and authority to override the system.
- Include requirements for explanation, logging and documentation in the requirements specification for new procurements. Choices made in procurement and the requirements specification behind a tender are a separate matter for discussion under
Basic Agreement for the Civil Service § 18(1)(e). - Keep minutes of the discussions, since the minutes will document in the annual report for 2027 that the level of ambition was set in cooperation with the employee representatives.
Questions and answers
Is the common directive on AI legally binding on central government agencies?
The common directive is an instruction from the superior ministry issued through the letter of allocation and does not have the status of an Act. The agency is obliged to follow it up and to report on it in the annual report. Failure to follow it up has no effect in relation to citizens, but may have consequences in the governance dialogue with the ministry.
Does the requirement also apply to the ministries themselves?
The circular expressly provides that the common directives apply to the ministries in the same way as to subordinate agencies. The ministries report on their own activities and also submit an overall assessment to the Ministry of Digitalisation and Public Governance by 15 August.
Must the agency reach agreement with the employee representatives on the level of ambition?
The circular states that the level of ambition is to be set ‘in cooperation with’ the employee representatives, without specifying whether this means discussion or negotiation. In our assessment, it is at least a matter for discussion under the Basic Agreement for the Civil Service. The employer takes the final decision, but must have held genuine discussions and be able to justify its choice.
What does this mean for suppliers of AI solutions to central government?
Government contracting authorities will ask for documentation that makes it possible to explain decisions, maintain human oversight and report on gains. Suppliers that can provide logging, explanations and measurable savings in time and cost will be better placed in competitions for public contracts.
- Ministry of Digitalisation and Public Governance, circular D-3/2026 Common directives in letters of allocation 2027–2029 (1 October 2026, Norwegian) Part I point 2 and Part II
- Government.no, More inclusion and more AI in central government (press release, 9 October 2026, Norwegian)
- Norwegian Public Administration Act 2025 (Act of 20 June 2025 No. 81, not yet in force) §§ 11, 12, 13 and 99
- General Data Protection Regulation (EU) 2016/679 Art. 22
- Artificial Intelligence Act, Regulation (EU) 2024/1689 Arts. 14, 26(2) and 27
- Basic Agreement for the Civil Service 2023–2025, extended to the end of 2026 (Norwegian) § 1(9) and § 18
- PM-2025-7 Continuation of the Basic Agreement for the Civil Service (Norwegian)
- Norwegian Working Environment Act (arbeidsmiljøloven) §§ 4-2 and 9-2
Next legal review: 15 January 2027