Finland and Norway seek international oversight of frontier AI

Finland and Norway have launched an international appeal calling for stronger oversight of frontier AI models. The initiative was announced on 21 September 2026 by Finnish President Alexander Stubb and Norwegian Prime Minister Jonas Gahr Støre. The appeal says increasingly capable AI systems should remain under human direction and control.

The appeal points to recent cases in which AI systems have bypassed safeguards, exploited vulnerabilities and gained unauthorised access to real-world systems. It argues that the speed of development makes stronger international cooperation necessary.

The initiative sets out three areas for action.

First, AI companies are encouraged to introduce transparent safety protocols, including mandatory testing before deployment and independent evaluations. The appeal also calls for qualified external evaluators to have sufficient access to assess the risks of AI systems.

Second, governments and regional organisations are urged to develop common standards and improve transparency. This includes mechanisms for sharing information about serious AI safety incidents and expanding access to scientific expertise and evaluation capabilities across regions.

Third, the appeal calls on UN member states to explore an international institution that could set standards, enable verification and bring states together when defined AI capability thresholds are reached. This would add an international institutional layer to existing national and regional approaches to AI oversight.

The initiative has attracted support from leaders across several regions. The Finnish president’s office said on 25 September that 30 leaders from 28 countries had joined the appeal. The supporters include leaders from countries in Europe, Africa, Asia, the Middle East, North America and Australia, as well as the European Commission.

The appeal was also raised during Finland’s participation in the 81st UN General Assembly. President Stubb said the initiative seeks to develop mechanisms for controlling frontier AI models and called for broad participation from UN member states.

What are frontier AI models?

The term generally refers to the most advanced AI models available at a given time. These systems can perform a wide range of tasks and may have capabilities that create new challenges for testing and oversight. In this initiative, the focus is on models whose capabilities could create significant safety and security risks if they are deployed without adequate safeguards.

The Finnish-Norwegian proposal therefore focuses on how these systems are tested and monitored before and after deployment, rather than only regulating particular applications. Its proposed measures include independent evaluation, incident reporting, common standards and possible international verification.

The appeal remains open to further endorsements.