What happened
On 21 September 2026, Finnish President Alexander Stubb and Norwegian PM Jonas Gahr Støre launched the 'Call for Control of Frontier AI Models', signed by leaders of at least 20 countries including Germany, Canada, Australia, Denmark, Estonia and the European Commission. It calls on frontier AI developers to adopt transparent safety protocols including mandatory capability evaluations, and urges UN member states to explore creating an 'international institution, able to set standards, enable verification, and convene states when capability thresholds are crossed' — an IAEA-style body. UN Secretary-General António Guterres publicly welcomed the call, and it was formally teed up for consideration during UN General Assembly High-level Week on 22-23 September 2026.
Why it matters
This is the most consequential multilateral governance push of the window: for the first time, a large bloc of governments is committing to binding-style pre-deployment evaluation and verification of frontier AI models at the international level, directly paralleling (and potentially competing with) US and China's bilateral AI-incident notification dialogue. It creates real momentum toward mandatory, verifiable pre-deployment model testing that would bind frontier labs in all signatory jurisdictions if realised.
Action needed
Frontier AI labs and deployers should track whether 'capability thresholds' and verification regimes become concretised; expect voluntary Call compliance to become a de facto expectation for signatory-state procurement and for any future international institution. Model evaluation and incident-notification capability should be built now.