Finland and Norway aim to bolster the oversight of the most advanced AI models. On the margins of the UN’s high-level week, they proposed a joint declaration titled “A Call for Control of Frontier AI Models” intended to influence the upcoming negotiations on global AI governance.
The document begins from a measured assessment. While AI yields substantial benefits, frontier systems pose tangible security risks when they escape adequate controls.
It cites cases where models managed to bypass testing protocols to infiltrate systems without authorization. A broader risk is also noted: the pace of technological progress may outpace the ability of states, companies, and civil society to manage its consequences.
It is notable that the text remains deliberately broad. It does not explicitly mention biological weapons, state cyber operations, disinformation, military uses, or the loss of control of autonomous agents. All of these topics were nonetheless debated in parallel, notably by OpenAI and Anthropic, during the same UN week.
This omission can be read as a diplomatic cautionary choice intended to build broad consensus rather than settle on sensitive points.
Three Audiences, Three Sets of Obligations
The declaration distinguishes three categories of recipients.
For leading-edge model developers, it calls for transparent safety protocols, mandatory testing before deployment, and independent evaluations, with sufficient access granted to qualified evaluators.
For governments and regional organizations, it urges converging toward common standards and better information-sharing on serious safety incidents. In this framework, equitable access to scientific expertise, including for the least-resourced countries, must be guaranteed.
For UN member states, finally, it suggests leveraging existing multilateral mechanisms and examining the creation of an international body charged with setting standards, verifying their implementation, and convening states in the event of crossing a threshold of capacity deemed critical.
A Still Limited Scope
Twenty-two countries*, alongside the European Commission, have signed on. The absence of certain major technological powers among the signatories underscores the limits of the exercise. The declaration is a political exhortation that provides neither sanctions, nor formal accession procedures, nor a precise timeline for implementation.
The UN Secretary-General, António Guterres, welcomed the initiative and endorsed the three proposed pillars, while urging AI laboratories to feed into the work of the UN’s Independent International Scientific Advisory Group on AI, the UN body charged with grounding the scientific debate on these risks.
*Among them Australia, Canada, Germany, Spain, Singapore, South Africa, Turkey and the United Arab Emirates.
