Scores
  • Premier Cup FT Riverside FC2 Northgate United1
  • Premier Cup LIVE 67' Harbor City1 Westfield Rovers1
  • Premier Cup FT Oakmont0 Lakeside Athletic3
  • Premier Cup 20:45 Port Ashby Kingsbridge
  • Basketball League FT Eastvale Tigers88 Summit Hawks84
  • National League Sat 15:00 Redwood Rangers Silverton Stars

What the National Assembly’s new AI model rules mean for developers and users

Makers of large AI models must publish safety tests and label generated content, with fines of up to 4 percent of turnover for breaches.

Stack of bound documents beside a closed laptop and a wooden gavel on a desk in soft daylight Higgsfield AI

The National Assembly has approved the country’s most detailed rules yet for artificial intelligence, setting out what companies that build and sell large AI models must do before their systems reach the public. The bill passed by 212 votes to 148, with the Civic Union and the Green Alliance in favor and the Progress Party opposed.

Here is what the new law covers, and what it means for the people who build and use AI tools.

Who is covered

The rules apply to developers of general-purpose models above a set computing threshold, as well as to any company that offers such a model to customers in the country, wherever it is based. Smaller businesses that simply use a popular chatbot or an image generator inside their own products face lighter obligations.

Read also

Covered developers will have to run and publish safety tests before release, including checks for dangerous misuse, bias in hiring or lending decisions and the leaking of personal data. They must also keep records of the data used to train their models and report serious incidents within 72 hours.

What changes for users

The most visible change is labeling. Images, audio and video created by AI will need a clear notice, and chatbots will have to tell people they are talking to a machine. A new AI Oversight Office within the digital ministry will take complaints from the public and can order a product to be withdrawn.

For the first time, an ordinary person who feels harmed by an AI system will know where to go.

Rasa Mikkelsen, deputy digital minister

Fines for the most serious breaches can reach 4 percent of a company’s annual turnover.

Critics and next steps

Opponents argue the law will burden local start-ups while large foreign firms can absorb the cost. Several founders in Elmbridge, home to a growing cluster of AI companies, warned during hearings that the testing rules were vague.

We support safety, but we need to know exactly which tests count. Otherwise lawyers, not engineers, will decide what we can ship.

Jonas Brekke, founder of an Elmbridge language-software company

The labeling rules take effect in six months. The testing and reporting duties follow a year after that, giving the oversight office time to publish detailed guidance. The digital ministry says it will consult industry, universities and consumer groups on the first drafts this winter.

Add News to your preferred sources on Google.

Trust and transparency

Published
Updated
Author
Megan Parker
Revision history
  • — Story updated.
  • — Article first published.

If you spot an error, let us know. We publish corrections clearly and visibly.

Comments 0

Write a comment

Join the discussion

Sign in or create a free account to comment.

Sign in Create free account

More news

In other news…

More news