Delar av dette innhaldet er ikkje tilgjengeleg på nynorsk enno.

Guidelines for the use of generative AI in communication

These guidelines are designed to support the responsible and ethical use of generative artificial intelligence (AI) in the field of communication.

The guidelines are intended to help build trust among users and ensure that the Tax Administration uses AI in a way that promotes efficiency, transparency, and fairness.

All use of generative artificial intelligence in the Tax Administration must comply with:

What we use AI for

In the Tax Administration, we can use generative artificial intelligence to:

Create text through writing support, inspiration, translation, summarisation, and subtitling of videos.
Analyse text from external users and answer questions via chatbot.
Process graphic material and illustrations, as long as the result follows the guidelines for design (our style guide).
Generate synthetic voices for use in frontline customer service. This must be clearly labelled.
Generate illustrations and other visual content for internal use, as long as this is clearly labelled.
Colour correct and relight photos.

What we do not use AI for

In the Tax Administration, we cannot use:

AI-generated images that can be mistaken for real photographs. This is to protect the credibility and documentary value of photography.
AI-generated images for press photos.
AI-generated videos depicting real people or that could be mistaken for a video of real people.
Text that's mainly generated by AI for meaningful content, such as an op-ed or speeches, with an employee as the author.

We must always label our content clearly

In the Tax Administration, we always label AI-generated content in the following way:

For text that's largely AI-generated:

Add a note at the beginning of the text: "This text has been prepared in whole or in part with artificial intelligence."
In documents, the labelling must be included in the metadata and in the document itself.

For audio:

AI-generated voices should be announced before playback: "You are talking to/listening to an AI-generated voice."

For illustrations and images:

Clear labelling in the caption: "Illustration generated with AI."
When sharing on social media, AI use must always be mentioned in the text accompanying the image.

For videos:

If a video contains AI-generated video clips, images, or audio, these elements must be labelled as they appear.

General requirements for labelling:

The labelling must be easily visible and understandable for all users.
The labelling must follow the content when sharing and reusing.
If there's any doubt about the need for labelling, the content must be labelled.
The labelling must be in the same language as the content.

All use must be checked

All AI-generated content that is to be used internally or externally must be checked by a colleague or manager. This is to reduce the risk of misinformation, hallucination, and bias. To have someone check the work is meant to ensure that the content is correct, relevant, and in line with the Tax Administration's style guide and other guidelines.

Employees will receive training

Employees will receive the necessary training in the use of AI tools and an understanding of the ethical and legal aspects of AI use. This is a management responsibility.

Policy updates

These guidelines are updated continuously due to rapid developments in the field of AI. Regular evaluations of AI tools must be carried out to ensure that they work as intended and that they comply with the established guidelines.

Versions

The second version of the guidelines was published on 27 February 2025.
The first version of the guidelines was published on 9 October 2024

Translation

Norwegian version